Merge pull request #95 from lordmathis/dependabot/npm_and_yarn/webui/npm-development-143683ff68

chore: bump typescript-eslint from 8.46.4 to 8.47.0 in /webui in the npm-development group

.github/dependabot.yml

@@ -0,0 +1,40 @@
+version: 2
+updates:
+  # Go modules
+  - package-ecosystem: "gomod"
+    directory: "/"
+    schedule:
+      interval: "weekly"
+      day: "monday"
+    open-pull-requests-limit: 5
+    commit-message:
+      prefix: "chore"
+    ignore:
+      - dependency-name: "*"
+        update-types: ["version-update:semver-patch"]
+    groups:
+      go-dependencies:
+        patterns:
+          - "*"
+    labels:
+      - "dependencies"
+
+  # npm dependencies for webui
+  - package-ecosystem: "npm"
+    directory: "/webui"
+    schedule:
+      interval: "weekly"
+      day: "monday"
+    open-pull-requests-limit: 5
+    commit-message:
+      prefix: "chore"
+    ignore:
+      - dependency-name: "*"
+        update-types: ["version-update:semver-patch"]
+    groups:
+      npm-production:
+        dependency-type: "production"
+      npm-development:
+        dependency-type: "development"
+    labels:
+      - "dependencies"

.gitignore

@@ -42,4 +42,7 @@ site/
 llamactl.dev.yaml
 
 # Debug files
-__debug*
+__debug*
+
+# Binary
+llamactl-*

cmd/server/main.go

@@ -5,6 +5,7 @@ import (
 	"llamactl/pkg/config"
 	"llamactl/pkg/manager"
 	"llamactl/pkg/server"
+	"log"
 	"net/http"
 	"os"
 	"os/signal"
@@ -38,8 +39,7 @@ func main() {
 	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.")
+		log.Printf("Error loading config: %v\nUsing default configuration.", err)
 	}
 
 	// Set version information
@@ -50,13 +50,11 @@ func main() {
 	// 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.")
+			log.Printf("Error creating config directory %s: %v\nPersistence will not be available.", cfg.Instances.InstancesDir, err)
 		}
 
 		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.")
+			log.Printf("Error creating log directory %s: %v\nInstance logs will not be available.", cfg.Instances.LogsDir, err)
 		}
 	}
 
@@ -81,7 +79,7 @@ func main() {
 	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)
+			log.Printf("Error starting server: %v\n", err)
 		}
 	}()
 
@@ -90,7 +88,7 @@ func main() {
 	fmt.Println("Shutting down server...")
 
 	if err := server.Close(); err != nil {
-		fmt.Printf("Error shutting down server: %v\n", err)
+		log.Printf("Error shutting down server: %v\n", err)
 	} else {
 		fmt.Println("Server shut down gracefully.")
 	}

docs/docs.go

@@ -256,6 +256,34 @@ const docTemplate = `{
                 }
             }
         },
+        "/api/v1/config": {
+            "get": {
+                "security": [
+                    {
+                        "ApiKeyAuth": []
+                    }
+                ],
+                "description": "Returns the current server configuration (sanitized)",
+                "tags": [
+                    "System"
+                ],
+                "summary": "Get server configuration",
+                "responses": {
+                    "200": {
+                        "description": "Sanitized configuration",
+                        "schema": {
+                            "$ref": "#/definitions/config.AppConfig"
+                        }
+                    },
+                    "500": {
+                        "description": "Internal Server Error",
+                        "schema": {
+                            "type": "string"
+                        }
+                    }
+                }
+            }
+        },
         "/api/v1/instances": {
             "get": {
                 "security": [
@@ -1475,6 +1503,247 @@ const docTemplate = `{
         }
     },
     "definitions": {
+        "config.AppConfig": {
+            "type": "object",
+            "properties": {
+                "auth": {
+                    "$ref": "#/definitions/config.AuthConfig"
+                },
+                "backends": {
+                    "$ref": "#/definitions/config.BackendConfig"
+                },
+                "build_time": {
+                    "type": "string"
+                },
+                "commit_hash": {
+                    "type": "string"
+                },
+                "instances": {
+                    "$ref": "#/definitions/config.InstancesConfig"
+                },
+                "local_node": {
+                    "type": "string"
+                },
+                "nodes": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "$ref": "#/definitions/config.NodeConfig"
+                    }
+                },
+                "server": {
+                    "$ref": "#/definitions/config.ServerConfig"
+                },
+                "version": {
+                    "type": "string"
+                }
+            }
+        },
+        "config.AuthConfig": {
+            "type": "object",
+            "properties": {
+                "inference_keys": {
+                    "description": "List of keys for OpenAI compatible inference endpoints",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "management_keys": {
+                    "description": "List of keys for management endpoints",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "require_inference_auth": {
+                    "description": "Require authentication for OpenAI compatible inference endpoints",
+                    "type": "boolean"
+                },
+                "require_management_auth": {
+                    "description": "Require authentication for management endpoints",
+                    "type": "boolean"
+                }
+            }
+        },
+        "config.BackendConfig": {
+            "type": "object",
+            "properties": {
+                "llama-cpp": {
+                    "$ref": "#/definitions/config.BackendSettings"
+                },
+                "mlx": {
+                    "$ref": "#/definitions/config.BackendSettings"
+                },
+                "vllm": {
+                    "$ref": "#/definitions/config.BackendSettings"
+                }
+            }
+        },
+        "config.BackendSettings": {
+            "type": "object",
+            "properties": {
+                "args": {
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "command": {
+                    "type": "string"
+                },
+                "docker": {
+                    "$ref": "#/definitions/config.DockerSettings"
+                },
+                "environment": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
+                "response_headers": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                }
+            }
+        },
+        "config.DockerSettings": {
+            "type": "object",
+            "properties": {
+                "args": {
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "enabled": {
+                    "type": "boolean"
+                },
+                "environment": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
+                "image": {
+                    "type": "string"
+                }
+            }
+        },
+        "config.InstancesConfig": {
+            "type": "object",
+            "properties": {
+                "auto_create_dirs": {
+                    "description": "Automatically create the data directory if it doesn't exist",
+                    "type": "boolean"
+                },
+                "configs_dir": {
+                    "description": "Instance config directory override",
+                    "type": "string"
+                },
+                "data_dir": {
+                    "description": "Directory where all llamactl data will be stored (instances.json, logs, etc.)",
+                    "type": "string"
+                },
+                "default_auto_restart": {
+                    "description": "Default auto-restart setting for new instances",
+                    "type": "boolean"
+                },
+                "default_max_restarts": {
+                    "description": "Default max restarts for new instances",
+                    "type": "integer"
+                },
+                "default_on_demand_start": {
+                    "description": "Default on-demand start setting for new instances",
+                    "type": "boolean"
+                },
+                "default_restart_delay": {
+                    "description": "Default restart delay for new instances (in seconds)",
+                    "type": "integer"
+                },
+                "enable_lru_eviction": {
+                    "description": "Enable LRU eviction for instance logs",
+                    "type": "boolean"
+                },
+                "logs_dir": {
+                    "description": "Logs directory override",
+                    "type": "string"
+                },
+                "max_instances": {
+                    "description": "Maximum number of instances that can be created",
+                    "type": "integer"
+                },
+                "max_running_instances": {
+                    "description": "Maximum number of instances that can be running at the same time",
+                    "type": "integer"
+                },
+                "on_demand_start_timeout": {
+                    "description": "How long to wait for an instance to start on demand (in seconds)",
+                    "type": "integer"
+                },
+                "port_range": {
+                    "description": "Port range for instances (e.g., 8000,9000)",
+                    "type": "array",
+                    "items": {
+                        "type": "integer"
+                    }
+                },
+                "timeout_check_interval": {
+                    "description": "Interval for checking instance timeouts (in minutes)",
+                    "type": "integer"
+                }
+            }
+        },
+        "config.NodeConfig": {
+            "type": "object",
+            "properties": {
+                "address": {
+                    "type": "string"
+                },
+                "api_key": {
+                    "type": "string"
+                }
+            }
+        },
+        "config.ServerConfig": {
+            "type": "object",
+            "properties": {
+                "allowed_headers": {
+                    "description": "Allowed headers for CORS (e.g., \"Accept\", \"Authorization\", \"Content-Type\", \"X-CSRF-Token\")",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "allowed_origins": {
+                    "description": "Allowed origins for CORS (e.g., \"http://localhost:3000\")",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "enable_swagger": {
+                    "description": "Enable Swagger UI for API documentation",
+                    "type": "boolean"
+                },
+                "host": {
+                    "description": "Server host to bind to",
+                    "type": "string"
+                },
+                "port": {
+                    "description": "Server port to bind to",
+                    "type": "integer"
+                },
+                "response_headers": {
+                    "description": "Response headers to send with responses",
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                }
+            }
+        },
         "instance.Instance": {
             "type": "object",
             "properties": {
@@ -1494,6 +1763,13 @@ const docTemplate = `{
                     "description": "Auto restart",
                     "type": "boolean"
                 },
+                "command_override": {
+                    "type": "string"
+                },
+                "docker_enabled": {
+                    "description": "Execution context overrides",
+                    "type": "boolean"
+                },
                 "environment": {
                     "description": "Environment variables",
                     "type": "object",

docs/managing-instances.md

@@ -42,32 +42,41 @@ Each instance is displayed as a card showing:
 ![Create Instance Screenshot](images/create_instance.png)
 
 1. Click the **"Create Instance"** button on the dashboard
-2. Enter a unique **Name** for your instance (only required field)
-3. **Select Target Node**: Choose which node to deploy the instance to from the dropdown
-4. **Choose Backend Type**:
-    - **llama.cpp**: For GGUF models using llama-server
-    - **MLX**: For MLX-optimized models (macOS only)
+2. *Optional*: Click **"Import"** to load a previously exported configuration
+
+**Instance Settings:**
+
+3. Enter a unique **Instance Name** (required)
+4. **Select Node**: Choose which node to deploy the instance to
+5. Configure **Auto Restart** settings:
+    - Enable automatic restart on failure
+    - Set max restarts and delay between attempts
+6. Configure basic instance options:
+    - **Idle Timeout**: Minutes before stopping idle instance
+    - **On Demand Start**: Start instance only when needed
+
+**Backend Configuration:**
+
+7. **Select Backend Type**:
+    - **Llama Server**: For GGUF models using llama-server
+    - **MLX LM**: For MLX-optimized models (macOS only)
     - **vLLM**: For distributed serving and high-throughput inference
-5. Configure model source:
-    - **For llama.cpp**: GGUF model path or HuggingFace repo
-    - **For MLX**: MLX model path or identifier (e.g., `mlx-community/Mistral-7B-Instruct-v0.3-4bit`)
-    - **For vLLM**: HuggingFace model identifier (e.g., `microsoft/DialoGPT-medium`)
-6. Configure optional instance management settings:
-    - **Auto Restart**: Automatically restart instance on failure
-    - **Max Restarts**: Maximum number of restart attempts
-    - **Restart Delay**: Delay in seconds between restart attempts
-    - **On Demand Start**: Start instance when receiving a request to the OpenAI compatible endpoint
-    - **Idle Timeout**: Minutes before stopping idle instance (set to 0 to disable)
-    - **Environment Variables**: Set custom environment variables for the instance process
-7. Configure backend-specific options:
-    - **llama.cpp**: Threads, context size, GPU layers, port, etc.
-    - **MLX**: Temperature, top-p, adapter path, Python environment, etc.
-    - **vLLM**: Tensor parallel size, GPU memory utilization, quantization, etc.
+8. *Optional*: Click **"Parse Command"** to import settings from an existing backend command
+9. Configure **Execution Context**:
+    - **Enable Docker**: Run backend in Docker container
+    - **Command Override**: Custom path to backend executable
+    - **Environment Variables**: Custom environment variables
 
 !!! tip "Auto-Assignment"
     Llamactl automatically assigns ports from the configured port range (default: 8000-9000) and generates API keys if authentication is enabled. You typically don't need to manually specify these values.
-
-8. Click **"Create"** to save the instance  
+  
+10. Configure **Basic Backend Options** (varies by backend):
+    - **llama.cpp**: Model path, threads, context size, GPU layers, etc.
+    - **MLX**: Model identifier, temperature, max tokens, etc.
+    - **vLLM**: Model identifier, tensor parallel size, GPU memory utilization, etc.
+11. *Optional*: Expand **Advanced Backend Options** for additional settings
+12. *Optional*: Add **Extra Args** as key-value pairs for custom command-line arguments
+13. Click **"Create"** to save the instance  
 
 **Via API**
 
@@ -82,11 +91,34 @@ curl -X POST http://localhost:8080/api/v1/instances/my-llama-instance \
       "model": "/path/to/model.gguf",
       "threads": 8,
       "ctx_size": 4096,
-      "gpu_layers": 32
+      "gpu_layers": 32,
+      "flash_attn": "on"
     },
+    "auto_restart": true,
+    "max_restarts": 3,
+    "docker_enabled": false,
+    "command_override": "/opt/llama-server-dev",
     "nodes": ["main"]
   }'
 
