CodeCrab Router

Native LLM Router for CodeCrab based on node-llama-cpp.

Local Configuration

The router uses a machine-specific configuration file that is deliberately excluded from version control to protect your local paths and settings.

Setup Instructions

1. Copy the example configuration:

   cp src/config/config.example.ts src/config/config.local.ts

2. Update your local paths: Open src/config/config.local.ts and set the absolute paths to your .gguf model files, your default workspace, and your runtime/cache directories.

   export default {
       models: {
           advisor:   '/absolute/path/to/qwen-advisor.gguf',
           backend:   '/absolute/path/to/qwen-backend.gguf',
           frontend:  '/absolute/path/to/qwen-frontend.gguf',
           embedding: '/absolute/path/to/embedding.gguf',
       },
       // ...
   };

3. Run the server and the VS Code wrapper: From the project root, you can start both components simultaneously using the provided scripts:

   Linux / macOS:

   ./start.sh

   Windows:

   start.bat

Note: The server will refuse to boot if config.local.ts is missing, or if any of the configured models do not physically exist on your disk.

Model Resolution

You do not need to configure every model path. If a specific model is missing from config.local.ts, the router will fall back to automatic discovery, searching for matching .gguf files in models/base/ and models/adapters/.

LLM Links

• Google Drive: Download Models