+# Create vLLM instance with environment variables
+curl -X POST http://localhost:8080/api/v1/instances/my-vllm-instance \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer <token>" \
+  -d '{
+    "backend_type": "vllm",
+    "backend_options": {
+      "model": "microsoft/DialoGPT-medium",
+      "tensor_parallel_size": 2,
+      "gpu_memory_utilization": 0.9
+    },
+    "on_demand_start": true,
+    "environment": {
+      "CUDA_VISIBLE_DEVICES": "0,1"
+    },
+    "nodes": ["worker1", "worker2"]
+  }'
+
 # Create MLX instance (macOS only)
 curl -X POST http://localhost:8080/api/v1/instances/my-mlx-instance \
   -H "Content-Type: application/json" \
@@ -96,74 +128,10 @@ curl -X POST http://localhost:8080/api/v1/instances/my-mlx-instance \
     "backend_options": {
       "model": "mlx-community/Mistral-7B-Instruct-v0.3-4bit",
       "temp": 0.7,
-      "top_p": 0.9,
       "max_tokens": 2048
     },
-    "auto_restart": true,
-    "max_restarts": 3,
     "nodes": ["main"]
   }'
-
-# Create vLLM instance
-curl -X POST http://localhost:8080/api/v1/instances/my-vllm-instance \
-  -H "Content-Type: application/json" \
-  -H "Authorization: Bearer <token>" \
-  -d '{
-    "backend_type": "vllm",
-    "backend_options": {
-      "model": "microsoft/DialoGPT-medium",
-      "tensor_parallel_size": 2,
-      "gpu_memory_utilization": 0.9
-    },
-    "auto_restart": true,
-    "on_demand_start": true,
-    "environment": {
-      "CUDA_VISIBLE_DEVICES": "0,1",
-      "NCCL_DEBUG": "INFO",
-      "PYTHONPATH": "/custom/path"
-    },
-    "nodes": ["main"]
-  }'
-
-# Create llama.cpp instance with HuggingFace model
-curl -X POST http://localhost:8080/api/v1/instances/gemma-3-27b \
-  -H "Content-Type: application/json" \
-  -H "Authorization: Bearer <token>" \
-  -d '{
-    "backend_type": "llama_cpp",
-    "backend_options": {
-      "hf_repo": "unsloth/gemma-3-27b-it-GGUF",
-      "hf_file": "gemma-3-27b-it-GGUF.gguf",
-      "gpu_layers": 32
-    },
-    "nodes": ["main"]
-  }'
-
-# Create instance on specific remote node
-curl -X POST http://localhost:8080/api/v1/instances/remote-llama \
-  -H "Content-Type: application/json" \
-  -H "Authorization: Bearer <token>" \
-  -d '{
-    "backend_type": "llama_cpp",
-    "backend_options": {
-      "model": "/models/llama-7b.gguf",
-      "gpu_layers": 32
-    },
-    "nodes": ["worker1"]
-  }'
-
-# Create instance on multiple nodes for high availability
-curl -X POST http://localhost:8080/api/v1/instances/multi-node-llama \
-  -H "Content-Type: application/json" \
-  -H "Authorization: Bearer <token>" \
-  -d '{
-    "backend_type": "llama_cpp",
-    "backend_options": {
-      "model": "/models/llama-7b.gguf",
-      "gpu_layers": 32
-    },
-    "nodes": ["worker1", "worker2", "worker3"]
-  }'
 ```
 
 ## Start Instance
@@ -219,6 +187,12 @@ curl -X PUT http://localhost:8080/api/v1/instances/{name} \
     Configuration changes require restarting the instance to take effect.
 
 
+## Export Instance
+
+**Via Web UI**
+1. Click the **"More actions"** button (three dots) on an instance card
+2. Click **"Export"** to download the instance configuration as a JSON file
+
 ## View Logs
 
 **Via Web UI**

docs/swagger.json

@@ -249,6 +249,34 @@
                 }
             }
         },
+        "/api/v1/config": {
+            "get": {
+                "security": [
+                    {
+                        "ApiKeyAuth": []
+                    }
+                ],
+                "description": "Returns the current server configuration (sanitized)",
+                "tags": [
+                    "System"
+                ],
+                "summary": "Get server configuration",
+                "responses": {
+                    "200": {
+                        "description": "Sanitized configuration",
+                        "schema": {
+                            "$ref": "#/definitions/config.AppConfig"
+                        }
+                    },
+                    "500": {
+                        "description": "Internal Server Error",
+                        "schema": {
+                            "type": "string"
+                        }
+                    }
+                }
+            }
+        },
         "/api/v1/instances": {
             "get": {
                 "security": [
@@ -1468,6 +1496,247 @@
         }
     },
     "definitions": {
+        "config.AppConfig": {
+            "type": "object",
+            "properties": {
+                "auth": {
+                    "$ref": "#/definitions/config.AuthConfig"
+                },
+                "backends": {
+                    "$ref": "#/definitions/config.BackendConfig"
+                },
+                "build_time": {
+                    "type": "string"
+                },
+                "commit_hash": {
+                    "type": "string"
+                },
+                "instances": {
+                    "$ref": "#/definitions/config.InstancesConfig"
+                },
+                "local_node": {
+                    "type": "string"
+                },
+                "nodes": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "$ref": "#/definitions/config.NodeConfig"
+                    }
+                },
+                "server": {
+                    "$ref": "#/definitions/config.ServerConfig"
+                },
+                "version": {
+                    "type": "string"
+                }
+            }
+        },
+        "config.AuthConfig": {
+            "type": "object",
+            "properties": {
+                "inference_keys": {
+                    "description": "List of keys for OpenAI compatible inference endpoints",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "management_keys": {
+                    "description": "List of keys for management endpoints",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "require_inference_auth": {
+                    "description": "Require authentication for OpenAI compatible inference endpoints",
+                    "type": "boolean"
+                },
+                "require_management_auth": {
+                    "description": "Require authentication for management endpoints",
+                    "type": "boolean"
+                }
+            }
+        },
+        "config.BackendConfig": {
+            "type": "object",
+            "properties": {
+                "llama-cpp": {
+                    "$ref": "#/definitions/config.BackendSettings"
+                },
+                "mlx": {
+                    "$ref": "#/definitions/config.BackendSettings"
+                },
+                "vllm": {
+                    "$ref": "#/definitions/config.BackendSettings"
+                }
+            }
+        },
+        "config.BackendSettings": {
+            "type": "object",
+            "properties": {
+                "args": {
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "command": {
+                    "type": "string"
+                },
+                "docker": {
+                    "$ref": "#/definitions/config.DockerSettings"
+                },
+                "environment": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
+                "response_headers": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                }
+            }
+        },
+        "config.DockerSettings": {
+            "type": "object",
+            "properties": {
+                "args": {
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "enabled": {
+                    "type": "boolean"
+                },
+                "environment": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                },
+                "image": {
+                    "type": "string"
+                }
+            }
+        },
+        "config.InstancesConfig": {
+            "type": "object",
+            "properties": {
+                "auto_create_dirs": {
+                    "description": "Automatically create the data directory if it doesn't exist",
+                    "type": "boolean"
+                },
+                "configs_dir": {
+                    "description": "Instance config directory override",
+                    "type": "string"
+                },
+                "data_dir": {
+                    "description": "Directory where all llamactl data will be stored (instances.json, logs, etc.)",
+                    "type": "string"
+                },
+                "default_auto_restart": {
+                    "description": "Default auto-restart setting for new instances",
+                    "type": "boolean"
+                },
+                "default_max_restarts": {
+                    "description": "Default max restarts for new instances",
+                    "type": "integer"
+                },
+                "default_on_demand_start": {
+                    "description": "Default on-demand start setting for new instances",
+                    "type": "boolean"
+                },
+                "default_restart_delay": {
+                    "description": "Default restart delay for new instances (in seconds)",
+                    "type": "integer"
+                },
+                "enable_lru_eviction": {
+                    "description": "Enable LRU eviction for instance logs",
+                    "type": "boolean"
+                },
+                "logs_dir": {
+                    "description": "Logs directory override",
+                    "type": "string"
+                },
+                "max_instances": {
+                    "description": "Maximum number of instances that can be created",
+                    "type": "integer"
+                },
+                "max_running_instances": {
+                    "description": "Maximum number of instances that can be running at the same time",
+                    "type": "integer"
+                },
+                "on_demand_start_timeout": {
+                    "description": "How long to wait for an instance to start on demand (in seconds)",
+                    "type": "integer"
+                },
+                "port_range": {
+                    "description": "Port range for instances (e.g., 8000,9000)",
+                    "type": "array",
+                    "items": {
+                        "type": "integer"
+                    }
+                },
+                "timeout_check_interval": {
+                    "description": "Interval for checking instance timeouts (in minutes)",
+                    "type": "integer"
+                }
+            }
+        },
+        "config.NodeConfig": {
+            "type": "object",
+            "properties": {
+                "address": {
+                    "type": "string"
+                },
+                "api_key": {
+                    "type": "string"
+                }
+            }
+        },
+        "config.ServerConfig": {
+            "type": "object",
+            "properties": {
+                "allowed_headers": {
+                    "description": "Allowed headers for CORS (e.g., \"Accept\", \"Authorization\", \"Content-Type\", \"X-CSRF-Token\")",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "allowed_origins": {
+                    "description": "Allowed origins for CORS (e.g., \"http://localhost:3000\")",
+                    "type": "array",
+                    "items": {
+                        "type": "string"
+                    }
+                },
+                "enable_swagger": {
+                    "description": "Enable Swagger UI for API documentation",
+                    "type": "boolean"
+                },
+                "host": {
+                    "description": "Server host to bind to",
+                    "type": "string"
+                },
+                "port": {
+                    "description": "Server port to bind to",
+                    "type": "integer"
+                },
+                "response_headers": {
+                    "description": "Response headers to send with responses",
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                }
+            }
+        },
         "instance.Instance": {
             "type": "object",
             "properties": {
@@ -1487,6 +1756,13 @@
                     "description": "Auto restart",
                     "type": "boolean"
                 },
+                "command_override": {
+                    "type": "string"
+                },
+                "docker_enabled": {
+                    "description": "Execution context overrides",
+                    "type": "boolean"
+                },
                 "environment": {
                     "description": "Environment variables",
                     "type": "object",

docs/swagger.yaml

@@ -1,5 +1,173 @@
 basePath: /api/v1
 definitions:
+  config.AppConfig:
+    properties:
+      auth:
+        $ref: '#/definitions/config.AuthConfig'
+      backends:
+        $ref: '#/definitions/config.BackendConfig'
+      build_time:
+        type: string
+      commit_hash:
+        type: string
+      instances:
+        $ref: '#/definitions/config.InstancesConfig'
+      local_node:
+        type: string
+      nodes:
+        additionalProperties:
+          $ref: '#/definitions/config.NodeConfig'
+        type: object
+      server:
+        $ref: '#/definitions/config.ServerConfig'
+      version:
+        type: string
+    type: object
+  config.AuthConfig:
+    properties:
+      inference_keys:
+        description: List of keys for OpenAI compatible inference endpoints
+        items:
+          type: string
+        type: array
+      management_keys:
+        description: List of keys for management endpoints
+        items:
+          type: string
+        type: array
+      require_inference_auth:
+        description: Require authentication for OpenAI compatible inference endpoints
+        type: boolean
+      require_management_auth:
+        description: Require authentication for management endpoints
+        type: boolean
+    type: object
+  config.BackendConfig:
+    properties:
+      llama-cpp:
+        $ref: '#/definitions/config.BackendSettings'
+      mlx:
+        $ref: '#/definitions/config.BackendSettings'
+      vllm:
+        $ref: '#/definitions/config.BackendSettings'
+    type: object
+  config.BackendSettings:
+    properties:
+      args:
+        items:
+          type: string
+        type: array
+      command:
+        type: string
+      docker:
+        $ref: '#/definitions/config.DockerSettings'
+      environment:
+        additionalProperties:
+          type: string
+        type: object
+      response_headers:
+        additionalProperties:
+          type: string
+        type: object
+    type: object
+  config.DockerSettings:
+    properties:
+      args:
+        items:
+          type: string
+        type: array
+      enabled:
+        type: boolean
+      environment:
+        additionalProperties:
+          type: string
+        type: object
+      image:
+        type: string
+    type: object
+  config.InstancesConfig:
+    properties:
+      auto_create_dirs:
+        description: Automatically create the data directory if it doesn't exist
+        type: boolean
+      configs_dir:
+        description: Instance config directory override
+        type: string
+      data_dir:
+        description: Directory where all llamactl data will be stored (instances.json,
+          logs, etc.)
+        type: string
+      default_auto_restart:
+        description: Default auto-restart setting for new instances
+        type: boolean
+      default_max_restarts:
+        description: Default max restarts for new instances
+        type: integer
+      default_on_demand_start:
+        description: Default on-demand start setting for new instances
+        type: boolean
+      default_restart_delay:
+        description: Default restart delay for new instances (in seconds)
+        type: integer
+      enable_lru_eviction:
+        description: Enable LRU eviction for instance logs
+        type: boolean
+      logs_dir:
+        description: Logs directory override
+        type: string
+      max_instances:
+        description: Maximum number of instances that can be created
+        type: integer
+      max_running_instances:
+        description: Maximum number of instances that can be running at the same time
+        type: integer
+      on_demand_start_timeout:
+        description: How long to wait for an instance to start on demand (in seconds)
+        type: integer
+      port_range:
+        description: Port range for instances (e.g., 8000,9000)
+        items:
+          type: integer
+        type: array
+      timeout_check_interval:
+        description: Interval for checking instance timeouts (in minutes)
+        type: integer
+    type: object
+  config.NodeConfig:
+    properties:
+      address:
+        type: string
+      api_key:
+        type: string
+    type: object
+  config.ServerConfig:
+    properties:
+      allowed_headers:
+        description: Allowed headers for CORS (e.g., "Accept", "Authorization", "Content-Type",
+          "X-CSRF-Token")
+        items:
+          type: string
+        type: array
+      allowed_origins:
+        description: Allowed origins for CORS (e.g., "http://localhost:3000")
+        items:
+          type: string
+        type: array
+      enable_swagger:
+        description: Enable Swagger UI for API documentation
+        type: boolean
+      host:
+        description: Server host to bind to
+        type: string
+      port:
+        description: Server port to bind to
+        type: integer
+      response_headers:
+        additionalProperties:
+          type: string
+        description: Response headers to send with responses
+        type: object
+    type: object
   instance.Instance:
     properties:
       created:
@@ -13,6 +181,11 @@ definitions:
       auto_restart:
         description: Auto restart
         type: boolean
+      command_override:
+        type: string
+      docker_enabled:
+        description: Execution context overrides
+        type: boolean
       environment:
         additionalProperties:
           type: string
@@ -216,6 +389,23 @@ paths:
       summary: Parse vllm serve command
       tags:
       - Backends
+  /api/v1/config:
+    get:
+      description: Returns the current server configuration (sanitized)
+      responses:
+        "200":
+          description: Sanitized configuration
+          schema:
+            $ref: '#/definitions/config.AppConfig'
+        "500":
+          description: Internal Server Error
+          schema:
+            type: string
+      security:
+      - ApiKeyAuth: []
+      summary: Get server configuration
+      tags:
+      - System
   /api/v1/instances:
     get:
       description: Returns a list of all instances managed by the server

pkg/backends/backend.go

@@ -79,26 +79,28 @@ func (o *Options) UnmarshalJSON(data []byte) error {
 }
 
 func (o *Options) MarshalJSON() ([]byte, error) {
-	type Alias Options
-	aux := &struct {
-		*Alias
-	}{
-		Alias: (*Alias)(o),
-	}
-
 	// Get backend and marshal it
+	var backendOptions map[string]any
 	backend := o.getBackend()
 	if backend != nil {
 		optionsData, err := json.Marshal(backend)
 		if err != nil {
 			return nil, fmt.Errorf("failed to marshal backend options: %w", err)
 		}
-		if err := json.Unmarshal(optionsData, &aux.BackendOptions); err != nil {
+		// Create a new map to avoid concurrent map writes
+		backendOptions = make(map[string]any)
+		if err := json.Unmarshal(optionsData, &backendOptions); err != nil {
 			return nil, fmt.Errorf("failed to unmarshal backend options to map: %w", err)
 		}
 	}
 
-	return json.Marshal(aux)
+	return json.Marshal(&struct {
+		BackendType    BackendType    `json:"backend_type"`
+		BackendOptions map[string]any `json:"backend_options,omitempty"`
+	}{
+		BackendType:    o.BackendType,
+		BackendOptions: backendOptions,
+	})
 }
 
 // setBackendOptions stores the backend in the appropriate typed field
@@ -140,32 +142,54 @@ func (o *Options) getBackend() backend {
 	}
 }
 
-func (o *Options) isDockerEnabled(backend *config.BackendSettings) bool {
-	if backend.Docker != nil && backend.Docker.Enabled && o.BackendType != BackendTypeMlxLm {
-		return true
+// isDockerEnabled checks if Docker is enabled with an optional override
+func (o *Options) isDockerEnabled(backend *config.BackendSettings, dockerEnabledOverride *bool) bool {
+	// Check if backend supports Docker
+	if backend.Docker == nil {
+		return false
 	}
-	return false
+
+	// MLX doesn't support Docker
+	if o.BackendType == BackendTypeMlxLm {
+		return false
+	}
+
+	// Check for instance-level override
+	if dockerEnabledOverride != nil {
+		return *dockerEnabledOverride
+	}
+
+	// Fall back to config value
+	return backend.Docker.Enabled
 }
 
-func (o *Options) IsDockerEnabled(backendConfig *config.BackendConfig) bool {
+func (o *Options) IsDockerEnabled(backendConfig *config.BackendConfig, dockerEnabled *bool) bool {
 	backendSettings := o.getBackendSettings(backendConfig)
-	return o.isDockerEnabled(backendSettings)
+	return o.isDockerEnabled(backendSettings, dockerEnabled)
 }
 
 // GetCommand builds the command to run the backend
-func (o *Options) GetCommand(backendConfig *config.BackendConfig) string {
-
+func (o *Options) GetCommand(backendConfig *config.BackendConfig, dockerEnabled *bool, commandOverride string) string {
 	backendSettings := o.getBackendSettings(backendConfig)
 
-	if o.isDockerEnabled(backendSettings) {
+	// Determine if Docker is enabled
+	useDocker := o.isDockerEnabled(backendSettings, dockerEnabled)
+
+	if useDocker {
 		return "docker"
 	}
 
+	// Check for command override (only applies when not in Docker mode)
+	if commandOverride != "" {
+		return commandOverride
+	}
+
+	// Fall back to config command
 	return backendSettings.Command
 }
 
 // buildCommandArgs builds command line arguments for the backend
-func (o *Options) BuildCommandArgs(backendConfig *config.BackendConfig) []string {
+func (o *Options) BuildCommandArgs(backendConfig *config.BackendConfig, dockerEnabled *bool) []string {
 
 	var args []string
 
@@ -175,7 +199,7 @@ func (o *Options) BuildCommandArgs(backendConfig *config.BackendConfig) []string
 		return args
 	}
 
-	if o.isDockerEnabled(backendSettings) {
+	if o.isDockerEnabled(backendSettings, dockerEnabled) {
 		// For Docker, start with Docker args
 		args = append(args, backendSettings.Docker.Args...)
 		args = append(args, backendSettings.Docker.Image)
@@ -191,7 +215,7 @@ func (o *Options) BuildCommandArgs(backendConfig *config.BackendConfig) []string
 }
 
 // BuildEnvironment builds the environment variables for the backend process
-func (o *Options) BuildEnvironment(backendConfig *config.BackendConfig, environment map[string]string) map[string]string {
+func (o *Options) BuildEnvironment(backendConfig *config.BackendConfig, dockerEnabled *bool, environment map[string]string) map[string]string {
 
 	backendSettings := o.getBackendSettings(backendConfig)
 	env := map[string]string{}
@@ -200,7 +224,7 @@ func (o *Options) BuildEnvironment(backendConfig *config.BackendConfig, environm
 		maps.Copy(env, backendSettings.Environment)
 	}
 
-	if o.isDockerEnabled(backendSettings) {
+	if o.isDockerEnabled(backendSettings, dockerEnabled) {
 		if backendSettings.Docker.Environment != nil {
 			maps.Copy(env, backendSettings.Docker.Environment)
 		}

pkg/backends/builder.go

@@ -93,3 +93,22 @@ func BuildDockerCommand(backendConfig *config.BackendSettings, instanceArgs []st
 
 	return "docker", dockerArgs, nil
 }
+
+// convertExtraArgsToFlags converts map[string]string to command flags
+// Empty values become boolean flags: {"flag": ""} → ["--flag"]
+// Non-empty values: {"flag": "value"} → ["--flag", "value"]
+func convertExtraArgsToFlags(extraArgs map[string]string) []string {
+	var args []string
+
+	for key, value := range extraArgs {
+		if value == "" {
+			// Boolean flag
+			args = append(args, "--"+key)
+		} else {
+			// Value flag
+			args = append(args, "--"+key, value)
+		}
+	}
+
+	return args
+}

pkg/backends/llama.go

@@ -5,7 +5,6 @@ import (
 	"fmt"
 	"llamactl/pkg/validation"
 	"reflect"
-	"strconv"
 )
 
 // llamaMultiValuedFlags defines flags that should be repeated for each value rather than comma-separated
@@ -41,7 +40,7 @@ type LlamaServerOptions struct {
 	BatchSize               int      `json:"batch_size,omitempty"`
 	UBatchSize              int      `json:"ubatch_size,omitempty"`
 	Keep                    int      `json:"keep,omitempty"`
-	FlashAttn               bool     `json:"flash_attn,omitempty"`
+	FlashAttn               string   `json:"flash_attn,omitempty"`
 	NoPerf                  bool     `json:"no_perf,omitempty"`
 	Escape                  bool     `json:"escape,omitempty"`
 	NoEscape                bool     `json:"no_escape,omitempty"`
@@ -187,6 +186,10 @@ type LlamaServerOptions struct {
 	FIMQwen7BDefault      bool `json:"fim_qwen_7b_default,omitempty"`
 	FIMQwen7BSpec         bool `json:"fim_qwen_7b_spec,omitempty"`
 	FIMQwen14BSpec        bool `json:"fim_qwen_14b_spec,omitempty"`
+
+	// ExtraArgs are additional command line arguments.
+	// Example: {"verbose": "", "log-file": "/logs/llama.log"}
+	ExtraArgs map[string]string `json:"extra_args,omitempty"`
 }
 
 // UnmarshalJSON implements custom JSON unmarshaling to support multiple field names
@@ -209,6 +212,15 @@ func (o *LlamaServerOptions) UnmarshalJSON(data []byte) error {
 	// Copy to our struct
 	*o = LlamaServerOptions(temp)
 
+	// Track which fields we've processed
+	processedFields := make(map[string]bool)
+
+	// Get all known canonical field names from struct tags
+	knownFields := getKnownFieldNames(o)
+	for field := range knownFields {
+		processedFields[field] = true
+	}
+
 	// Handle alternative field names
 	fieldMappings := map[string]string{
 		// Common params
@@ -220,7 +232,7 @@ func (o *LlamaServerOptions) UnmarshalJSON(data []byte) error {
 		"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
+		"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
@@ -234,7 +246,7 @@ func (o *LlamaServerOptions) UnmarshalJSON(data []byte) error {
 		"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
+		"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
@@ -250,9 +262,9 @@ func (o *LlamaServerOptions) UnmarshalJSON(data []byte) error {
 		"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
+		"log_verbose":   "verbose",         // --log-verbose
 		"lv":            "verbosity",       // -lv, --verbosity, --log-verbosity N
-		"log-verbosity": "verbosity",       // --log-verbosity N
+		"log_verbosity": "verbosity",       // --log-verbosity N
 
 		// Sampling params
 		"s":  "seed",             // -s, --seed SEED
@@ -269,21 +281,23 @@ func (o *LlamaServerOptions) UnmarshalJSON(data []byte) error {
 		"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
+		"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
+		"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
+	// Process alternative field names and mark them as processed
 	for altName, canonicalName := range fieldMappings {
+		processedFields[altName] = true // Mark alternatives as known
+
 		if value, exists := raw[altName]; exists {
 			// Use reflection to set the field value
 			v := reflect.ValueOf(o).Elem()
@@ -294,36 +308,21 @@ func (o *LlamaServerOptions) UnmarshalJSON(data []byte) error {
 			})
 
 			if field.IsValid() && field.CanSet() {
-				switch field.Kind() {
-				case reflect.Int:
-					if intVal, ok := value.(float64); ok {
-						field.SetInt(int64(intVal))
-					} else if strVal, ok := value.(string); ok {
-						if intVal, err := strconv.Atoi(strVal); err == nil {
-							field.SetInt(int64(intVal))
-						}
-					}
-				case reflect.Float64:
-					if floatVal, ok := value.(float64); ok {
-						field.SetFloat(floatVal)
-					} else if strVal, ok := value.(string); ok {
-						if floatVal, err := strconv.ParseFloat(strVal, 64); err == nil {
-							field.SetFloat(floatVal)
-						}
-					}
-				case reflect.String:
-					if strVal, ok := value.(string); ok {
-						field.SetString(strVal)
-					}
-				case reflect.Bool:
-					if boolVal, ok := value.(bool); ok {
-						field.SetBool(boolVal)
-					}
-				}
+				setFieldValue(field, value)
 			}
 		}
 	}
 
+	// Collect unknown fields into ExtraArgs
+	if o.ExtraArgs == nil {
+		o.ExtraArgs = make(map[string]string)
+	}
+	for key, value := range raw {
+		if !processedFields[key] {
+			o.ExtraArgs[key] = fmt.Sprintf("%v", value)
+		}
+	}
+
 	return nil
 }
 
@@ -354,6 +353,18 @@ func (o *LlamaServerOptions) Validate() error {
 		return validation.ValidationError(fmt.Errorf("invalid port range: %d", o.Port))
 	}
 
+	// Validate extra_args keys and values
+	for key, value := range o.ExtraArgs {
+		if err := validation.ValidateStringForInjection(key); err != nil {
+			return validation.ValidationError(fmt.Errorf("extra_args key %q: %w", key, err))
+		}
+		if value != "" {
+			if err := validation.ValidateStringForInjection(value); err != nil {
+				return validation.ValidationError(fmt.Errorf("extra_args value for %q: %w", key, err))
+			}
+		}
+	}
+
 	return nil
 }
 
@@ -361,7 +372,12 @@ func (o *LlamaServerOptions) Validate() error {
 func (o *LlamaServerOptions) BuildCommandArgs() []string {
 	// Llama uses multiple flags for arrays by default (not comma-separated)
 	// Use package-level llamaMultiValuedFlags variable
-	return BuildCommandArgs(o, llamaMultiValuedFlags)
+	args := BuildCommandArgs(o, llamaMultiValuedFlags)
+
+	// Append extra args at the end
+	args = append(args, convertExtraArgsToFlags(o.ExtraArgs)...)
+
+	return args
 }
 
 func (o *LlamaServerOptions) BuildDockerArgs() []string {

pkg/backends/llama_test.go

@@ -4,6 +4,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"llamactl/pkg/backends"
+	"llamactl/pkg/config"
 	"llamactl/pkg/testutil"
 	"reflect"
 	"testing"
@@ -33,12 +34,11 @@ func TestLlamaCppBuildCommandArgs_BooleanFields(t *testing.T) {
 		{
 			name: "multiple booleans",
 			options: backends.LlamaServerOptions{
-				Verbose:   true,
-				FlashAttn: true,
-				Mlock:     false,
-				NoMmap:    true,
+				Verbose: true,
+				Mlock:   false,
+				NoMmap:  true,
 			},
-			expected: []string{"--verbose", "--flash-attn", "--no-mmap"},
+			expected: []string{"--verbose", "--no-mmap"},
 			excluded: []string{"--mlock"},
 		},
 	}
@@ -346,7 +346,7 @@ func TestParseLlamaCommand(t *testing.T) {
 		},
 		{
 			name:      "multiple value types",
-			command:   "llama-server --model /test/model.gguf --gpu-layers 32 --temp 0.7 --verbose --no-mmap",
+			command:   "llama-server --model /test/model.gguf --n-gpu-layers 32 --temp 0.7 --verbose --no-mmap",
 			expectErr: false,
 			validate: func(t *testing.T, opts *backends.LlamaServerOptions) {
 				if opts.Model != "/test/model.gguf" {
@@ -434,3 +434,195 @@ func TestParseLlamaCommandArrays(t *testing.T) {
 		}
 	}
 }
+
+func TestLlamaCppBuildCommandArgs_ExtraArgs(t *testing.T) {
+	options := backends.LlamaServerOptions{
+		Model: "/models/test.gguf",
+		ExtraArgs: map[string]string{
+			"flash-attn": "",               // boolean flag
+			"log-file":   "/logs/test.log", // value flag
+		},
+	}
+
+	args := options.BuildCommandArgs()
+
+	// Check that extra args are present
+	if !testutil.Contains(args, "--flash-attn") {
+		t.Error("Expected --flash-attn flag not found")
+	}
+	if !testutil.Contains(args, "--log-file") || !testutil.Contains(args, "/logs/test.log") {
+		t.Error("Expected --log-file flag or value not found")
+	}
+}
+
+func TestParseLlamaCommand_ExtraArgs(t *testing.T) {
+	tests := []struct {
+		name      string
+		command   string
+		expectErr bool
+		validate  func(*testing.T, *backends.LlamaServerOptions)
+	}{
+		{
+			name:      "extra args with known fields",
+			command:   "llama-server --model /path/to/model.gguf --gpu-layers 32 --unknown-flag value --another-bool-flag",
+			expectErr: false,
+			validate: func(t *testing.T, opts *backends.LlamaServerOptions) {
+				if opts.Model != "/path/to/model.gguf" {
+					t.Errorf("expected model '/path/to/model.gguf', got '%s'", opts.Model)
+				}
+				if opts.GPULayers != 32 {
+					t.Errorf("expected gpu_layers 32, got %d", opts.GPULayers)
+				}
+				if opts.ExtraArgs == nil {
+					t.Fatal("expected extra_args to be non-nil")
+				}
+				if val, ok := opts.ExtraArgs["unknown_flag"]; !ok || val != "value" {
+					t.Errorf("expected extra_args[unknown_flag]='value', got '%s'", val)
+				}
+				if val, ok := opts.ExtraArgs["another_bool_flag"]; !ok || val != "true" {
+					t.Errorf("expected extra_args[another_bool_flag]='true', got '%s'", val)
+				}
+			},
+		},
+		{
+			name:      "extra args with alternative field names",
+			command:   "llama-server -m /model.gguf -ngl 16 --custom-arg test --new-feature",
+			expectErr: false,
+			validate: func(t *testing.T, opts *backends.LlamaServerOptions) {
+				// Check that alternative names worked for known fields
+				if opts.Model != "/model.gguf" {
+					t.Errorf("expected model '/model.gguf', got '%s'", opts.Model)
+				}
+				if opts.GPULayers != 16 {
+					t.Errorf("expected gpu_layers 16, got %d", opts.GPULayers)
+				}
+				// Check that unknown args went to ExtraArgs
+				if opts.ExtraArgs == nil {
+					t.Fatal("expected extra_args to be non-nil")
+				}
+				if val, ok := opts.ExtraArgs["custom_arg"]; !ok || val != "test" {
+					t.Errorf("expected extra_args[custom_arg]='test', got '%s'", val)
+				}
+				if val, ok := opts.ExtraArgs["new_feature"]; !ok || val != "true" {
+					t.Errorf("expected extra_args[new_feature]='true', got '%s'", val)
+				}
+			},
+		},
+		{
+			name:      "only extra args",
+			command:   "llama-server --experimental-feature --beta-mode enabled",
+			expectErr: false,
+			validate: func(t *testing.T, opts *backends.LlamaServerOptions) {
+				if opts.ExtraArgs == nil {
+					t.Fatal("expected extra_args to be non-nil")
+				}
+				if val, ok := opts.ExtraArgs["experimental_feature"]; !ok || val != "true" {
+					t.Errorf("expected extra_args[experimental_feature]='true', got '%s'", val)
+				}
+				if val, ok := opts.ExtraArgs["beta_mode"]; !ok || val != "enabled" {
+					t.Errorf("expected extra_args[beta_mode]='enabled', got '%s'", val)
+				}
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var opts backends.LlamaServerOptions
+			result, err := opts.ParseCommand(tt.command)
+
+			if tt.expectErr && err == nil {
+				t.Error("expected error but got none")
+				return
+			}
+			if !tt.expectErr && err != nil {
+				t.Errorf("unexpected error: %v", err)
+				return
+			}
+
+			if !tt.expectErr && tt.validate != nil {
+				llamaOpts, ok := result.(*backends.LlamaServerOptions)
+				if !ok {
+					t.Fatal("result is not *LlamaServerOptions")
+				}
+				tt.validate(t, llamaOpts)
+			}
+		})
+	}
+}
+func TestLlamaCppGetCommand_WithOverrides(t *testing.T) {
+	tests := []struct {
+		name            string
+		dockerInConfig  bool
+		dockerEnabled   *bool
+		commandOverride string
+		expected        string
+	}{
+		{
+			name:            "no overrides - use config command",
+			dockerInConfig:  false,
+			dockerEnabled:   nil,
+			commandOverride: "",
+			expected:        "/usr/bin/llama-server",
+		},
+		{
+			name:            "override to enable docker",
+			dockerInConfig:  false,
+			dockerEnabled:   boolPtr(true),
+			commandOverride: "",
+			expected:        "docker",
+		},
+		{
+			name:            "override to disable docker",
+			dockerInConfig:  true,
+			dockerEnabled:   boolPtr(false),
+			commandOverride: "",
+			expected:        "/usr/bin/llama-server",
+		},
+		{
+			name:            "command override",
+			dockerInConfig:  false,
+			dockerEnabled:   nil,
+			commandOverride: "/custom/llama-server",
+			expected:        "/custom/llama-server",
+		},
+		{
+			name:            "docker takes precedence over command override",
+			dockerInConfig:  false,
+			dockerEnabled:   boolPtr(true),
+			commandOverride: "/custom/llama-server",
+			expected:        "docker",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			backendConfig := &config.BackendConfig{
+				LlamaCpp: config.BackendSettings{
+					Command: "/usr/bin/llama-server",
+					Docker: &config.DockerSettings{
+						Enabled: tt.dockerInConfig,
+						Image:   "test-image",
+					},
+				},
+			}
+
+			opts := backends.Options{
+				BackendType: backends.BackendTypeLlamaCpp,
+				LlamaServerOptions: &backends.LlamaServerOptions{
+					Model: "test-model.gguf",
+				},
+			}
+
+			result := opts.GetCommand(backendConfig, tt.dockerEnabled, tt.commandOverride)
+			if result != tt.expected {
+				t.Errorf("GetCommand() = %v, want %v", result, tt.expected)
+			}
+		})
+	}
+}
+
+// Helper function to create bool pointer
+func boolPtr(b bool) *bool {
+	return &b
+}

pkg/backends/mlx.go

@@ -1,6 +1,7 @@
 package backends
 
 import (
+	"encoding/json"
 	"fmt"
 	"llamactl/pkg/validation"
 )
@@ -29,6 +30,46 @@ type MlxServerOptions struct {
 	TopK      int     `json:"top_k,omitempty"`
 	MinP      float64 `json:"min_p,omitempty"`
 	MaxTokens int     `json:"max_tokens,omitempty"`
+
+	// ExtraArgs are additional command line arguments.
+	// Example: {"verbose": "", "log-file": "/logs/mlx.log"}
+	ExtraArgs map[string]string `json:"extra_args,omitempty"`
+}
+
+// UnmarshalJSON implements custom JSON unmarshaling to collect unknown fields into ExtraArgs
+func (o *MlxServerOptions) UnmarshalJSON(data []byte) error {
+	// First unmarshal into a map to capture all fields
+	var raw map[string]any
+	if err := json.Unmarshal(data, &raw); err != nil {
+		return err
+	}
+
+	// Create a temporary struct for standard unmarshaling
+	type tempOptions MlxServerOptions
+	temp := tempOptions{}
+
+	// Standard unmarshal first
+	if err := json.Unmarshal(data, &temp); err != nil {
+		return err
+	}
+
+	// Copy to our struct
+	*o = MlxServerOptions(temp)
+
+	// Get all known canonical field names from struct tags
+	knownFields := getKnownFieldNames(o)
+
+	// Collect unknown fields into ExtraArgs
+	if o.ExtraArgs == nil {
+		o.ExtraArgs = make(map[string]string)
+	}
+	for key, value := range raw {
+		if !knownFields[key] {
+			o.ExtraArgs[key] = fmt.Sprintf("%v", value)
+		}
+	}
+
+	return nil
 }
 
 func (o *MlxServerOptions) GetPort() int {
@@ -57,13 +98,30 @@ func (o *MlxServerOptions) Validate() error {
 		return validation.ValidationError(fmt.Errorf("invalid port range: %d", o.Port))
 	}
 
+	// Validate extra_args keys and values
+	for key, value := range o.ExtraArgs {
+		if err := validation.ValidateStringForInjection(key); err != nil {
+			return validation.ValidationError(fmt.Errorf("extra_args key %q: %w", key, err))
+		}
+		if value != "" {
+			if err := validation.ValidateStringForInjection(value); err != nil {
+				return validation.ValidationError(fmt.Errorf("extra_args value for %q: %w", key, err))
+			}
+		}
+	}
+
 	return nil
 }
 
 // BuildCommandArgs converts to command line arguments
 func (o *MlxServerOptions) BuildCommandArgs() []string {
 	multipleFlags := map[string]struct{}{} // MLX doesn't currently have []string fields
-	return BuildCommandArgs(o, multipleFlags)
+	args := BuildCommandArgs(o, multipleFlags)
+
+	// Append extra args at the end
+	args = append(args, convertExtraArgsToFlags(o.ExtraArgs)...)
+
+	return args
 }
 
 func (o *MlxServerOptions) BuildDockerArgs() []string {

pkg/backends/mlx_test.go

@@ -2,6 +2,7 @@ package backends_test
 
 import (
 	"llamactl/pkg/backends"
+	"llamactl/pkg/config"
 	"llamactl/pkg/testutil"
 	"testing"
 )
@@ -202,3 +203,129 @@ func TestMlxBuildCommandArgs_ZeroValues(t *testing.T) {
 		}
 	}
 }
+
+func TestParseMlxCommand_ExtraArgs(t *testing.T) {
+	tests := []struct {
+		name      string
+		command   string
+		expectErr bool
+		validate  func(*testing.T, *backends.MlxServerOptions)
+	}{
+		{
+			name:      "extra args with known fields",
+			command:   "mlx_lm.server --model /path/to/model --port 8080 --unknown-flag value --new-bool-flag",
+			expectErr: false,
+			validate: func(t *testing.T, opts *backends.MlxServerOptions) {
+				if opts.Model != "/path/to/model" {
+					t.Errorf("expected model '/path/to/model', got '%s'", opts.Model)
+				}
+				if opts.Port != 8080 {
+					t.Errorf("expected port 8080, got %d", opts.Port)
+				}
+				if opts.ExtraArgs == nil {
+					t.Fatal("expected extra_args to be non-nil")
+				}
+				if val, ok := opts.ExtraArgs["unknown_flag"]; !ok || val != "value" {
+					t.Errorf("expected extra_args[unknown_flag]='value', got '%s'", val)
+				}
+				if val, ok := opts.ExtraArgs["new_bool_flag"]; !ok || val != "true" {
+					t.Errorf("expected extra_args[new_bool_flag]='true', got '%s'", val)
+				}
+			},
+		},
+		{
+			name:      "only extra args",
+			command:   "mlx_lm.server --experimental-feature --custom-param test",
+			expectErr: false,
+			validate: func(t *testing.T, opts *backends.MlxServerOptions) {
+				if opts.ExtraArgs == nil {
+					t.Fatal("expected extra_args to be non-nil")
+				}
+				if val, ok := opts.ExtraArgs["experimental_feature"]; !ok || val != "true" {
+					t.Errorf("expected extra_args[experimental_feature]='true', got '%s'", val)
+				}
+				if val, ok := opts.ExtraArgs["custom_param"]; !ok || val != "test" {
+					t.Errorf("expected extra_args[custom_param]='test', got '%s'", val)
+				}
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var opts backends.MlxServerOptions
+			result, err := opts.ParseCommand(tt.command)
+
+			if tt.expectErr && err == nil {
+				t.Error("expected error but got none")
+				return
+			}
+			if !tt.expectErr && err != nil {
+				t.Errorf("unexpected error: %v", err)
+				return
+			}
+
+			if !tt.expectErr && tt.validate != nil {
+				mlxOpts, ok := result.(*backends.MlxServerOptions)
+				if !ok {
+					t.Fatal("result is not *MlxServerOptions")
+				}
+				tt.validate(t, mlxOpts)
+			}
+		})
+	}
+}
+func TestMlxGetCommand_NoDocker(t *testing.T) {
+	// MLX backend should never use Docker
+	backendConfig := &config.BackendConfig{
+		MLX: config.BackendSettings{
+			Command: "/usr/bin/mlx-server",
+			Docker: &config.DockerSettings{
+				Enabled: true, // Even if enabled in config
+				Image:   "test-image",
+			},
+		},
+	}
+
+	opts := backends.Options{
+		BackendType: backends.BackendTypeMlxLm,
+		MlxServerOptions: &backends.MlxServerOptions{
+			Model: "test-model",
+		},
+	}
+
+	tests := []struct {
+		name            string
+		dockerEnabled   *bool
+		commandOverride string
+		expected        string
+	}{
+		{
+			name:            "ignores docker in config",
+			dockerEnabled:   nil,
+			commandOverride: "",
+			expected:        "/usr/bin/mlx-server",
+		},
+		{
+			name:            "ignores docker override",
+			dockerEnabled:   boolPtr(true),
+			commandOverride: "",
+			expected:        "/usr/bin/mlx-server",
+		},
+		{
+			name:            "respects command override",
+			dockerEnabled:   nil,
+			commandOverride: "/custom/mlx-server",
+			expected:        "/custom/mlx-server",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := opts.GetCommand(backendConfig, tt.dockerEnabled, tt.commandOverride)
+			if result != tt.expected {
+				t.Errorf("GetCommand() = %v, want %v", result, tt.expected)
+			}
+		})
+	}
+}

pkg/backends/parser.go

@@ -4,6 +4,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"path/filepath"
+	"reflect"
 	"regexp"
 	"strconv"
 	"strings"
@@ -211,3 +212,65 @@ func parseValue(value string) any {
 	// Return as string
 	return value
 }
+
+// setFieldValue sets a field value using reflection, handling type conversions
+// Used by UnmarshalJSON implementations to handle alternative field names
+func setFieldValue(field reflect.Value, value any) {
+	switch field.Kind() {
+	case reflect.Int:
+		if intVal, ok := value.(float64); ok {
+			field.SetInt(int64(intVal))
+		} else if strVal, ok := value.(string); ok {
+			if intVal, err := strconv.Atoi(strVal); err == nil {
+				field.SetInt(int64(intVal))
+			}
+		}
+	case reflect.Float64:
+		if floatVal, ok := value.(float64); ok {
+			field.SetFloat(floatVal)
+		} else if strVal, ok := value.(string); ok {
+			if floatVal, err := strconv.ParseFloat(strVal, 64); err == nil {
+				field.SetFloat(floatVal)
+			}
+		}
+	case reflect.String:
+		if strVal, ok := value.(string); ok {
+			field.SetString(strVal)
+		}
+	case reflect.Bool:
+		if boolVal, ok := value.(bool); ok {
+			field.SetBool(boolVal)
+		}
+	case reflect.Slice:
+		// Handle string slices
+		if field.Type().Elem().Kind() == reflect.String {
+			if slice, ok := value.([]any); ok {
+				strSlice := make([]string, 0, len(slice))
+				for _, v := range slice {
+					if s, ok := v.(string); ok {
+						strSlice = append(strSlice, s)
+					}
+				}
+				field.Set(reflect.ValueOf(strSlice))
+			}
+		}
+	}
+}
+
+// getKnownFieldNames extracts all known field names from struct json tags
+// Used by UnmarshalJSON implementations to identify unknown fields for ExtraArgs
+func getKnownFieldNames(v any) map[string]bool {
+	fields := make(map[string]bool)
+	t := reflect.TypeOf(v).Elem()
+
+	for i := 0; i < t.NumField(); i++ {
+		field := t.Field(i)
+		jsonTag := field.Tag.Get("json")
+		if jsonTag != "" && jsonTag != "-" {
+			// Handle "name,omitempty" format
+			name := strings.Split(jsonTag, ",")[0]
+			fields[name] = true
+		}
+	}
+	return fields
+}

pkg/backends/vllm.go

@@ -1,6 +1,7 @@
 package backends
 
 import (
+	"encoding/json"
 	"fmt"
 	"llamactl/pkg/validation"
 )
@@ -142,6 +143,46 @@ type VllmServerOptions struct {
 	OverridePoolingConfig     string `json:"override_pooling_config,omitempty"`
 	OverrideNeuronConfig      string `json:"override_neuron_config,omitempty"`
 	OverrideKVCacheALIGNSize  int    `json:"override_kv_cache_align_size,omitempty"`
+
+	// ExtraArgs are additional command line arguments.
+	// Example: {"verbose": "", "log-file": "/logs/vllm.log"}
+	ExtraArgs map[string]string `json:"extra_args,omitempty"`
+}
+
+// UnmarshalJSON implements custom JSON unmarshaling to collect unknown fields into ExtraArgs
+func (o *VllmServerOptions) UnmarshalJSON(data []byte) error {
+	// First unmarshal into a map to capture all fields
+	var raw map[string]any
+	if err := json.Unmarshal(data, &raw); err != nil {
+		return err
+	}
+
+	// Create a temporary struct for standard unmarshaling
+	type tempOptions VllmServerOptions
+	temp := tempOptions{}
+
+	// Standard unmarshal first
+	if err := json.Unmarshal(data, &temp); err != nil {
+		return err
+	}
+
+	// Copy to our struct
+	*o = VllmServerOptions(temp)
+
+	// Get all known canonical field names from struct tags
+	knownFields := getKnownFieldNames(o)
+
+	// Collect unknown fields into ExtraArgs
+	if o.ExtraArgs == nil {
+		o.ExtraArgs = make(map[string]string)
+	}
+	for key, value := range raw {
+		if !knownFields[key] {
+			o.ExtraArgs[key] = fmt.Sprintf("%v", value)
+		}
+	}
+
+	return nil
 }
 
 func (o *VllmServerOptions) GetPort() int {
@@ -171,6 +212,18 @@ func (o *VllmServerOptions) Validate() error {
 		return validation.ValidationError(fmt.Errorf("invalid port range: %d", o.Port))
 	}
 
+	// Validate extra_args keys and values
+	for key, value := range o.ExtraArgs {
+		if err := validation.ValidateStringForInjection(key); err != nil {
+			return validation.ValidationError(fmt.Errorf("extra_args key %q: %w", key, err))
+		}
+		if value != "" {
+			if err := validation.ValidateStringForInjection(value); err != nil {
+				return validation.ValidationError(fmt.Errorf("extra_args value for %q: %w", key, err))
+			}
+		}
+	}
+
 	return nil
 }
 
@@ -193,6 +246,9 @@ func (o *VllmServerOptions) BuildCommandArgs() []string {
 	flagArgs := BuildCommandArgs(&optionsCopy, vllmMultiValuedFlags)
 	args = append(args, flagArgs...)
 
+	// Append extra args at the end
+	args = append(args, convertExtraArgsToFlags(o.ExtraArgs)...)
+
 	return args
 }
 
@@ -203,6 +259,9 @@ func (o *VllmServerOptions) BuildDockerArgs() []string {
 	flagArgs := BuildCommandArgs(o, vllmMultiValuedFlags)
 	args = append(args, flagArgs...)
 
+	// Append extra args at the end
+	args = append(args, convertExtraArgsToFlags(o.ExtraArgs)...)
+
 	return args
 }
 

pkg/backends/vllm_test.go

@@ -321,3 +321,94 @@ func TestVllmBuildCommandArgs_PositionalModel(t *testing.T) {
 		t.Errorf("Expected --port 8080 not found in %v", args)
 	}
 }
+
+func TestParseVllmCommand_ExtraArgs(t *testing.T) {
+	tests := []struct {
+		name      string
+		command   string
+		expectErr bool
+		validate  func(*testing.T, *backends.VllmServerOptions)
+	}{
+		{
+			name:      "extra args with known fields",
+			command:   "vllm serve llama-model --tensor-parallel-size 2 --unknown-flag value --new-bool-flag",
+			expectErr: false,
+			validate: func(t *testing.T, opts *backends.VllmServerOptions) {
+				if opts.Model != "llama-model" {
+					t.Errorf("expected model 'llama-model', got '%s'", opts.Model)
+				}
+				if opts.TensorParallelSize != 2 {
+					t.Errorf("expected tensor_parallel_size 2, got %d", opts.TensorParallelSize)
+				}
+				if opts.ExtraArgs == nil {
+					t.Fatal("expected extra_args to be non-nil")
+				}
+				if val, ok := opts.ExtraArgs["unknown_flag"]; !ok || val != "value" {
+					t.Errorf("expected extra_args[unknown_flag]='value', got '%s'", val)
+				}
+				if val, ok := opts.ExtraArgs["new_bool_flag"]; !ok || val != "true" {
+					t.Errorf("expected extra_args[new_bool_flag]='true', got '%s'", val)
+				}
+			},
+		},
+		{
+			name:      "only extra args",
+			command:   "vllm serve model --experimental-feature --custom-param test",
+			expectErr: false,
+			validate: func(t *testing.T, opts *backends.VllmServerOptions) {
+				if opts.ExtraArgs == nil {
+					t.Fatal("expected extra_args to be non-nil")
+				}
+				if val, ok := opts.ExtraArgs["experimental_feature"]; !ok || val != "true" {
+					t.Errorf("expected extra_args[experimental_feature]='true', got '%s'", val)
+				}
+				if val, ok := opts.ExtraArgs["custom_param"]; !ok || val != "test" {
+					t.Errorf("expected extra_args[custom_param]='test', got '%s'", val)
+				}
+			},
+		},
+		{
+			name:      "extra args without model positional",
+			command:   "vllm serve --model my-model --new-feature enabled --beta-flag",
+			expectErr: false,
+			validate: func(t *testing.T, opts *backends.VllmServerOptions) {
+				if opts.Model != "my-model" {
+					t.Errorf("expected model 'my-model', got '%s'", opts.Model)
+				}
+				if opts.ExtraArgs == nil {
+					t.Fatal("expected extra_args to be non-nil")
+				}
+				if val, ok := opts.ExtraArgs["new_feature"]; !ok || val != "enabled" {
+					t.Errorf("expected extra_args[new_feature]='enabled', got '%s'", val)
+				}
+				if val, ok := opts.ExtraArgs["beta_flag"]; !ok || val != "true" {
+					t.Errorf("expected extra_args[beta_flag]='true', got '%s'", val)
+				}
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var opts backends.VllmServerOptions
+			result, err := opts.ParseCommand(tt.command)
+
+			if tt.expectErr && err == nil {
+				t.Error("expected error but got none")
+				return
+			}
+			if !tt.expectErr && err != nil {
+				t.Errorf("unexpected error: %v", err)
+				return
+			}
+
+			if !tt.expectErr && tt.validate != nil {
+				vllmOpts, ok := result.(*backends.VllmServerOptions)
+				if !ok {
+					t.Fatal("result is not *VllmServerOptions")
+				}
+				tt.validate(t, vllmOpts)
+			}
+		})
+	}
+}

pkg/config/config.go

@@ -1,6 +1,8 @@
 package config
 
 import (
+	"encoding/json"
+	"fmt"
 	"log"
 	"os"
 	"path/filepath"
@@ -13,126 +15,126 @@ import (
 
 // BackendSettings contains structured backend configuration
 type BackendSettings struct {
-	Command         string            `yaml:"command"`
-	Args            []string          `yaml:"args"`
-	Environment     map[string]string `yaml:"environment,omitempty"`
-	Docker          *DockerSettings   `yaml:"docker,omitempty"`
-	ResponseHeaders map[string]string `yaml:"response_headers,omitempty"`
+	Command         string            `yaml:"command" json:"command"`
+	Args            []string          `yaml:"args" json:"args"`
+	Environment     map[string]string `yaml:"environment,omitempty" json:"environment,omitempty"`
+	Docker          *DockerSettings   `yaml:"docker,omitempty" json:"docker,omitempty"`
+	ResponseHeaders map[string]string `yaml:"response_headers,omitempty" json:"response_headers,omitempty"`
 }
 
 // DockerSettings contains Docker-specific configuration
 type DockerSettings struct {
-	Enabled     bool              `yaml:"enabled"`
-	Image       string            `yaml:"image"`
-	Args        []string          `yaml:"args"`
-	Environment map[string]string `yaml:"environment,omitempty"`
+	Enabled     bool              `yaml:"enabled" json:"enabled"`
+	Image       string            `yaml:"image" json:"image"`
+	Args        []string          `yaml:"args" json:"args"`
+	Environment map[string]string `yaml:"environment,omitempty" json:"environment,omitempty"`
 }
 
 // BackendConfig contains backend executable configurations
 type BackendConfig struct {
-	LlamaCpp BackendSettings `yaml:"llama-cpp"`
-	VLLM     BackendSettings `yaml:"vllm"`
-	MLX      BackendSettings `yaml:"mlx"`
+	LlamaCpp BackendSettings `yaml:"llama-cpp" json:"llama-cpp"`
+	VLLM     BackendSettings `yaml:"vllm" json:"vllm"`
+	MLX      BackendSettings `yaml:"mlx" json:"mlx"`
 }
 
 // AppConfig represents the configuration for llamactl
 type AppConfig struct {
-	Server     ServerConfig          `yaml:"server"`
-	Backends   BackendConfig         `yaml:"backends"`
-	Instances  InstancesConfig       `yaml:"instances"`
-	Auth       AuthConfig            `yaml:"auth"`
-	LocalNode  string                `yaml:"local_node,omitempty"`
-	Nodes      map[string]NodeConfig `yaml:"nodes,omitempty"`
-	Version    string                `yaml:"-"`
-	CommitHash string                `yaml:"-"`
-	BuildTime  string                `yaml:"-"`
+	Server     ServerConfig          `yaml:"server" json:"server"`
+	Backends   BackendConfig         `yaml:"backends" json:"backends"`
+	Instances  InstancesConfig       `yaml:"instances" json:"instances"`
+	Auth       AuthConfig            `yaml:"auth" json:"auth"`
+	LocalNode  string                `yaml:"local_node,omitempty" json:"local_node,omitempty"`
+	Nodes      map[string]NodeConfig `yaml:"nodes,omitempty" json:"nodes,omitempty"`
+	Version    string                `yaml:"-" json:"version"`
+	CommitHash string                `yaml:"-" json:"commit_hash"`
+	BuildTime  string                `yaml:"-" json:"build_time"`
 }
 
 // ServerConfig contains HTTP server configuration
 type ServerConfig struct {
 	// Server host to bind to
-	Host string `yaml:"host"`
+	Host string `yaml:"host" json:"host"`
 
 	// Server port to bind to
-	Port int `yaml:"port"`
+	Port int `yaml:"port" json:"port"`
 
 	// Allowed origins for CORS (e.g., "http://localhost:3000")
-	AllowedOrigins []string `yaml:"allowed_origins"`
+	AllowedOrigins []string `yaml:"allowed_origins" json:"allowed_origins"`
 
 	// Allowed headers for CORS (e.g., "Accept", "Authorization", "Content-Type", "X-CSRF-Token")
-	AllowedHeaders []string `yaml:"allowed_headers"`
+	AllowedHeaders []string `yaml:"allowed_headers" json:"allowed_headers"`
 
 	// Enable Swagger UI for API documentation
-	EnableSwagger bool `yaml:"enable_swagger"`
+	EnableSwagger bool `yaml:"enable_swagger" json:"enable_swagger"`
 
 	// Response headers to send with responses
-	ResponseHeaders map[string]string `yaml:"response_headers,omitempty"`
+	ResponseHeaders map[string]string `yaml:"response_headers,omitempty" json:"response_headers,omitempty"`
 }
 
 // InstancesConfig contains instance management configuration
 type InstancesConfig struct {
 	// Port range for instances (e.g., 8000,9000)
-	PortRange [2]int `yaml:"port_range"`
+	PortRange [2]int `yaml:"port_range" json:"port_range"`
 
 	// Directory where all llamactl data will be stored (instances.json, logs, etc.)
-	DataDir string `yaml:"data_dir"`
+	DataDir string `yaml:"data_dir" json:"data_dir"`
 
 	// Instance config directory override
-	InstancesDir string `yaml:"configs_dir"`
+	InstancesDir string `yaml:"configs_dir" json:"configs_dir"`
 
 	// Logs directory override
-	LogsDir string `yaml:"logs_dir"`
+	LogsDir string `yaml:"logs_dir" json:"logs_dir"`
 
 	// Automatically create the data directory if it doesn't exist
-	AutoCreateDirs bool `yaml:"auto_create_dirs"`
+	AutoCreateDirs bool `yaml:"auto_create_dirs" json:"auto_create_dirs"`
 
 	// Maximum number of instances that can be created
-	MaxInstances int `yaml:"max_instances"`
+	MaxInstances int `yaml:"max_instances" json:"max_instances"`
 
 	// Maximum number of instances that can be running at the same time
-	MaxRunningInstances int `yaml:"max_running_instances,omitempty"`
+	MaxRunningInstances int `yaml:"max_running_instances,omitempty" json:"max_running_instances,omitempty"`
 
 	// Enable LRU eviction for instance logs
-	EnableLRUEviction bool `yaml:"enable_lru_eviction"`
+	EnableLRUEviction bool `yaml:"enable_lru_eviction" json:"enable_lru_eviction"`
 
 	// Default auto-restart setting for new instances
-	DefaultAutoRestart bool `yaml:"default_auto_restart"`
+	DefaultAutoRestart bool `yaml:"default_auto_restart" json:"default_auto_restart"`
 
 	// Default max restarts for new instances
-	DefaultMaxRestarts int `yaml:"default_max_restarts"`
+	DefaultMaxRestarts int `yaml:"default_max_restarts" json:"default_max_restarts"`
 
 	// Default restart delay for new instances (in seconds)
-	DefaultRestartDelay int `yaml:"default_restart_delay"`
+	DefaultRestartDelay int `yaml:"default_restart_delay" json:"default_restart_delay"`
 
 	// Default on-demand start setting for new instances
-	DefaultOnDemandStart bool `yaml:"default_on_demand_start"`
+	DefaultOnDemandStart bool `yaml:"default_on_demand_start" json:"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"`
+	OnDemandStartTimeout int `yaml:"on_demand_start_timeout,omitempty" json:"on_demand_start_timeout,omitempty"`
 
 	// Interval for checking instance timeouts (in minutes)
-	TimeoutCheckInterval int `yaml:"timeout_check_interval"`
+	TimeoutCheckInterval int `yaml:"timeout_check_interval" json:"timeout_check_interval"`
 }
 
 // AuthConfig contains authentication settings
 type AuthConfig struct {
 
 	// Require authentication for OpenAI compatible inference endpoints
-	RequireInferenceAuth bool `yaml:"require_inference_auth"`
+	RequireInferenceAuth bool `yaml:"require_inference_auth" json:"require_inference_auth"`
 
 	// List of keys for OpenAI compatible inference endpoints
-	InferenceKeys []string `yaml:"inference_keys"`
+	InferenceKeys []string `yaml:"inference_keys" json:"inference_keys"`
 
 	// Require authentication for management endpoints
-	RequireManagementAuth bool `yaml:"require_management_auth"`
+	RequireManagementAuth bool `yaml:"require_management_auth" json:"require_management_auth"`
 
 	// List of keys for management endpoints
-	ManagementKeys []string `yaml:"management_keys"`
+	ManagementKeys []string `yaml:"management_keys" json:"management_keys"`
 }
 
 type NodeConfig struct {
-	Address string `yaml:"address"`
-	APIKey  string `yaml:"api_key,omitempty"`
+	Address string `yaml:"address" json:"address"`
+	APIKey  string `yaml:"api_key,omitempty" json:"api_key,omitempty"`
 }
 
 // LoadConfig loads configuration with the following precedence:
@@ -231,6 +233,11 @@ func LoadConfig(configPath string) (AppConfig, error) {
 		cfg.Instances.LogsDir = filepath.Join(cfg.Instances.DataDir, "logs")
 	}
 
+	// Validate port range
+	if cfg.Instances.PortRange[0] <= 0 || cfg.Instances.PortRange[1] <= 0 || cfg.Instances.PortRange[0] >= cfg.Instances.PortRange[1] {
+		return AppConfig{}, fmt.Errorf("invalid port range: %v", cfg.Instances.PortRange)
+	}
+
 	return cfg, nil
 }
 
@@ -604,3 +611,31 @@ func getDefaultConfigLocations() []string {
 
 	return locations
 }
+
+// SanitizedCopy returns a copy of the AppConfig with sensitive information removed
+func (cfg *AppConfig) SanitizedCopy() (AppConfig, error) {
+	// Deep copy via JSON marshal/unmarshal to avoid concurrent map access
+	data, err := json.Marshal(cfg)
+	if err != nil {
+		log.Printf("Failed to marshal config for sanitization: %v", err)
+		return AppConfig{}, err
+	}
+
+	var sanitized AppConfig
+	if err := json.Unmarshal(data, &sanitized); err != nil {
+		log.Printf("Failed to unmarshal config for sanitization: %v", err)
+		return AppConfig{}, err
+	}
+
+	// Clear sensitive information
+	sanitized.Auth.InferenceKeys = []string{}
+	sanitized.Auth.ManagementKeys = []string{}
+
+	// Clear API keys from nodes
+	for nodeName, node := range sanitized.Nodes {
+		node.APIKey = ""
+		sanitized.Nodes[nodeName] = node
+	}
+
+	return sanitized, nil
+}

pkg/instance/instance.go

@@ -5,7 +5,7 @@ import (
 	"fmt"
 	"llamactl/pkg/config"
 	"log"
-	"net/http/httputil"
+	"net/http"
 	"time"
 )
 
@@ -182,15 +182,6 @@ func (i *Instance) GetPort() int {
 	return i.options.GetPort()
 }
 
-// GetProxy returns the reverse proxy for this instance
-func (i *Instance) GetProxy() (*httputil.ReverseProxy, error) {
-	if i.proxy == nil {
-		return nil, fmt.Errorf("instance %s has no proxy component", i.Name)
-	}
-
-	return i.proxy.get()
-}
-
 func (i *Instance) IsRemote() bool {
 	opts := i.GetOptions()
 	if opts == nil {
@@ -242,13 +233,29 @@ func (i *Instance) ShouldTimeout() bool {
 	return i.proxy.shouldTimeout()
 }
 
+// GetInflightRequests returns the current number of inflight requests
+func (i *Instance) GetInflightRequests() int32 {
+	if i.proxy == nil {
+		return 0
+	}
+	return i.proxy.getInflightRequests()
+}
+
+// ServeHTTP serves HTTP requests through the proxy with request tracking and shutdown handling
+func (i *Instance) ServeHTTP(w http.ResponseWriter, r *http.Request) error {
+	if i.proxy == nil {
+		return fmt.Errorf("instance %s has no proxy component", i.Name)
+	}
+	return i.proxy.serveHTTP(w, r)
+}
+
 func (i *Instance) getCommand() string {
 	opts := i.GetOptions()
 	if opts == nil {
 		return ""
 	}
 
-	return opts.BackendOptions.GetCommand(i.globalBackendSettings)
+	return opts.BackendOptions.GetCommand(i.globalBackendSettings, opts.DockerEnabled, opts.CommandOverride)
 }
 
 func (i *Instance) buildCommandArgs() []string {
@@ -257,7 +264,7 @@ func (i *Instance) buildCommandArgs() []string {
 		return nil
 	}
 
-	return opts.BackendOptions.BuildCommandArgs(i.globalBackendSettings)
+	return opts.BackendOptions.BuildCommandArgs(i.globalBackendSettings, opts.DockerEnabled)
 }
 
 func (i *Instance) buildEnvironment() map[string]string {
@@ -266,29 +273,21 @@ func (i *Instance) buildEnvironment() map[string]string {
 		return nil
 	}
 
-	return opts.BackendOptions.BuildEnvironment(i.globalBackendSettings, opts.Environment)
+	return opts.BackendOptions.BuildEnvironment(i.globalBackendSettings, opts.DockerEnabled, opts.Environment)
 }
 
 // MarshalJSON implements json.Marshaler for Instance
 func (i *Instance) MarshalJSON() ([]byte, error) {
-	// Get options
-	opts := i.GetOptions()
-
-	// Determine if docker is enabled for this instance's backend
-	dockerEnabled := opts.BackendOptions.IsDockerEnabled(i.globalBackendSettings)
-
 	return json.Marshal(&struct {
-		Name          string   `json:"name"`
-		Status        *status  `json:"status"`
-		Created       int64    `json:"created,omitempty"`
-		Options       *options `json:"options,omitempty"`
-		DockerEnabled bool     `json:"docker_enabled,omitempty"`
+		Name    string   `json:"name"`
+		Status  *status  `json:"status"`
+		Created int64    `json:"created,omitempty"`
+		Options *options `json:"options,omitempty"`
 	}{
-		Name:          i.Name,
-		Status:        i.status,
-		Created:       i.Created,
-		Options:       i.options,
-		DockerEnabled: dockerEnabled,
+		Name:    i.Name,
+		Status:  i.status,
+		Created: i.Created,
+		Options: i.options,
 	})
 }
 

pkg/instance/instance_test.go

@@ -171,64 +171,6 @@ func TestSetOptions(t *testing.T) {
 	}
 }
 
-func TestGetProxy(t *testing.T) {
-	globalConfig := &config.AppConfig{
-		Backends: config.BackendConfig{
-			LlamaCpp: config.BackendSettings{
-				Command: "llama-server",
-				Args:    []string{},
-			},
-			MLX: config.BackendSettings{
-				Command: "mlx_lm.server",
-				Args:    []string{},
-			},
-			VLLM: config.BackendSettings{
-				Command: "vllm",
-				Args:    []string{"serve"},
-			},
-		},
-		Instances: config.InstancesConfig{
-			LogsDir: "/tmp/test",
-		},
-		Nodes:     map[string]config.NodeConfig{},
-		LocalNode: "main",
-	}
-
-	options := &instance.Options{
-		Nodes: map[string]struct{}{"main": {}},
-		BackendOptions: backends.Options{
-			BackendType: backends.BackendTypeLlamaCpp,
-			LlamaServerOptions: &backends.LlamaServerOptions{
-				Host: "localhost",
-				Port: 8080,
-			},
-		},
-	}
-
-	// Mock onStatusChange function
-	mockOnStatusChange := func(oldStatus, newStatus instance.Status) {}
-
-	inst := instance.New("test-instance", globalConfig, options, mockOnStatusChange)
-
-	// 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) {
 	globalConfig := &config.AppConfig{
 		Backends: config.BackendConfig{
@@ -613,11 +555,6 @@ func TestRemoteInstanceOperations(t *testing.T) {
 		t.Error("Expected error when restarting remote instance")
 	}
 
-	// GetProxy should fail for remote instance
-	if _, err := inst.GetProxy(); err != nil {
-		t.Error("Expected no error when getting proxy for remote instance")
-	}
-
 	// GetLogs should fail for remote instance
 	if _, err := inst.GetLogs(10); err == nil {
 		t.Error("Expected error when getting logs for remote instance")

pkg/instance/logger.go

@@ -7,13 +7,14 @@ import (
 	"os"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 )
 
 type logger struct {
 	name        string
 	logDir      string
-	logFile     *os.File
+	logFile     atomic.Pointer[os.File]
 	logFilePath string
 	mu          sync.RWMutex
 }
@@ -47,11 +48,11 @@ func (i *logger) create() error {
 		return fmt.Errorf("failed to create stdout log file: %w", err)
 	}
 
-	i.logFile = logFile
+	i.logFile.Store(logFile)
 
 	// Write a startup marker to both files
 	timestamp := time.Now().Format("2006-01-02 15:04:05")
-	fmt.Fprintf(i.logFile, "\n=== Instance %s started at %s ===\n", i.name, timestamp)
+	fmt.Fprintf(logFile, "\n=== Instance %s started at %s ===\n", i.name, timestamp)
 
 	return nil
 }
@@ -102,11 +103,12 @@ func (i *logger) close() {
 	i.mu.Lock()
 	defer i.mu.Unlock()
 
-	if i.logFile != nil {
+	logFile := i.logFile.Swap(nil)
+	if logFile != nil {
 		timestamp := time.Now().Format("2006-01-02 15:04:05")
-		fmt.Fprintf(i.logFile, "=== Instance %s stopped at %s ===\n\n", i.name, timestamp)
-		i.logFile.Close()
-		i.logFile = nil
+		fmt.Fprintf(logFile, "=== Instance %s stopped at %s ===\n\n", i.name, timestamp)
+		logFile.Sync() // Ensure all buffered data is written to disk
+		logFile.Close()
 	}
 }
 
@@ -117,9 +119,9 @@ func (i *logger) readOutput(reader io.ReadCloser) {
 	scanner := bufio.NewScanner(reader)
 	for scanner.Scan() {
 		line := scanner.Text()
-		if i.logFile != nil {
-			fmt.Fprintln(i.logFile, line)
-			i.logFile.Sync() // Ensure data is written to disk
+		// Use atomic load to avoid lock contention on every line
+		if logFile := i.logFile.Load(); logFile != nil {
+			fmt.Fprintln(logFile, line)
 		}
 	}
 }

pkg/instance/options.go

@@ -5,7 +5,9 @@ import (
 	"fmt"
 	"llamactl/pkg/backends"
 	"llamactl/pkg/config"
+	"llamactl/pkg/validation"
 	"log"
+	"maps"
 	"slices"
 	"sync"
 )
@@ -22,6 +24,11 @@ type Options struct {
 	IdleTimeout *int `json:"idle_timeout,omitempty"` // minutes
 	// Environment variables
 	Environment map[string]string `json:"environment,omitempty"`
+
+	// Execution context overrides
+	DockerEnabled   *bool  `json:"docker_enabled,omitempty"`
+	CommandOverride string `json:"command_override,omitempty"`
+
 	// Assigned nodes
 	Nodes map[string]struct{} `json:"-"`
 	// Backend options
@@ -138,15 +145,25 @@ func (c *Options) UnmarshalJSON(data []byte) error {
 
 // MarshalJSON implements custom JSON marshaling for Options
 func (c *Options) MarshalJSON() ([]byte, error) {
-	// Use anonymous struct to avoid recursion
 	type Alias Options
-	aux := struct {
+
+	// Make a copy of the struct
+	temp := *c
+
+	// Copy environment map to avoid concurrent access issues
+	if temp.Environment != nil {
+		envCopy := make(map[string]string, len(temp.Environment))
+		maps.Copy(envCopy, temp.Environment)
+		temp.Environment = envCopy
+	}
+
+	aux := &struct {
 		Nodes          []string             `json:"nodes,omitempty"` // Output as JSON array
 		BackendType    backends.BackendType `json:"backend_type"`
 		BackendOptions map[string]any       `json:"backend_options,omitempty"`
 		*Alias
 	}{
-		Alias: (*Alias)(c),
+		Alias: (*Alias)(&temp),
 	}
 
 	// Convert nodes map to array (sorted for consistency)
@@ -163,13 +180,12 @@ func (c *Options) MarshalJSON() ([]byte, error) {
 	aux.BackendType = c.BackendOptions.BackendType
 
 	// Marshal the backends.Options struct to get the properly formatted backend options
-	// Marshal a pointer to trigger the pointer receiver MarshalJSON method
 	backendData, err := json.Marshal(&c.BackendOptions)
 	if err != nil {
 		return nil, fmt.Errorf("failed to marshal backend options: %w", err)
 	}
 
-	// Unmarshal into a temporary struct to extract the backend_options map
+	// Unmarshal into a new temporary map to extract the backend_options
 	var tempBackend struct {
 		BackendOptions map[string]any `json:"backend_options,omitempty"`
 	}
@@ -200,6 +216,28 @@ func (c *Options) validateAndApplyDefaults(name string, globalSettings *config.I
 		*c.IdleTimeout = 0
 	}
 
+	// Validate docker_enabled and command_override relationship
+	if c.DockerEnabled != nil && *c.DockerEnabled && c.CommandOverride != "" {
+		log.Printf("Instance %s: command_override cannot be set when docker_enabled is true, ignoring command_override", name)
+		c.CommandOverride = "" // Clear invalid configuration
+	}
+
+	// Validate command_override if set
+	if c.CommandOverride != "" {
+		if err := validation.ValidateStringForInjection(c.CommandOverride); err != nil {
+			log.Printf("Instance %s: invalid command_override: %v, clearing value", name, err)
+			c.CommandOverride = "" // Clear invalid value
+		}
+	}
+
+	// Validate docker_enabled for MLX backend
+	if c.BackendOptions.BackendType == backends.BackendTypeMlxLm {
+		if c.DockerEnabled != nil && *c.DockerEnabled {
+			log.Printf("Instance %s: docker_enabled is not supported for MLX backend, ignoring", name)
+			c.DockerEnabled = nil // Clear invalid configuration
+		}
+	}
+
 	// Apply defaults from global settings for nil fields
 	if globalSettings != nil {
 		if c.AutoRestart == nil {

pkg/instance/process.go

@@ -132,14 +132,28 @@ func (p *process) stop() error {
 		p.restartCancel = nil
 	}
 
-	// Set status to stopped first to signal intentional stop
-	p.instance.SetStatus(Stopped)
+	// Set status to ShuttingDown first to reject new requests
+	p.instance.SetStatus(ShuttingDown)
 
 	// Get the monitor done channel before releasing the lock
 	monitorDone := p.monitorDone
 
 	p.mu.Unlock()
 
+	// Wait for inflight requests to complete (max 30 seconds)
+	log.Printf("Instance %s shutting down, waiting for inflight requests to complete...", p.instance.Name)
+	deadline := time.Now().Add(30 * time.Second)
+	for time.Now().Before(deadline) {
+		inflight := p.instance.GetInflightRequests()
+		if inflight == 0 {
+			break
+		}
+		time.Sleep(100 * time.Millisecond)
+	}
+
+	// Now set status to stopped to signal intentional stop
+	p.instance.SetStatus(Stopped)
+
 	// Stop the process with SIGINT if cmd exists
 	if p.cmd != nil && p.cmd.Process != nil {
 		if err := p.cmd.Process.Signal(syscall.SIGINT); err != nil {
@@ -156,6 +170,7 @@ func (p *process) stop() error {
 	select {
 	case <-monitorDone:
 		// Process exited normally
+		log.Printf("Instance %s shut down gracefully", p.instance.Name)
 	case <-time.After(30 * time.Second):
 		// Force kill if it doesn't exit within 30 seconds
 		if p.cmd != nil && p.cmd.Process != nil {

pkg/instance/proxy.go

@@ -37,8 +37,9 @@ type proxy struct {
 	proxyOnce sync.Once
 	proxyErr  error
 
-	lastRequestTime atomic.Int64
-	timeProvider    TimeProvider
+	lastRequestTime  atomic.Int64
+	inflightRequests atomic.Int32
+	timeProvider     TimeProvider
 }
 
 // newProxy creates a new Proxy for the given instance
@@ -153,6 +154,23 @@ func (p *proxy) build() (*httputil.ReverseProxy, error) {
 	return proxy, nil
 }
 
+// serveHTTP handles HTTP requests with inflight tracking
+func (p *proxy) serveHTTP(w http.ResponseWriter, r *http.Request) error {
+	// Get the reverse proxy
+	reverseProxy, err := p.get()
+	if err != nil {
+		return err
+	}
+
+	// Track inflight requests
+	p.incInflightRequests()
+	defer p.decInflightRequests()
+
+	// Serve the request
+	reverseProxy.ServeHTTP(w, r)
+	return nil
+}
+
 // clear resets the proxy, allowing it to be recreated when options change.
 func (p *proxy) clear() {
 	p.mu.Lock()
@@ -160,7 +178,7 @@ func (p *proxy) clear() {
 
 	p.proxy = nil
 	p.proxyErr = nil
-	p.proxyOnce = sync.Once{} // Reset Once for next GetProxy call
+	p.proxyOnce = sync.Once{}
 }
 
 // updateLastRequestTime updates the last request access time for the instance
@@ -199,3 +217,18 @@ func (p *proxy) shouldTimeout() bool {
 func (p *proxy) setTimeProvider(tp TimeProvider) {
 	p.timeProvider = tp
 }
+
+// incInflightRequests increments the inflight request counter
+func (p *proxy) incInflightRequests() {
+	p.inflightRequests.Add(1)
+}
+
+// decInflightRequests decrements the inflight request counter
+func (p *proxy) decInflightRequests() {
+	p.inflightRequests.Add(-1)
+}
+
+// getInflightRequests returns the current number of inflight requests
+func (p *proxy) getInflightRequests() int32 {
+	return p.inflightRequests.Load()
+}

pkg/instance/status.go

@@ -14,20 +14,23 @@ const (
 	Running
 	Failed
 	Restarting
+	ShuttingDown
 )
 
 var nameToStatus = map[string]Status{
-	"stopped":    Stopped,
-	"running":    Running,
-	"failed":     Failed,
-	"restarting": Restarting,
+	"stopped":       Stopped,
+	"running":       Running,
+	"failed":        Failed,
+	"restarting":    Restarting,
+	"shutting_down": ShuttingDown,
 }
 
 var statusToName = map[Status]string{
-	Stopped:    "stopped",
-	Running:    "running",
-	Failed:     "failed",
-	Restarting: "restarting",
+	Stopped:      "stopped",
+	Running:      "running",
+	Failed:       "failed",
+	Restarting:   "restarting",
+	ShuttingDown: "shutting_down",
 }
 
 // Status enum JSON marshaling methods

pkg/manager/lifecycle_test.go

@@ -10,7 +10,7 @@ import (
 )
 
 func TestInstanceTimeoutLogic(t *testing.T) {
-	testManager := createTestManager()
+	testManager := createTestManager(t)
 	defer testManager.Shutdown()
 
 	idleTimeout := 1 // 1 minute
@@ -42,7 +42,7 @@ func TestInstanceTimeoutLogic(t *testing.T) {
 }
 
 func TestInstanceWithoutTimeoutNeverExpires(t *testing.T) {
-	testManager := createTestManager()
+	testManager := createTestManager(t)
 	defer testManager.Shutdown()
 
 	noTimeoutInst := createInstanceWithTimeout(t, testManager, "no-timeout-test", "/path/to/model.gguf", nil)
@@ -64,7 +64,7 @@ func TestInstanceWithoutTimeoutNeverExpires(t *testing.T) {
 }
 
 func TestEvictLRUInstance_Success(t *testing.T) {
-	manager := createTestManager()
+	manager := createTestManager(t)
 	defer manager.Shutdown()
 
 	// Create 3 instances with idle timeout enabled (value doesn't matter for LRU logic)
@@ -121,7 +121,7 @@ func TestEvictLRUInstance_Success(t *testing.T) {
 }
 
 func TestEvictLRUInstance_NoRunningInstances(t *testing.T) {
-	manager := createTestManager()
+	manager := createTestManager(t)
 	defer manager.Shutdown()
 
 	err := manager.EvictLRUInstance()
@@ -134,7 +134,7 @@ func TestEvictLRUInstance_NoRunningInstances(t *testing.T) {
 }
 
 func TestEvictLRUInstance_OnlyEvictsTimeoutEnabledInstances(t *testing.T) {
-	manager := createTestManager()
+	manager := createTestManager(t)
 	defer manager.Shutdown()
 
 	// Create mix of instances: some with timeout enabled, some disabled

pkg/manager/manager.go

@@ -54,16 +54,10 @@ func New(globalConfig *config.AppConfig) InstanceManager {
 
 	// Initialize port allocator
 	portRange := globalConfig.Instances.PortRange
-	ports, err := newPortAllocator(portRange[0], portRange[1])
-	if err != nil {
-		log.Fatalf("Failed to create port allocator: %v", err)
-	}
+	ports := newPortAllocator(portRange[0], portRange[1])
 
 	// Initialize persistence
-	persistence, err := newInstancePersister(globalConfig.Instances.InstancesDir)
-	if err != nil {
-		log.Fatalf("Failed to create instance persister: %v", err)
-	}
+	persistence := newInstancePersister(globalConfig.Instances.InstancesDir)
 
 	// Initialize remote manager
 	remote := newRemoteManager(globalConfig.Nodes, 30*time.Second)
@@ -116,7 +110,7 @@ func (im *instanceManager) Shutdown() {
 				defer wg.Done()
 				fmt.Printf("Stopping instance %s...\n", inst.Name)
 				if err := inst.Stop(); err != nil {
-					fmt.Printf("Error stopping instance %s: %v\n", inst.Name, err)
+					log.Printf("Error stopping instance %s: %v\n", inst.Name, err)
 				}
 			}(inst)
 		}

pkg/manager/manager_test.go

@@ -85,7 +85,7 @@ func TestDeleteInstance_RemovesPersistenceFile(t *testing.T) {
 }
 
 func TestConcurrentAccess(t *testing.T) {
-	mgr := createTestManager()
+	mgr := createTestManager(t)
 	defer mgr.Shutdown()
 
 	// Test concurrent operations
@@ -113,7 +113,7 @@ func TestConcurrentAccess(t *testing.T) {
 	}
 
 	// Concurrent list operations
-	for i := 0; i < 3; i++ {
+	for range 3 {
 		wg.Add(1)
 		go func() {
 			defer wg.Done()
@@ -134,16 +134,17 @@ func TestConcurrentAccess(t *testing.T) {
 
 // Helper functions for test configuration
 func createTestAppConfig(instancesDir string) *config.AppConfig {
-	// Use 'sleep' as a test command instead of 'llama-server'
-	// This allows tests to run in CI environments without requiring actual LLM binaries
-	// The sleep command will be invoked with model paths and other args, which it ignores
+	// Use 'sh -c "sleep 999999"' as a test command instead of 'llama-server'
+	// The shell ignores all additional arguments passed after the command
 	return &config.AppConfig{
 		Backends: config.BackendConfig{
 			LlamaCpp: config.BackendSettings{
-				Command: "sleep",
+				Command: "sh",
+				Args:    []string{"-c", "sleep 999999"},
 			},
 			MLX: config.BackendSettings{
-				Command: "sleep",
+				Command: "sh",
+				Args:    []string{"-c", "sleep 999999"},
 			},
 		},
 		Instances: config.InstancesConfig{
@@ -162,28 +163,8 @@ func createTestAppConfig(instancesDir string) *config.AppConfig {
 	}
 }
 
-func createTestManager() manager.InstanceManager {
-	appConfig := &config.AppConfig{
-		Backends: config.BackendConfig{
-			LlamaCpp: config.BackendSettings{
-				Command: "sleep",
-			},
-			MLX: config.BackendSettings{
-				Command: "sleep",
-			},
-		},
-		Instances: config.InstancesConfig{
-			PortRange:            [2]int{8000, 9000},
-			LogsDir:              "/tmp/test",
-			MaxInstances:         10,
-			MaxRunningInstances:  10,
-			DefaultAutoRestart:   true,
-			DefaultMaxRestarts:   3,
-			DefaultRestartDelay:  5,
-			TimeoutCheckInterval: 5,
-		},
-		LocalNode: "main",
-		Nodes:     map[string]config.NodeConfig{},
-	}
+func createTestManager(t *testing.T) manager.InstanceManager {
+	tempDir := t.TempDir()
+	appConfig := createTestAppConfig(tempDir)
 	return manager.New(appConfig)
 }

pkg/manager/operations.go

@@ -330,7 +330,8 @@ func (im *instanceManager) DeleteInstance(name string) error {
 	lock.Lock()
 	defer im.unlockAndCleanup(name)
 
-	if inst.IsRunning() {
+	status := inst.GetStatus()
+	if status == instance.Running || status == instance.Restarting {
 		return fmt.Errorf("instance with name %s is still running, stop it before deleting", name)
 	}
 

pkg/manager/operations_test.go

@@ -10,7 +10,7 @@ import (
 )
 
 func TestCreateInstance_FailsWithDuplicateName(t *testing.T) {
-	mngr := createTestManager()
+	mngr := createTestManager(t)
 	options := &instance.Options{
 		BackendOptions: backends.Options{
 			BackendType: backends.BackendTypeLlamaCpp,
@@ -36,6 +36,7 @@ func TestCreateInstance_FailsWithDuplicateName(t *testing.T) {
 }
 
 func TestCreateInstance_FailsWhenMaxInstancesReached(t *testing.T) {
+	tempDir := t.TempDir()
 	appConfig := &config.AppConfig{
 		Backends: config.BackendConfig{
 			LlamaCpp: config.BackendSettings{
@@ -44,6 +45,7 @@ func TestCreateInstance_FailsWhenMaxInstancesReached(t *testing.T) {
 		},
 		Instances: config.InstancesConfig{
 			PortRange:            [2]int{8000, 9000},
+			InstancesDir:         tempDir,
 			MaxInstances:         1, // Very low limit for testing
 			TimeoutCheckInterval: 5,
 		},
@@ -77,7 +79,7 @@ func TestCreateInstance_FailsWhenMaxInstancesReached(t *testing.T) {
 }
 
 func TestCreateInstance_FailsWithPortConflict(t *testing.T) {
-	manager := createTestManager()
+	manager := createTestManager(t)
 
 	options1 := &instance.Options{
 		BackendOptions: backends.Options{
@@ -115,7 +117,7 @@ func TestCreateInstance_FailsWithPortConflict(t *testing.T) {
 }
 
 func TestInstanceOperations_FailWithNonExistentInstance(t *testing.T) {
-	manager := createTestManager()
+	manager := createTestManager(t)
 
 	options := &instance.Options{
 		BackendOptions: backends.Options{
@@ -143,7 +145,7 @@ func TestInstanceOperations_FailWithNonExistentInstance(t *testing.T) {
 }
 
 func TestDeleteInstance_RunningInstanceFails(t *testing.T) {
-	mgr := createTestManager()
+	mgr := createTestManager(t)
 	defer mgr.Shutdown()
 
 	options := &instance.Options{
@@ -155,15 +157,13 @@ func TestDeleteInstance_RunningInstanceFails(t *testing.T) {
 		},
 	}
 
-	_, err := mgr.CreateInstance("test-instance", options)
+	inst, err := mgr.CreateInstance("test-instance", options)
 	if err != nil {
 		t.Fatalf("CreateInstance failed: %v", err)
 	}
 
-	_, err = mgr.StartInstance("test-instance")
-	if err != nil {
-		t.Fatalf("StartInstance failed: %v", err)
-	}
+	// Simulate starting the instance
+	inst.SetStatus(instance.Running)
 
 	// Should fail to delete running instance
 	err = mgr.DeleteInstance("test-instance")
@@ -173,7 +173,7 @@ func TestDeleteInstance_RunningInstanceFails(t *testing.T) {
 }
 
 func TestUpdateInstance(t *testing.T) {
-	mgr := createTestManager()
+	mgr := createTestManager(t)
 	defer mgr.Shutdown()
 
 	options := &instance.Options{
@@ -186,14 +186,14 @@ func TestUpdateInstance(t *testing.T) {
 		},
 	}
 
-	_, err := mgr.CreateInstance("test-instance", options)
+	inst, err := mgr.CreateInstance("test-instance", options)
 	if err != nil {
 		t.Fatalf("CreateInstance failed: %v", err)
 	}
 
-	_, err = mgr.StartInstance("test-instance")
-	if err != nil {
-		t.Fatalf("StartInstance failed: %v", err)
+	// Start the instance (will use 'yes' command from test config)
+	if err := inst.Start(); err != nil {
+		t.Fatalf("Failed to start instance: %v", err)
 	}
 
 	// Update running instance with new model
@@ -212,9 +212,9 @@ func TestUpdateInstance(t *testing.T) {
 		t.Fatalf("UpdateInstance failed: %v", err)
 	}
 
-	// Should still be running after update
+	// Should be running after update (was running before, should be restarted)
 	if !updated.IsRunning() {
-		t.Error("Instance should be running after update")
+		t.Errorf("Instance should be running after update, got: %v", updated.GetStatus())
 	}
 
 	if updated.GetOptions().BackendOptions.LlamaServerOptions.Model != "/path/to/new-model.gguf" {
@@ -223,7 +223,7 @@ func TestUpdateInstance(t *testing.T) {
 }
 
 func TestUpdateInstance_ReleasesOldPort(t *testing.T) {
-	mgr := createTestManager()
+	mgr := createTestManager(t)
 	defer mgr.Shutdown()
 
 	options := &instance.Options{

pkg/manager/persistence.go

@@ -15,35 +15,18 @@ import (
 type instancePersister struct {
 	mu           sync.Mutex
 	instancesDir string
-	enabled      bool
 }
 
 // newInstancePersister creates a new instance persister.
 // If instancesDir is empty, persistence is disabled.
-func newInstancePersister(instancesDir string) (*instancePersister, error) {
-	if instancesDir == "" {
-		return &instancePersister{
-			enabled: false,
-		}, nil
-	}
-
-	// Ensure the instances directory exists
-	if err := os.MkdirAll(instancesDir, 0755); err != nil {
-		return nil, fmt.Errorf("failed to create instances directory: %w", err)
-	}
-
+func newInstancePersister(instancesDir string) *instancePersister {
 	return &instancePersister{
 		instancesDir: instancesDir,
-		enabled:      true,
-	}, nil
+	}
 }
 
 // Save persists an instance to disk with atomic write
 func (p *instancePersister) save(inst *instance.Instance) error {
-	if !p.enabled {
-		return nil
-	}
-
 	if inst == nil {
 		return fmt.Errorf("cannot save nil instance")
 	}
@@ -103,10 +86,6 @@ func (p *instancePersister) save(inst *instance.Instance) error {
 
 // Delete removes an instance's persistence file from disk.
 func (p *instancePersister) delete(name string) error {
-	if !p.enabled {
-		return nil
-	}
-
 	validatedName, err := p.validateInstanceName(name)
 	if err != nil {
 		return err
@@ -131,10 +110,6 @@ func (p *instancePersister) delete(name string) error {
 // LoadAll loads all persisted instances from disk.
 // Returns a slice of instances and any errors encountered during loading.
 func (p *instancePersister) loadAll() ([]*instance.Instance, error) {
-	if !p.enabled {
-		return nil, nil
-	}
-
 	p.mu.Lock()
 	defer p.mu.Unlock()
 

pkg/manager/ports.go

@@ -24,15 +24,7 @@ type portAllocator struct {
 }
 
 // newPortAllocator creates a new port allocator for the given port range.
-// Returns an error if the port range is invalid.
-func newPortAllocator(minPort, maxPort int) (*portAllocator, error) {
-	if minPort <= 0 || maxPort <= 0 {
-		return nil, fmt.Errorf("invalid port range: min=%d, max=%d (must be > 0)", minPort, maxPort)
-	}
-	if minPort > maxPort {
-		return nil, fmt.Errorf("invalid port range: min=%d > max=%d", minPort, maxPort)
-	}
-
+func newPortAllocator(minPort, maxPort int) *portAllocator {
 	rangeSize := maxPort - minPort + 1
 	bitmapSize := (rangeSize + 63) / 64 // Round up to nearest uint64
 
@@ -42,7 +34,7 @@ func newPortAllocator(minPort, maxPort int) (*portAllocator, error) {
 		minPort:   minPort,
 		maxPort:   maxPort,
 		rangeSize: rangeSize,
-	}, nil
+	}
 }
 
 // allocate finds and allocates the first available port for the given instance.

pkg/server/handlers_backends.go

@@ -66,17 +66,16 @@ func (h *Handler) LlamaCppUIProxy() http.HandlerFunc {
 			return
 		}
 
-		proxy, err := inst.GetProxy()
-		if err != nil {
-			writeError(w, http.StatusInternalServerError, "failed to get proxy", err.Error())
-			return
-		}
-
 		if !inst.IsRemote() {
 			h.stripLlamaCppPrefix(r, inst.Name)
 		}
 
-		proxy.ServeHTTP(w, r)
+		// Use instance's ServeHTTP which tracks inflight requests and handles shutting down state
+		err = inst.ServeHTTP(w, r)
+		if err != nil {
+			// Error is already handled in ServeHTTP (response written)
+			return
+		}
 	}
 }
 
@@ -110,6 +109,12 @@ func (h *Handler) LlamaCppProxy() http.HandlerFunc {
 			return
 		}
 
+		// Check if instance is shutting down before autostart logic
+		if inst.GetStatus() == instance.ShuttingDown {
+			writeError(w, http.StatusServiceUnavailable, "instance_shutting_down", "Instance is shutting down")
+			return
+		}
+
 		if !inst.IsRemote() && !inst.IsRunning() {
 			err := h.ensureInstanceRunning(inst)
 			if err != nil {
@@ -118,17 +123,16 @@ func (h *Handler) LlamaCppProxy() http.HandlerFunc {
 			}
 		}
 
-		proxy, err := inst.GetProxy()
-		if err != nil {
-			writeError(w, http.StatusInternalServerError, "failed to get proxy", err.Error())
-			return
-		}
-
 		if !inst.IsRemote() {
 			h.stripLlamaCppPrefix(r, inst.Name)
 		}
 
-		proxy.ServeHTTP(w, r)
+		// Use instance's ServeHTTP which tracks inflight requests and handles shutting down state
+		err = inst.ServeHTTP(w, r)
+		if err != nil {
+			// Error is already handled in ServeHTTP (response written)
+			return
+		}
 	}
 }
 

pkg/server/handlers_instances.go

@@ -332,12 +332,6 @@ func (h *Handler) InstanceProxy() http.HandlerFunc {
 			return
 		}
 
-		proxy, err := inst.GetProxy()
-		if err != nil {
-			writeError(w, http.StatusInternalServerError, "proxy_failed", "Failed to get proxy: "+err.Error())
-			return
-		}
-
 		if !inst.IsRemote() {
 			// Strip the "/api/v1/instances/<name>/proxy" prefix from the request URL
 			prefix := fmt.Sprintf("/api/v1/instances/%s/proxy", inst.Name)
@@ -348,6 +342,11 @@ func (h *Handler) InstanceProxy() http.HandlerFunc {
 		r.Header.Set("X-Forwarded-Host", r.Header.Get("Host"))
 		r.Header.Set("X-Forwarded-Proto", "http")
 
-		proxy.ServeHTTP(w, r)
+		// Use instance's ServeHTTP which tracks inflight requests and handles shutting down state
+		err = inst.ServeHTTP(w, r)
+		if err != nil {
+			// Error is already handled in ServeHTTP (response written)
+			return
+		}
 	}
 }

pkg/server/handlers_openai.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"encoding/json"
 	"io"
+	"llamactl/pkg/instance"
 	"llamactl/pkg/validation"
 	"net/http"
 )
@@ -106,6 +107,12 @@ func (h *Handler) OpenAIProxy() http.HandlerFunc {
 			return
 		}
 
+		// Check if instance is shutting down before autostart logic
+		if inst.GetStatus() == instance.ShuttingDown {
+			writeError(w, http.StatusServiceUnavailable, "instance_shutting_down", "Instance is shutting down")
+			return
+		}
+
 		if !inst.IsRemote() && !inst.IsRunning() {
 			err := h.ensureInstanceRunning(inst)
 			if err != nil {
@@ -114,16 +121,15 @@ func (h *Handler) OpenAIProxy() http.HandlerFunc {
 			}
 		}
 
-		proxy, err := inst.GetProxy()
-		if err != nil {
-			writeError(w, http.StatusInternalServerError, "proxy_failed", err.Error())
-			return
-		}
-
 		// Recreate the request body from the bytes we read
 		r.Body = io.NopCloser(bytes.NewReader(bodyBytes))
 		r.ContentLength = int64(len(bodyBytes))
 
-		proxy.ServeHTTP(w, r)
+		// Use instance's ServeHTTP which tracks inflight requests and handles shutting down state
+		err = inst.ServeHTTP(w, r)
+		if err != nil {
+			// Error is already handled in ServeHTTP (response written)
+			return
+		}
 	}
 }

pkg/server/handlers_system.go

@@ -20,3 +20,23 @@ func (h *Handler) VersionHandler() http.HandlerFunc {
 		writeText(w, http.StatusOK, versionInfo)
 	}
 }
+
+// ConfigHandler godoc
+// @Summary Get server configuration
+// @Description Returns the current server configuration (sanitized)
+// @Tags System
+// @Security ApiKeyAuth
+// @Produces application/json
+// @Success 200 {object} config.AppConfig "Sanitized configuration"
+// @Failure 500 {string} string "Internal Server Error"
+// @Router /api/v1/config [get]
+func (h *Handler) ConfigHandler() http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		sanitizedConfig, err := h.cfg.SanitizedCopy()
+		if err != nil {
+			writeError(w, http.StatusInternalServerError, "sanitized_copy_error", "Failed to get sanitized config")
+			return
+		}
+		writeJSON(w, http.StatusOK, sanitizedConfig)
+	}
+}

pkg/server/routes.go

@@ -1,7 +1,7 @@
 package server
 
 import (
-	"fmt"
+	"log"
 
 	"github.com/go-chi/chi/v5"
 	"github.com/go-chi/chi/v5/middleware"
@@ -42,7 +42,9 @@ func SetupRouter(handler *Handler) *chi.Mux {
 			r.Use(authMiddleware.AuthMiddleware(KeyTypeManagement))
 		}
 
-		r.Get("/version", handler.VersionHandler()) // Get server version
+		r.Get("/version", handler.VersionHandler())
+
+		r.Get("/config", handler.ConfigHandler())
 
 		// Backend-specific endpoints
 		r.Route("/backends", func(r chi.Router) {
@@ -159,7 +161,7 @@ func SetupRouter(handler *Handler) *chi.Mux {
 
 	// Serve WebUI files
 	if err := webui.SetupWebUI(r); err != nil {
-		fmt.Printf("Failed to set up WebUI: %v\n", err)
+		log.Printf("Failed to set up WebUI: %v\n", err)
 	}
 
 	return r

pkg/validation/validation_test.go

@@ -239,25 +239,3 @@ func TestValidateInstanceOptions_MultipleFieldInjection(t *testing.T) {
 		})
 	}
 }
-
-func TestValidateInstanceOptions_NonStringFields(t *testing.T) {
-	// Test that non-string fields don't interfere with validation
-	options := backends.Options{
-		BackendType: backends.BackendTypeLlamaCpp,
-		LlamaServerOptions: &backends.LlamaServerOptions{
-			Port:        8080,
-			GPULayers:   32,
-			CtxSize:     4096,
-			Temperature: 0.7,
-			TopK:        40,
-			TopP:        0.9,
-			Verbose:     true,
-			FlashAttn:   false,
-		},
-	}
-
-	err := options.ValidateInstanceOptions()
-	if err != nil {
-		t.Errorf("ValidateInstanceOptions with non-string fields should not error, got: %v", err)
-	}
-}

webui/package.json

@@ -25,34 +25,34 @@
     "@tailwindcss/vite": "^4.1.11",
     "class-variance-authority": "^0.7.1",
     "clsx": "^2.1.1",
-    "lucide-react": "^0.525.0",
-    "react": "^19.1.0",
-    "react-dom": "^19.1.0",
+    "lucide-react": "^0.554.0",
+    "react": "^19.2.0",
+    "react-dom": "^19.2.0",
     "sonner": "^2.0.7",
-    "tailwind-merge": "^3.3.1",
+    "tailwind-merge": "^3.4.0",
     "tailwindcss": "^4.1.11",
-    "zod": "^4.0.5"
+    "zod": "^4.1.12"
   },
   "devDependencies": {
-    "@eslint/js": "^9.32.0",
-    "@testing-library/jest-dom": "^6.6.3",
+    "@eslint/js": "^9.39.1",
+    "@testing-library/jest-dom": "^6.9.1",
     "@testing-library/react": "^16.3.0",
     "@testing-library/user-event": "^14.6.1",
-    "@types/eslint__js": "^8.42.3",
-    "@types/node": "^24.0.15",
-    "@types/react": "^19.1.8",
-    "@types/react-dom": "^19.1.6",
-    "@vitejs/plugin-react": "^4.7.0",
-    "@vitest/ui": "^3.2.4",
-    "eslint": "^9.32.0",
+    "@types/eslint__js": "^9.14.0",
+    "@types/node": "^24.10.1",
+    "@types/react": "^19.2.4",
+    "@types/react-dom": "^19.2.3",
+    "@vitejs/plugin-react": "^5.1.1",
+    "@vitest/ui": "^4.0.8",
+    "eslint": "^9.39.1",
     "eslint-plugin-react": "^7.37.5",
-    "eslint-plugin-react-hooks": "^5.2.0",
+    "eslint-plugin-react-hooks": "^7.0.1",
     "eslint-plugin-react-refresh": "^0.4.20",
-    "jsdom": "^26.1.0",
-    "tw-animate-css": "^1.3.5",
-    "typescript": "^5.8.3",
-    "typescript-eslint": "^8.38.0",
-    "vite": "^7.1.11",
-    "vitest": "^3.2.4"
+    "jsdom": "^27.2.0",
+    "tw-animate-css": "^1.4.0",
+    "typescript": "^5.9.3",
+    "typescript-eslint": "^8.47.0",
+    "vite": "^7.2.2",
+    "vitest": "^4.0.8"
   }
 }

webui/src/__tests__/App.test.tsx

@@ -4,8 +4,7 @@ import userEvent from '@testing-library/user-event'
 import App from '@/App'
 import { InstancesProvider } from '@/contexts/InstancesContext'
 import { instancesApi } from '@/lib/api'
-import type { Instance } from '@/types/instance'
-import { BackendType } from '@/types/instance'
+import {BackendType, type Instance } from '@/types/instance'
 import { AuthProvider } from '@/contexts/AuthContext'
 
 // Mock the API
@@ -49,6 +48,21 @@ vi.mock('@/lib/healthService', () => ({
   })),
 }))
 
+// Mock the ConfigContext helper hooks
+vi.mock('@/hooks/useConfig', () => ({
+  useInstanceDefaults: () => ({
+    autoRestart: true,
+    maxRestarts: 3,
+    restartDelay: 5,
+    onDemandStart: false,
+  }),
+  useBackendSettings: () => ({
+    command: '/usr/bin/llama-server',
+    dockerEnabled: false,
+    dockerImage: '',
+  }),
+}))
+
 function renderApp() {
   return render(
     <AuthProvider>
@@ -119,8 +133,12 @@ describe('App Component - Critical Business Logic Only', () => {
       // Verify correct API call
       await waitFor(() => {
         expect(instancesApi.create).toHaveBeenCalledWith('new-test-instance', {
-          auto_restart: true, // Default value
-          backend_type: BackendType.LLAMA_CPP
+          auto_restart: true, // Default value from config
+          backend_type: BackendType.LLAMA_CPP,
+          docker_enabled: false,
+          max_restarts: 3,
+          on_demand_start: false,
+          restart_delay: 5
         })
       })
 

webui/src/components/BackendFormField.tsx

@@ -3,22 +3,36 @@ import { Input } from '@/components/ui/input'
 import { Label } from '@/components/ui/label'
 import { Checkbox } from '@/components/ui/checkbox'
 import { getBackendFieldType, basicBackendFieldsConfig } from '@/lib/zodFormUtils'
+import ExtraArgsInput from '@/components/form/ExtraArgsInput'
 
 interface BackendFormFieldProps {
   fieldKey: string
-  value: string | number | boolean | string[] | undefined
-  onChange: (key: string, value: string | number | boolean | string[] | undefined) => void
+  value: string | number | boolean | string[] | Record<string, string> | undefined
+  onChange: (key: string, value: string | number | boolean | string[] | Record<string, string> | undefined) => void
 }
 
 const BackendFormField: React.FC<BackendFormFieldProps> = ({ fieldKey, value, onChange }) => {
+  // Special handling for extra_args
+  if (fieldKey === 'extra_args') {
+    return (
+      <ExtraArgsInput
+        id={fieldKey}
+        label="Extra Arguments"
+        value={value as Record<string, string> | undefined}
+        onChange={(newValue) => onChange(fieldKey, newValue)}
+        description="Additional command line arguments to pass to the backend"
+      />
+    )
+  }
+
   // Get configuration for basic fields, or use field name for advanced fields
-  const config = basicBackendFieldsConfig[fieldKey as string] || { label: fieldKey }
-  
+  const config = basicBackendFieldsConfig[fieldKey] || { label: fieldKey }
+
   // Get type from Zod schema
   const fieldType = getBackendFieldType(fieldKey)
 
   const handleChange = (newValue: string | number | boolean | string[] | undefined) => {
-    onChange(fieldKey as string, newValue)
+    onChange(fieldKey, newValue)
   }
 
   const renderField = () => {

webui/src/components/HealthBadge.tsx

@@ -21,6 +21,8 @@ const HealthBadge: React.FC<HealthBadgeProps> = ({ health }) => {
         return <Loader2 className="h-3 w-3 animate-spin" />;
       case "restarting":
         return <Loader2 className="h-3 w-3 animate-spin" />;
+      case "shutting_down":
+        return <Loader2 className="h-3 w-3 animate-spin" />;
       case "stopped":
         return <Clock className="h-3 w-3" />;
       case "failed":
@@ -36,6 +38,8 @@ const HealthBadge: React.FC<HealthBadgeProps> = ({ health }) => {
         return "outline";
       case "restarting":
         return "outline";
+      case "shutting_down":
+        return "outline";
       case "stopped":
         return "secondary";
       case "failed":
@@ -51,6 +55,8 @@ const HealthBadge: React.FC<HealthBadgeProps> = ({ health }) => {
         return "Starting";
       case "restarting":
         return "Restarting";
+      case "shutting_down":
+        return "Shutting Down";
       case "stopped":
         return "Stopped";
       case "failed":

webui/src/components/InstanceCard.tsx

@@ -2,12 +2,13 @@
 import { Button } from "@/components/ui/button";
 import { Card, CardContent, CardHeader, CardTitle } from "@/components/ui/card";
 import type { Instance } from "@/types/instance";
-import { Edit, FileText, Play, Square, Trash2, MoreHorizontal } from "lucide-react";
+import { Edit, FileText, Play, Square, Trash2, MoreHorizontal, Download } from "lucide-react";
 import LogsDialog from "@/components/LogDialog";
 import HealthBadge from "@/components/HealthBadge";
 import BackendBadge from "@/components/BackendBadge";
 import { useState } from "react";
 import { useInstanceHealth } from "@/hooks/useInstanceHealth";
+import { instancesApi } from "@/lib/api";
 
 interface InstanceCardProps {
   instance: Instance;
@@ -52,6 +53,36 @@ function InstanceCard({
     setIsLogsOpen(true);
   };
 
+  const handleExport = () => {
+    void (async () => {
+      try {
+        // Fetch the most up-to-date instance data from the backend
+        const instanceData = await instancesApi.get(instance.name);
+
+        // Convert to JSON string with pretty formatting (matching backend format)
+        const jsonString = JSON.stringify(instanceData, null, 2);
+
+        // Create a blob and download link
+        const blob = new Blob([jsonString], { type: "application/json" });
+        const url = URL.createObjectURL(blob);
+        const link = document.createElement("a");
+        link.href = url;
+        link.download = `${instance.name}.json`;
+
+        // Trigger download
+        document.body.appendChild(link);
+        link.click();
+
+        // Cleanup
+        document.body.removeChild(link);
+        URL.revokeObjectURL(url);
+      } catch (error) {
+        console.error("Failed to export instance:", error);
+        alert(`Failed to export instance: ${error instanceof Error ? error.message : "Unknown error"}`);
+      }
+    })();
+  };
+
   const running = instance.status === "running";
 
   return (
@@ -66,7 +97,7 @@ function InstanceCard({
             
             {/* Badges row */}
             <div className="flex items-center gap-2 flex-wrap">
-              <BackendBadge backend={instance.options?.backend_type} docker={instance.docker_enabled} />
+              <BackendBadge backend={instance.options?.backend_type} docker={instance.options?.docker_enabled} />
               {running && <HealthBadge health={health} />}
             </div>
           </div>
@@ -131,6 +162,18 @@ function InstanceCard({
                 Logs
               </Button>
 
+              <Button
+                size="sm"
+                variant="outline"
+                onClick={handleExport}
+                title="Export instance"
+                data-testid="export-instance-button"
+                className="flex-1"
+              >
+                <Download className="h-4 w-4 mr-1" />
+                Export
+              </Button>
+
               <Button
                 size="sm"
                 variant="destructive"

webui/src/components/InstanceDialog.tsx

@@ -1,4 +1,4 @@
-import React, { useState, useEffect } from "react";
+import React, { useState, useEffect, useRef } from "react";
 import { Button } from "@/components/ui/button";
 import {
   Dialog,
@@ -9,9 +9,12 @@ import {
   DialogTitle,
 } from "@/components/ui/dialog";
 import { BackendType, type CreateInstanceOptions, type Instance } from "@/types/instance";
+import type { BackendOptions } from "@/schemas/instanceOptions";
 import ParseCommandDialog from "@/components/ParseCommandDialog";
 import InstanceSettingsCard from "@/components/instance/InstanceSettingsCard";
 import BackendConfigurationCard from "@/components/instance/BackendConfigurationCard";
+import { Upload } from "lucide-react";
+import { useInstanceDefaults, useBackendSettings } from "@/hooks/useConfig";
 
 interface InstanceDialogProps {
   open: boolean;
@@ -27,12 +30,18 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
   instance,
 }) => {
   const isEditing = !!instance;
+  const instanceDefaults = useInstanceDefaults();
 
   const [instanceName, setInstanceName] = useState("");
   const [formData, setFormData] = useState<CreateInstanceOptions>({});
   const [nameError, setNameError] = useState("");
   const [showParseDialog, setShowParseDialog] = useState(false);
+  const fileInputRef = useRef<HTMLInputElement>(null);
 
+  // Get backend settings for all backends (we'll use this to update docker_enabled on backend type change)
+  const llamaCppSettings = useBackendSettings(BackendType.LLAMA_CPP);
+  const vllmSettings = useBackendSettings(BackendType.VLLM);
+  const mlxSettings = useBackendSettings(BackendType.MLX_LM);
 
   // Reset form when dialog opens/closes or when instance changes
   useEffect(() => {
@@ -42,43 +51,58 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
         setInstanceName(instance.name);
         setFormData(instance.options || {});
       } else {
-        // Reset form for new instance
+        // Reset form for new instance with defaults from config
         setInstanceName("");
         setFormData({
-          auto_restart: true, // Default value
+          auto_restart: instanceDefaults?.autoRestart ?? true,
+          max_restarts: instanceDefaults?.maxRestarts,
+          restart_delay: instanceDefaults?.restartDelay,
+          on_demand_start: instanceDefaults?.onDemandStart,
           backend_type: BackendType.LLAMA_CPP, // Default backend type
+          docker_enabled: llamaCppSettings?.dockerEnabled ?? false,
           backend_options: {},
         });
       }
       setNameError(""); // Reset any name errors
     }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
   }, [open, instance]);
 
-  const handleFieldChange = (key: keyof CreateInstanceOptions, value: any) => {
+  const handleFieldChange = (key: keyof CreateInstanceOptions, value: unknown) => {
     setFormData((prev) => {
-      // If backend_type is changing, clear backend_options
+      // If backend_type is changing, update docker_enabled default and clear backend_options
       if (key === 'backend_type' && prev.backend_type !== value) {
+        let dockerEnabled = false;
+        if (value === BackendType.LLAMA_CPP) {
+          dockerEnabled = llamaCppSettings?.dockerEnabled ?? false;
+        } else if (value === BackendType.VLLM) {
+          dockerEnabled = vllmSettings?.dockerEnabled ?? false;
+        } else if (value === BackendType.MLX_LM) {
+          dockerEnabled = mlxSettings?.dockerEnabled ?? false;
+        }
+
         return {
           ...prev,
-          [key]: value,
+          backend_type: value as CreateInstanceOptions['backend_type'],
+          docker_enabled: dockerEnabled,
           backend_options: {}, // Clear backend options when backend type changes
         };
       }
-      
+
       return {
         ...prev,
         [key]: value,
-      };
+      } as CreateInstanceOptions;
     });
   };
 
-  const handleBackendFieldChange = (key: string, value: any) => {
+  const handleBackendFieldChange = (key: string, value: unknown) => {
     setFormData((prev) => ({
       ...prev,
       backend_options: {
         ...prev.backend_options,
         [key]: value,
-      } as any,
+      } as BackendOptions,
     }));
   };
 
@@ -103,12 +127,22 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
       return;
     }
 
+    // Validate docker_enabled and command_override relationship
+    if (formData.backend_type !== BackendType.MLX_LM) {
+      if (formData.docker_enabled === true && formData.command_override) {
+        setNameError("Command override cannot be set when Docker is enabled");
+        return;
+      }
+    }
+
     // Clean up undefined values to avoid sending empty fields
-    const cleanOptions: CreateInstanceOptions = {};
+    const cleanOptions: CreateInstanceOptions = {} as CreateInstanceOptions;
     Object.entries(formData).forEach(([key, value]) => {
+      const typedKey = key as keyof CreateInstanceOptions;
+
       if (key === 'backend_options' && value && typeof value === 'object' && !Array.isArray(value)) {
         // Handle backend_options specially - clean nested object
-        const cleanBackendOptions: any = {};
+        const cleanBackendOptions: Record<string, unknown> = {};
         Object.entries(value).forEach(([backendKey, backendValue]) => {
           if (backendValue !== undefined && backendValue !== null && (typeof backendValue !== 'string' || backendValue.trim() !== "")) {
             // Handle arrays - don't include empty arrays
@@ -121,7 +155,7 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
 
         // Only include backend_options if it has content
         if (Object.keys(cleanBackendOptions).length > 0) {
-          (cleanOptions as any)[key] = cleanBackendOptions;
+          (cleanOptions as Record<string, unknown>)[typedKey] = cleanBackendOptions as BackendOptions;
         }
       } else if (value !== undefined && value !== null) {
         // Skip empty strings
@@ -132,7 +166,7 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
         if (Array.isArray(value) && value.length === 0) {
           return;
         }
-        (cleanOptions as any)[key] = value;
+        (cleanOptions as Record<string, unknown>)[typedKey] = value;
       }
     });
 
@@ -153,6 +187,49 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
     setShowParseDialog(false);
   };
 
+  const handleImportFile = () => {
+    fileInputRef.current?.click();
+  };
+
+  const handleFileChange = (event: React.ChangeEvent<HTMLInputElement>) => {
+    const file = event.target.files?.[0];
+    if (!file) return;
+
+    const reader = new FileReader();
+    reader.onload = (e) => {
+      try {
+        const content = e.target?.result as string;
+        const importedData = JSON.parse(content) as { name?: string; options?: CreateInstanceOptions };
+
+        // Validate that it's an instance export
+        if (!importedData.name || !importedData.options) {
+          alert('Invalid instance file: Missing required fields (name, options)');
+          return;
+        }
+
+        // Set the instance name (only for new instances, not editing)
+        if (!isEditing && typeof importedData.name === 'string') {
+          handleNameChange(importedData.name);
+        }
+
+        // Populate all the options from the imported file
+        if (importedData.options) {
+          setFormData(prev => ({
+            ...prev,
+            ...importedData.options,
+          }));
+        }
+
+        // Reset the file input
+        event.target.value = '';
+      } catch (error) {
+        console.error('Failed to parse instance file:', error);
+        alert(`Failed to parse instance file: ${error instanceof Error ? error.message : 'Invalid JSON'}`);
+      }
+    };
+
+    reader.readAsText(file);
+  };
 
   // Save button label logic
   let saveButtonLabel = "Create Instance";
@@ -168,14 +245,38 @@ const InstanceDialog: React.FC<InstanceDialogProps> = ({
     <Dialog open={open} onOpenChange={onOpenChange}>
       <DialogContent className="sm:max-w-[600px] max-h-[80vh] overflow-hidden flex flex-col">
         <DialogHeader>
-          <DialogTitle>
-            {isEditing ? "Edit Instance" : "Create New Instance"}
-          </DialogTitle>
-          <DialogDescription>
-            {isEditing
-              ? "Modify the instance configuration below."
-              : "Configure your new llama-server instance below."}
-          </DialogDescription>
+          <div className="flex items-center justify-between">
+            <div className="flex-1">
+              <DialogTitle>
+                {isEditing ? "Edit Instance" : "Create New Instance"}
+              </DialogTitle>
+              <DialogDescription>
+                {isEditing
+                  ? "Modify the instance configuration below."
+                  : "Configure your new llama-server instance below."}
+              </DialogDescription>
+            </div>
+            {!isEditing && (
+              <Button
+                type="button"
+                variant="ghost"
+                size="sm"
+                onClick={handleImportFile}
+                title="Import instance configuration from JSON file"
+                className="ml-2"
+              >
+                <Upload className="h-4 w-4 mr-2" />
+                Import
+              </Button>
+            )}
+          </div>
+          <input
+            ref={fileInputRef}
+            type="file"
+            accept=".json"
+            onChange={handleFileChange}
+            className="hidden"
+          />
         </DialogHeader>
 
         <div className="flex-1 overflow-y-auto">

webui/src/components/InstanceList.tsx

@@ -56,9 +56,9 @@ function InstanceList({ editInstance }: InstanceListProps) {
           <MemoizedInstanceCard 
             key={instance.name} 
             instance={instance}
-            startInstance={startInstance}
-            stopInstance={stopInstance}
-            deleteInstance={deleteInstance}
+            startInstance={() => { void startInstance(instance.name) }}
+            stopInstance={() => { void stopInstance(instance.name) }}
+            deleteInstance={() => { void deleteInstance(instance.name) }}
             editInstance={editInstance}
           />
         ))}

webui/src/components/ParseCommandDialog.tsx

@@ -54,7 +54,7 @@ const ParseCommandDialog: React.FC<ParseCommandDialogProps> = ({
           options = await backendsApi.vllm.parseCommand(command);
           break;
         default:
-          throw new Error(`Unsupported backend type: ${backendType}`);
+          throw new Error(`Unsupported backend type: ${String(backendType)}`);
       }
 
       onParsed(options);

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

@@ -2,8 +2,22 @@ import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest'
 import { render, screen, waitFor } from '@testing-library/react'
 import userEvent from '@testing-library/user-event'
 import InstanceDialog from '@/components/InstanceDialog'
-import type { Instance } from '@/types/instance'
-import { BackendType } from '@/types/instance'
+import { BackendType, type Instance } from '@/types/instance'
+
+// Mock the ConfigContext helper hooks
+vi.mock('@/hooks/useConfig', () => ({
+  useInstanceDefaults: () => ({
+    autoRestart: true,
+    maxRestarts: 3,
+    restartDelay: 5,
+    onDemandStart: false,
+  }),
+  useBackendSettings: () => ({
+    command: '/usr/bin/llama-server',
+    dockerEnabled: false,
+    dockerImage: '',
+  }),
+}))
 
 describe('InstanceModal - Form Logic and Validation', () => {
   const mockOnSave = vi.fn()
@@ -75,7 +89,7 @@ afterEach(() => {
 
     it('submits form with correct data structure', async () => {
       const user = userEvent.setup()
-      
+
       render(
         <InstanceDialog
           open={true}
@@ -86,13 +100,17 @@ afterEach(() => {
 
       // Fill required name
       await user.type(screen.getByLabelText(/Instance Name/), 'my-instance')
-      
+
       // Submit form
       await user.click(screen.getByTestId('dialog-save-button'))
 
       expect(mockOnSave).toHaveBeenCalledWith('my-instance', {
-        auto_restart: true, // Default value
-        backend_type: BackendType.LLAMA_CPP
+        auto_restart: true, // Default value from config
+        backend_type: BackendType.LLAMA_CPP,
+        docker_enabled: false,
+        max_restarts: 3,
+        on_demand_start: false,
+        restart_delay: 5
       })
     })
 
@@ -253,7 +271,7 @@ afterEach(() => {
 
     it('includes restart options in form submission when enabled', async () => {
       const user = userEvent.setup()
-      
+
       render(
         <InstanceDialog
           open={true}
@@ -264,17 +282,23 @@ afterEach(() => {
 
       // Fill form
       await user.type(screen.getByLabelText(/Instance Name/), 'test-instance')
-      
-      // Set restart options
-      await user.type(screen.getByLabelText(/Max Restarts/), '5')
-      await user.type(screen.getByLabelText(/Restart Delay/), '10')
+
+      // Clear default values and set new restart options
+      const maxRestartsInput = screen.getByLabelText(/Max Restarts/)
+      const restartDelayInput = screen.getByLabelText(/Restart Delay/)
+      await user.clear(maxRestartsInput)
+      await user.type(maxRestartsInput, '5')
+      await user.clear(restartDelayInput)
+      await user.type(restartDelayInput, '10')
 
       await user.click(screen.getByTestId('dialog-save-button'))
 
       expect(mockOnSave).toHaveBeenCalledWith('test-instance', {
         auto_restart: true,
         backend_type: BackendType.LLAMA_CPP,
+        docker_enabled: false,
         max_restarts: 5,
+        on_demand_start: false,
         restart_delay: 10
       })
     })
@@ -284,7 +308,7 @@ afterEach(() => {
   describe('Form Data Handling', () => {
     it('cleans up undefined values before submission', async () => {
       const user = userEvent.setup()
-      
+
       render(
         <InstanceDialog
           open={true}
@@ -298,16 +322,20 @@ afterEach(() => {
 
       await user.click(screen.getByTestId('dialog-save-button'))
 
-      // Should only include non-empty values
+      // Should include default values from config
       expect(mockOnSave).toHaveBeenCalledWith('clean-instance', {
-        auto_restart: true, // Only this default value should be included
-        backend_type: BackendType.LLAMA_CPP
+        auto_restart: true,
+        backend_type: BackendType.LLAMA_CPP,
+        docker_enabled: false,
+        max_restarts: 3,
+        on_demand_start: false,
+        restart_delay: 5
       })
     })
 
     it('handles numeric fields correctly', async () => {
       const user = userEvent.setup()
-      
+
       render(
         <InstanceDialog
           open={true}
@@ -317,7 +345,7 @@ afterEach(() => {
       )
 
       await user.type(screen.getByLabelText(/Instance Name/), 'numeric-test')
-      
+
       // Test GPU layers field (numeric)
       const gpuLayersInput = screen.getByLabelText(/GPU Layers/)
       await user.type(gpuLayersInput, '15')
@@ -328,6 +356,10 @@ afterEach(() => {
         auto_restart: true,
         backend_type: BackendType.LLAMA_CPP,
         backend_options: { gpu_layers: 15 }, // Should be number, not string
+        docker_enabled: false,
+        max_restarts: 3,
+        on_demand_start: false,
+        restart_delay: 5
       })
     })
   })

webui/src/components/form/EnvVarsInput.tsx

@@ -0,0 +1,26 @@
+import React from 'react'
+import KeyValueInput from './KeyValueInput'
+
+interface EnvVarsInputProps {
+  id: string
+  label: string
+  value: Record<string, string> | undefined
+  onChange: (value: Record<string, string> | undefined) => void
+  description?: string
+  disabled?: boolean
+  className?: string
+}
+
+const EnvVarsInput: React.FC<EnvVarsInputProps> = (props) => {
+  return (
+    <KeyValueInput
+      {...props}
+      keyPlaceholder="Variable name"
+      valuePlaceholder="Variable value"
+      addButtonText="Add Variable"
+      allowEmptyValues={false}
+    />
+  )
+}
+
+export default EnvVarsInput

webui/src/components/form/EnvironmentVariablesInput.tsx

@@ -1,144 +0,0 @@
-import React, { useState } from 'react'
-import { Input } from '@/components/ui/input'
-import { Label } from '@/components/ui/label'
-import { Button } from '@/components/ui/button'
-import { X, Plus } from 'lucide-react'
-
-interface EnvironmentVariablesInputProps {
-  id: string
-  label: string
-  value: Record<string, string> | undefined
-  onChange: (value: Record<string, string> | undefined) => void
-  description?: string
-  disabled?: boolean
-  className?: string
-}
-
-interface EnvVar {
-  key: string
-  value: string
-}
-
-const EnvironmentVariablesInput: React.FC<EnvironmentVariablesInputProps> = ({
-  id,
-  label,
-  value,
-  onChange,
-  description,
-  disabled = false,
-  className
-}) => {
-  // Convert the value object to an array of key-value pairs for editing
-  const envVarsFromValue = value
-    ? Object.entries(value).map(([key, val]) => ({ key, value: val }))
-    : []
-
-  const [envVars, setEnvVars] = useState<EnvVar[]>(
-    envVarsFromValue.length > 0 ? envVarsFromValue : [{ key: '', value: '' }]
-  )
-
-  // Update parent component when env vars change
-  const updateParent = (newEnvVars: EnvVar[]) => {
-    // Filter out empty entries
-    const validVars = newEnvVars.filter(env => env.key.trim() !== '' && env.value.trim() !== '')
-
-    if (validVars.length === 0) {
-      onChange(undefined)
-    } else {
-      const envObject = validVars.reduce((acc, env) => {
-        acc[env.key.trim()] = env.value.trim()
-        return acc
-      }, {} as Record<string, string>)
-      onChange(envObject)
-    }
-  }
-
-  const handleKeyChange = (index: number, newKey: string) => {
-    const newEnvVars = [...envVars]
-    newEnvVars[index].key = newKey
-    setEnvVars(newEnvVars)
-    updateParent(newEnvVars)
-  }
-
-  const handleValueChange = (index: number, newValue: string) => {
-    const newEnvVars = [...envVars]
-    newEnvVars[index].value = newValue
-    setEnvVars(newEnvVars)
-    updateParent(newEnvVars)
-  }
-
-  const addEnvVar = () => {
-    const newEnvVars = [...envVars, { key: '', value: '' }]
-    setEnvVars(newEnvVars)
-  }
-
-  const removeEnvVar = (index: number) => {
-    if (envVars.length === 1) {
-      // Reset to empty if it's the last one
-      const newEnvVars = [{ key: '', value: '' }]
-      setEnvVars(newEnvVars)
-      updateParent(newEnvVars)
-    } else {
-      const newEnvVars = envVars.filter((_, i) => i !== index)
-      setEnvVars(newEnvVars)
-      updateParent(newEnvVars)
-    }
-  }
-
-  return (
-    <div className={`grid gap-2 ${className || ''}`}>
-      <Label htmlFor={id}>
-        {label}
-      </Label>
-      <div className="space-y-2">
-        {envVars.map((envVar, index) => (
-          <div key={index} className="flex gap-2 items-center">
-            <Input
-              placeholder="Variable name"
-              value={envVar.key}
-              onChange={(e) => handleKeyChange(index, e.target.value)}
-              disabled={disabled}
-              className="flex-1"
-            />
-            <Input
-              placeholder="Variable value"
-              value={envVar.value}
-              onChange={(e) => handleValueChange(index, e.target.value)}
-              disabled={disabled}
-              className="flex-1"
-            />
-            <Button
-              type="button"
-              variant="outline"
-              size="sm"
-              onClick={() => removeEnvVar(index)}
-              disabled={disabled}
-              className="shrink-0"
-            >
-              <X className="h-4 w-4" />
-            </Button>
-          </div>
-        ))}
-        <Button
-          type="button"
-          variant="outline"
-          size="sm"
-          onClick={addEnvVar}
-          disabled={disabled}
-          className="w-fit"
-        >
-          <Plus className="h-4 w-4 mr-2" />
-          Add Variable
-        </Button>
-      </div>
-      {description && (
-        <p className="text-sm text-muted-foreground">{description}</p>
-      )}
-      <p className="text-xs text-muted-foreground">
-        Environment variables that will be passed to the backend process
-      </p>
-    </div>
-  )
-}
-
-export default EnvironmentVariablesInput

webui/src/components/form/ExtraArgsInput.tsx

@@ -0,0 +1,27 @@
+import React from 'react'
+import KeyValueInput from './KeyValueInput'
+
+interface ExtraArgsInputProps {
+  id: string
+  label: string
+  value: Record<string, string> | undefined
+  onChange: (value: Record<string, string> | undefined) => void
+  description?: string
+  disabled?: boolean
+  className?: string
+}
+
+const ExtraArgsInput: React.FC<ExtraArgsInputProps> = (props) => {
+  return (
+    <KeyValueInput
+      {...props}
+      keyPlaceholder="Flag name (without --)"
+      valuePlaceholder="Value (empty for boolean flags)"
+      addButtonText="Add Argument"
+      helperText="Additional command line arguments to pass to the backend. Leave value empty for boolean flags."
+      allowEmptyValues={true}
+    />
+  )
+}
+
+export default ExtraArgsInput

webui/src/components/form/KeyValueInput.tsx

@@ -0,0 +1,171 @@
+import React, { useState, useEffect } from 'react'
+import { Input } from '@/components/ui/input'
+import { Label } from '@/components/ui/label'
+import { Button } from '@/components/ui/button'
+import { X, Plus } from 'lucide-react'
+
+interface KeyValueInputProps {
+  id: string
+  label: string
+  value: Record<string, string> | undefined
+  onChange: (value: Record<string, string> | undefined) => void
+  description?: string
+  disabled?: boolean
+  className?: string
+  keyPlaceholder?: string
+  valuePlaceholder?: string
+  addButtonText?: string
+  helperText?: string
+  allowEmptyValues?: boolean // If true, entries with empty values are considered valid
+}
+
+interface KeyValuePair {
+  key: string
+  value: string
+}
+
+const KeyValueInput: React.FC<KeyValueInputProps> = ({
+  id,
+  label,
+  value,
+  onChange,
+  description,
+  disabled = false,
+  className,
+  keyPlaceholder = 'Key',
+  valuePlaceholder = 'Value',
+  addButtonText = 'Add Entry',
+  helperText,
+  allowEmptyValues = false
+}) => {
+  // Convert the value object to an array of key-value pairs for editing
+  const pairsFromValue = value
+    ? Object.entries(value).map(([key, val]) => ({ key, value: val }))
+    : []
+
+  const [pairs, setPairs] = useState<KeyValuePair[]>(
+    pairsFromValue.length > 0 ? pairsFromValue : [{ key: '', value: '' }]
+  )
+
+  // Sync internal state when value prop changes
+  useEffect(() => {
+    const newPairsFromValue = value
+      ? Object.entries(value).map(([key, val]) => ({ key, value: val }))
+      : []
+
+    if (newPairsFromValue.length > 0) {
+      setPairs(newPairsFromValue)
+    } else if (!value) {
+      // Reset to single empty row if value is explicitly undefined/null
+      setPairs([{ key: '', value: '' }])
+    }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [value])
+
+  // Update parent component when pairs change
+  const updateParent = (newPairs: KeyValuePair[]) => {
+    // Filter based on validation rules
+    const validPairs = allowEmptyValues
+      ? newPairs.filter(pair => pair.key.trim() !== '')
+      : newPairs.filter(pair => pair.key.trim() !== '' && pair.value.trim() !== '')
+
+    if (validPairs.length === 0) {
+      onChange(undefined)
+    } else {
+      const pairsObject = validPairs.reduce((acc, pair) => {
+        acc[pair.key.trim()] = pair.value.trim()
+        return acc
+      }, {} as Record<string, string>)
+      onChange(pairsObject)
+    }
+  }
+
+  const handleKeyChange = (index: number, newKey: string) => {
+    const newPairs = [...pairs]
+    newPairs[index].key = newKey
+    setPairs(newPairs)
+    updateParent(newPairs)
+  }
+
+  const handleValueChange = (index: number, newValue: string) => {
+    const newPairs = [...pairs]
+    newPairs[index].value = newValue
+    setPairs(newPairs)
+    updateParent(newPairs)
+  }
+
+  const addPair = () => {
+    const newPairs = [...pairs, { key: '', value: '' }]
+    setPairs(newPairs)
+  }
+
+  const removePair = (index: number) => {
+    if (pairs.length === 1) {
+      // Reset to empty if it's the last one
+      const newPairs = [{ key: '', value: '' }]
+      setPairs(newPairs)
+      updateParent(newPairs)
+    } else {
+      const newPairs = pairs.filter((_, i) => i !== index)
+      setPairs(newPairs)
+      updateParent(newPairs)
+    }
+  }
+
+  return (
+    <div className={`grid gap-2 ${className || ''}`}>
+      <Label htmlFor={id}>
+        {label}
+      </Label>
+      <div className="space-y-2">
+        {pairs.map((pair, index) => (
+          <div key={index} className="flex gap-2 items-center">
+            <Input
+              placeholder={keyPlaceholder}
+              value={pair.key}
+              onChange={(e) => handleKeyChange(index, e.target.value)}
+              disabled={disabled}
+              className="flex-1"
+            />
+            <Input
+              placeholder={valuePlaceholder}
+              value={pair.value}
+              onChange={(e) => handleValueChange(index, e.target.value)}
+              disabled={disabled}
+              className="flex-1"
+            />
+            <Button
+              type="button"
+              variant="outline"
+              size="sm"
+              onClick={() => removePair(index)}
+              disabled={disabled}
+              className="shrink-0"
+            >
+              <X className="h-4 w-4" />
+            </Button>
+          </div>
+        ))}
+        <Button
+          type="button"
+          variant="outline"
+          size="sm"
+          onClick={addPair}
+          disabled={disabled}
+          className="w-fit"
+        >
+          <Plus className="h-4 w-4 mr-2" />
+          {addButtonText}
+        </Button>
+      </div>
+      {description && (
+        <p className="text-sm text-muted-foreground">{description}</p>
+      )}
+      {helperText && (
+        <p className="text-xs text-muted-foreground">{helperText}</p>
+      )}
+    </div>
+  )
+}
+
+export default KeyValueInput

webui/src/components/instance/BackendConfiguration.tsx

@@ -47,8 +47,18 @@ const BackendConfiguration: React.FC<BackendConfigurationProps> = ({
             ))}
         </div>
       )}
+
+      {/* Extra Args - Always visible as a separate section */}
+      <div className="space-y-4">
+        <BackendFormField
+          key="extra_args"
+          fieldKey="extra_args"
+          value={(formData.backend_options as any)?.extra_args}
+          onChange={onBackendFieldChange}
+        />
+      </div>
     </div>
   )
 }
 
-export default BackendConfiguration
+export default BackendConfiguration

webui/src/components/instance/BackendConfigurationCard.tsx

@@ -6,6 +6,7 @@ import { Terminal, ChevronDown, ChevronRight } from 'lucide-react'
 import { getBasicBackendFields, getAdvancedBackendFields } from '@/lib/zodFormUtils'
 import BackendFormField from '@/components/BackendFormField'
 import SelectInput from '@/components/form/SelectInput'
+import ExecutionContextSection from '@/components/instance/ExecutionContextSection'
 
 interface BackendConfigurationCardProps {
   formData: CreateInstanceOptions
@@ -59,6 +60,12 @@ const BackendConfigurationCard: React.FC<BackendConfigurationCardProps> = ({
           </p>
         </div>
 
+        {/* Execution Context Section */}
+        <ExecutionContextSection
+          formData={formData}
+          onChange={onChange}
+        />
+
         {/* Basic Backend Options */}
         {basicBackendFields.length > 0 && (
           <div className="space-y-4">
@@ -109,6 +116,16 @@ const BackendConfigurationCard: React.FC<BackendConfigurationCardProps> = ({
             )}
           </div>
         )}
+
+        {/* Extra Arguments - Always visible */}
+        <div className="space-y-4">
+          <BackendFormField
+            key="extra_args"
+            fieldKey="extra_args"
+            value={(formData.backend_options as Record<string, unknown>)?.extra_args as Record<string, string> | undefined}
+            onChange={onBackendFieldChange}
+          />
+        </div>
       </CardContent>
     </Card>
   )

webui/src/components/instance/ExecutionContextSection.tsx

@@ -0,0 +1,76 @@
+import React from 'react'
+import { BackendType, type CreateInstanceOptions } from '@/types/instance'
+import CheckboxInput from '@/components/form/CheckboxInput'
+import TextInput from '@/components/form/TextInput'
+import EnvVarsInput from '@/components/form/EnvVarsInput'
+import { useBackendSettings } from '@/hooks/useConfig'
+
+interface ExecutionContextSectionProps {
+  formData: CreateInstanceOptions
+  onChange: (key: keyof CreateInstanceOptions, value: unknown) => void
+}
+
+const ExecutionContextSection: React.FC<ExecutionContextSectionProps> = ({
+  formData,
+  onChange
+}) => {
+  const backendSettings = useBackendSettings(formData.backend_type)
+
+  // Get placeholder for command override based on backend type and config
+  const getCommandPlaceholder = () => {
+    if (backendSettings?.command) {
+      return backendSettings.command
+    }
+
+    // Fallback placeholders if config is not loaded
+    switch (formData.backend_type) {
+      case BackendType.LLAMA_CPP:
+        return "llama-server"
+      case BackendType.VLLM:
+        return "vllm"
+      case BackendType.MLX_LM:
+        return "mlx_lm.server"
+      default:
+        return ""
+    }
+  }
+
+  return (
+    <div className="space-y-4">
+      <h3 className="text-md font-medium">Execution Context</h3>
+
+      {/* Docker Mode Toggle - only for backends that support Docker */}
+      {formData.backend_type !== BackendType.MLX_LM && (
+        <CheckboxInput
+          id="docker_enabled"
+          label="Enable Docker"
+          value={formData.docker_enabled}
+          onChange={(value) => onChange('docker_enabled', value)}
+          description="Run backend in Docker container"
+        />
+      )}
+
+      {/* Command Override - only shown when Docker is disabled or backend is MLX */}
+      {(formData.backend_type === BackendType.MLX_LM || formData.docker_enabled !== true) && (
+        <TextInput
+          id="command_override"
+          label="Command Override"
+          value={formData.command_override || ''}
+          onChange={(value) => onChange('command_override', value)}
+          placeholder={getCommandPlaceholder()}
+          description="Custom path to backend executable (leave empty to use config default)"
+        />
+      )}
+
+      <EnvVarsInput
+        id="environment"
+        label="Environment Variables"
+        value={formData.environment}
+        onChange={(value) => onChange('environment', value)}
+        description="Custom environment variables for the instance"
+      />
+    </div>
+  )
+}
+
+export default ExecutionContextSection

webui/src/components/instance/InstanceSettingsCard.tsx

@@ -1,12 +1,11 @@
 import React, { useState, useEffect } from 'react'
-import type { CreateInstanceOptions } from '@/types/instance'
+import { type CreateInstanceOptions } from '@/types/instance'
 import { Card, CardContent, CardHeader, CardTitle } from '@/components/ui/card'
 import { Label } from '@/components/ui/label'
 import { Input } from '@/components/ui/input'
 import AutoRestartConfiguration from '@/components/instance/AutoRestartConfiguration'
 import NumberInput from '@/components/form/NumberInput'
 import CheckboxInput from '@/components/form/CheckboxInput'
-import EnvironmentVariablesInput from '@/components/form/EnvironmentVariablesInput'
 import SelectInput from '@/components/form/SelectInput'
 import { nodesApi, type NodesMap } from '@/lib/api'
 
@@ -131,14 +130,6 @@ const InstanceSettingsCard: React.FC<InstanceSettingsCardProps> = ({
             onChange={(value) => onChange('on_demand_start', value)}
             description="Start instance only when needed"
           />
-
-          <EnvironmentVariablesInput
-            id="environment"
-            label="Environment Variables"
-            value={formData.environment}
-            onChange={(value) => onChange('environment', value)}
-            description="Custom environment variables for the instance"
-          />
         </div>
       </CardContent>
     </Card>

webui/src/contexts/AuthContext.tsx

@@ -147,16 +147,3 @@ export const useAuth = (): AuthContextType => {
   }
   return context
 }
-
-// Helper hook for getting auth headers
-export const useAuthHeaders = (): HeadersInit => {
-  const { apiKey, isAuthenticated } = useAuth()
-  
-  if (!isAuthenticated || !apiKey) {
-    return {}
-  }
-  
-  return {
-    'Authorization': `Bearer ${apiKey}`
-  }
-}

webui/src/contexts/ConfigContext.tsx

@@ -0,0 +1,62 @@
+import { type ReactNode, createContext, useContext, useEffect, useState, useRef } from 'react'
+import { serverApi } from '@/lib/api'
+import type { AppConfig } from '@/types/config'
+import { useAuth } from './AuthContext'
+
+interface ConfigContextType {
+  config: AppConfig | null
+  isLoading: boolean
+  error: string | null
+}
+
+const ConfigContext = createContext<ConfigContextType | undefined>(undefined)
+
+interface ConfigProviderProps {
+  children: ReactNode
+}
+
+export const ConfigProvider = ({ children }: ConfigProviderProps) => {
+  const { isAuthenticated } = useAuth()
+  const [config, setConfig] = useState<AppConfig | null>(null)
+  const [isLoading, setIsLoading] = useState(true)
+  const [error, setError] = useState<string | null>(null)
+  const loadedRef = useRef(false)
+
+  useEffect(() => {
+    if (!isAuthenticated || loadedRef.current) {
+      setIsLoading(false)
+      return
+    }
+
+    loadedRef.current = true
+
+    const loadConfig = async () => {
+      try {
+        const data = await serverApi.getConfig()
+        setConfig(data)
+      } catch (err) {
+        const errorMessage = err instanceof Error ? err.message : 'Failed to load configuration'
+        setError(errorMessage)
+        console.error('Error loading config:', err)
+      } finally {
+        setIsLoading(false)
+      }
+    }
+
+    void loadConfig()
+  }, [isAuthenticated])
+
+  return (
+    <ConfigContext.Provider value={{ config, isLoading, error }}>
+      {children}
+    </ConfigContext.Provider>
+  )
+}
+
+export const useConfig = (): ConfigContextType => {
+  const context = useContext(ConfigContext)
+  if (context === undefined) {
+    throw new Error('useConfig must be used within a ConfigProvider')
+  }
+  return context
+}

webui/src/hooks/useConfig.ts

@@ -0,0 +1,51 @@
+import { useConfig } from '@/contexts/ConfigContext'
+
+// Helper hook to get instance default values from config
+export const useInstanceDefaults = () => {
+  const { config } = useConfig()
+
+  if (!config || !config.instances) {
+    return null
+  }
+
+  return {
+    autoRestart: config.instances.default_auto_restart,
+    maxRestarts: config.instances.default_max_restarts,
+    restartDelay: config.instances.default_restart_delay,
+    onDemandStart: config.instances.default_on_demand_start,
+  }
+}
+
+// Helper hook to get specific backend settings by backend type
+export const useBackendSettings = (backendType: string | undefined) => {
+  const { config } = useConfig()
+
+  if (!config || !config.backends || !backendType) {
+    return null
+  }
+
+  // Map backend type to config key
+  const backendKey = backendType === 'llama_cpp'
+    ? 'llama-cpp'
+    : backendType === 'mlx_lm'
+    ? 'mlx'
+    : backendType === 'vllm'
+    ? 'vllm'
+    : null
+
+  if (!backendKey) {
+    return null
+  }
+
+  const backendConfig = config.backends[backendKey as keyof typeof config.backends]
+
+  if (!backendConfig) {
+    return null
+  }
+
+  return {
+    command: backendConfig.command || '',
+    dockerEnabled: backendConfig.docker?.enabled ?? false,
+    dockerImage: backendConfig.docker?.image || '',
+  }
+}

webui/src/hooks/useInstanceHealth.ts

@@ -18,7 +18,7 @@ export function useInstanceHealth(instanceName: string, instanceStatus: Instance
 
   // Trigger health check when instance status changes to active states
   useEffect(() => {
-    if (instanceStatus === 'running' || instanceStatus === 'restarting') {
+    if (instanceStatus === 'running' || instanceStatus === 'restarting' || instanceStatus === 'shutting_down') {
       healthService.refreshHealth(instanceName).catch(error => {
         console.error(`Failed to refresh health for ${instanceName}:`, error)
       })

webui/src/lib/api.ts

@@ -1,4 +1,5 @@
 import type { CreateInstanceOptions, Instance } from "@/types/instance";
+import type { AppConfig } from "@/types/config";
 import { handleApiError } from "./errorUtils";
 
 // Adding baseURI as a prefix to support being served behind a subpath
@@ -73,6 +74,9 @@ export const serverApi = {
 
   // GET /backends/llama-cpp/devices
   getDevices: () => apiCall<string>("/backends/llama-cpp/devices", {}, "text"),
+
+  // GET /config
+  getConfig: () => apiCall<AppConfig>("/config"),
 };
 
 // Backend API functions

webui/src/lib/healthService.ts

@@ -5,11 +5,12 @@ type HealthCallback = (health: HealthStatus) => void
 
 // Polling intervals based on health state (in milliseconds)
 const POLLING_INTERVALS: Record<HealthState, number> = {
-  'starting': 5000,    // 5 seconds - frequent during startup
-  'restarting': 5000,  // 5 seconds - restart in progress
-  'ready': 60000,      // 60 seconds - stable state
-  'stopped': 0,        // No polling
-  'failed': 0,         // No polling
+  'starting': 5000,      // 5 seconds - frequent during startup
+  'restarting': 5000,    // 5 seconds - restart in progress
+  'shutting_down': 3000, // 3 seconds - monitor shutdown progress
+  'ready': 60000,        // 60 seconds - stable state
+  'stopped': 0,          // No polling
+  'failed': 0,           // No polling
 }
 
 class HealthService {
@@ -96,6 +97,7 @@ class HealthService {
       case 'running': return 'starting' // Should not happen as we check HTTP for running
       case 'failed': return 'failed'
       case 'restarting': return 'restarting'
+      case 'shutting_down': return 'shutting_down'
     }
   }
 

webui/src/lib/zodFormUtils.ts

@@ -126,7 +126,7 @@ export function getAdvancedBackendFields(backendType?: string): string[] {
   const fieldGetter = backendFieldGetters[normalizedType] || getAllLlamaCppFieldKeys
   const basicConfig = backendFieldConfigs[normalizedType] || basicLlamaCppFieldsConfig
 
-  return fieldGetter().filter(key => !(key in basicConfig))
+  return fieldGetter().filter(key => !(key in basicConfig) && key !== 'extra_args')
 }
 
 // Combined backend fields config for use in BackendFormField

webui/src/main.tsx

@@ -4,13 +4,16 @@ import App from './App'
 import { InstancesProvider } from './contexts/InstancesContext'
 import './index.css'
 import { AuthProvider } from './contexts/AuthContext'
+import { ConfigProvider } from './contexts/ConfigContext'
 
 ReactDOM.createRoot(document.getElementById('root')!).render(
   <React.StrictMode>
     <AuthProvider>
-      <InstancesProvider>
-        <App />
-      </InstancesProvider>
+      <ConfigProvider>
+        <InstancesProvider>
+          <App />
+        </InstancesProvider>
+      </ConfigProvider>
     </AuthProvider>
   </React.StrictMode>,
 )

webui/src/schemas/backends/llamacpp.ts

@@ -167,6 +167,9 @@ export const LlamaCppBackendOptionsSchema = z.object({
   fim_qwen_7b_default: z.boolean().optional(),
   fim_qwen_7b_spec: z.boolean().optional(),
   fim_qwen_14b_spec: z.boolean().optional(),
+
+  // Extra args
+  extra_args: z.record(z.string(), z.string()).optional(),
 })
 
 // Infer the TypeScript type from the schema

webui/src/schemas/backends/mlx.ts

@@ -25,6 +25,9 @@ export const MlxBackendOptionsSchema = z.object({
   top_k: z.number().optional(),
   min_p: z.number().optional(),
   max_tokens: z.number().optional(),
+
+  // Extra args
+  extra_args: z.record(z.string(), z.string()).optional(),
 })
 
 // Infer the TypeScript type from the schema

webui/src/schemas/backends/vllm.ts

@@ -125,6 +125,9 @@ export const VllmBackendOptionsSchema = z.object({
   override_pooling_config: z.string().optional(),
   override_neuron_config: z.string().optional(),
   override_kv_cache_align_size: z.number().optional(),
+
+  // Extra args
+  extra_args: z.record(z.string(), z.string()).optional(),
 })
 
 // Infer the TypeScript type from the schema

webui/src/schemas/instanceOptions.ts

@@ -36,6 +36,10 @@ export const CreateInstanceOptionsSchema = z.object({
   // Environment variables
   environment: z.record(z.string(), z.string()).optional(),
 
+  // Execution context overrides
+  docker_enabled: z.boolean().optional(),
+  command_override: z.string().optional(),
+
   // Backend configuration
   backend_type: z.enum([BackendType.LLAMA_CPP, BackendType.MLX_LM, BackendType.VLLM]).optional(),
   backend_options: BackendOptionsSchema.optional(),

webui/src/types/config.ts

@@ -0,0 +1,70 @@
+export interface BackendSettings {
+  command: string
+  args: string[]
+  environment?: Record<string, string>
+  docker?: DockerSettings
+  response_headers?: Record<string, string>
+}
+
+export interface DockerSettings {
+  enabled: boolean
+  image: string
+  args: string[]
+  environment?: Record<string, string>
+}
+
+export interface BackendConfig {
+  'llama-cpp': BackendSettings
+  vllm: BackendSettings
+  mlx: BackendSettings
+}
+
+export interface ServerConfig {
+  host: string
+  port: number
+  allowed_origins: string[]
+  allowed_headers: string[]
+  enable_swagger: boolean
+  response_headers?: Record<string, string>
+}
+
+export interface InstancesConfig {
+  port_range: [number, number]
+  data_dir: string
+  configs_dir: string
+  logs_dir: string
+  auto_create_dirs: boolean
+  max_instances: number
+  max_running_instances: number
+  enable_lru_eviction: boolean
+  default_auto_restart: boolean
+  default_max_restarts: number
+  default_restart_delay: number
+  default_on_demand_start: boolean
+  on_demand_start_timeout: number
+  timeout_check_interval: number
+}
+
+export interface AuthConfig {
+  require_inference_auth: boolean
+  inference_keys: string[] // Will be empty in sanitized response
+  require_management_auth: boolean
+  management_keys: string[] // Will be empty in sanitized response
+}
+
+export interface NodeConfig {
+  address: string
+  api_key: string // Will be empty in sanitized response
+}
+
+export interface AppConfig {
+  server: ServerConfig
+  backends: BackendConfig
+  instances: InstancesConfig
+  auth: AuthConfig
+  local_node: string
+  nodes: Record<string, NodeConfig>
+  version?: string
+  commit_hash?: string
+  build_time?: string
+}

webui/src/types/instance.ts

@@ -11,9 +11,9 @@ export const BackendType = {
 
 export type BackendTypeValue = typeof BackendType[keyof typeof BackendType]
 
-export type InstanceStatus = 'running' | 'stopped' | 'failed' | 'restarting'
+export type InstanceStatus = 'running' | 'stopped' | 'failed' | 'restarting' | 'shutting_down'
 
-export type HealthState = 'stopped' | 'starting' | 'ready' | 'failed' | 'restarting'
+export type HealthState = 'stopped' | 'starting' | 'ready' | 'failed' | 'restarting' | 'shutting_down'
 
 export interface HealthStatus {
   state: HealthState
@@ -27,5 +27,4 @@ export interface Instance {
   name: string;
   status: InstanceStatus;
   options?: CreateInstanceOptions;
-  docker_enabled?: boolean; // indicates backend is running via Docker
 }

webui/tsconfig.json

@@ -19,7 +19,7 @@
     "paths": {
       "@/*": ["./src/*"]
     },
-    "types": ["vite/client"]
+    "types": ["vite/client", "@types/node"]
   },
   "include": ["src", "src/vite-env.d.ts"],
   "references": [{ "path": "./tsconfig.node.json" }]