Local AI Bridge runs a localhost server inside VS Code and exposes chat-capable vscode.lm models through:

• HTTP/WebSocket endpoints for local clients
• streamed responses (SSE or WebSocket)
• a built-in browser chat UI at http://127.0.0.1:<port>/

The bridge is provider-agnostic. It discovers models via vscode.lm.selectChatModels() and does not hardcode model vendors.

• Developers who want a local bridge between VS Code vscode.lm models and external clients
• Teams validating chat integration behavior (streaming, stop/cancel, connection lifecycle)
• Developers testing OpenAI-ready apps against local model providers with a thin adapter layer

Local AI Bridge exposes a custom bridge API (/chat, /models, /chat/stream, /chat/ws), not a full OpenAI API clone. If your app expects OpenAI endpoints directly, add an adapter/proxy layer.

• Binds to 127.0.0.1 only (never 0.0.0.0)
• Activity Bar integration with a dedicated Local AI Bridge icon
• Native collapsible sidebar sections:
  • Connection (status + Start/Stop)
  • Controls (port/transport/autoStart + Save/Reset)
  • Event Logs (live bridge events)
• Command Palette support:
  • Local AI: Start Server
  • Local AI: Stop Server
• Streaming transport modes:
  • SSE (/chat/stream)
  • WebSocket (/chat/ws)
• Built-in UI + standalone external demo web app

• VS Code ^1.95.0
• Node.js 18+ (recommended for local build)
• At least one VS Code extension that exposes models through vscode.lm

1. Open this project in VS Code.
2. Run:

   npm install
   npm run compile

3. Press F5 to launch an Extension Development Host window.
4. In the host window, click the Local AI Bridge icon in the Activity Bar.

1. Build and package:

   npm install
   npm run compile
   npx @vscode/vsce package

2. Install:
   • Command Palette: Extensions: Install from VSIX...
   • choose the generated .vsix (for example local-ai-bridge-0.0.1.vsix)
3. Reload VS Code and open the Local AI Bridge Activity Bar view.

• Shows current status (Ready / Stopped / error)
• Shows URL, transport, and port
• Start and Stop control the bridge server

• Edit runtime settings:
  • Port
  • Transport (sse or websocket)
  • Auto start on launch
• Save persists settings
• Reset re-syncs fields from current extension state

• Displays live bridge-wide events (/events)
• Includes request/response, stream, and connection lifecycle logs

{
  "localAI.transport": "sse",
  "localAI.port": 3000,
  "localAI.autoStart": true
}

• localAI.transport: "sse" or "websocket" (default: "sse")
• localAI.port: 1024-65535 (default: 3000)
• localAI.autoStart: starts server on VS Code launch

If the server is running, changing localAI.port or localAI.transport restarts it automatically.

Use either client depending on your goal:

• Built-in UI (webview/index.html)
  • URL: http://127.0.0.1:<port>/
  • Best for quick validation of extension behavior
• External demo web (demo-web/index.html)
  • Run: npm run demo
  • URL: http://127.0.0.1:3001
  • Best for testing an external browser app against the bridge API

Base URL: http://127.0.0.1:<port>

• GET /health -> { "ok": true }
• GET /config -> { "transport": "sse|websocket", "port": <number>, "version": "..." }
• GET /models -> { "models": [...] }
• POST /chat
  • request: { "prompt": "...", "model": "...", "clientId": "..." }
  • response: 202 { "accepted": true, "clientId": "..." }
• POST /chat/stop
  • request: { "clientId": "..." }
  • response: { "stopped": true|false, "clientId": "..." }
• GET /chat/stream?clientId=... (SSE mode only)
• WS /chat/ws?clientId=... (WebSocket mode only)
  • optional client message: { "type": "stop" }

{ "delta": "partial text", "done": false }

{ "delta": "", "done": true }

{ "error": "message", "done": true }

• npm run compile build TypeScript to out/
• npm run watch TypeScript watch mode
• npm run demo run external demo web at http://127.0.0.1:3001
• npm run vscode:prepublish prepublish compile

vscode.lm stream
      |
      v
 StreamTransport (interface)
   |-- SseTransport       -> GET /chat/stream
   `-- WebSocketTransport -> WS  /chat/ws

Key files:

• src/extension.ts activation, commands, config lifecycle, sidebar wiring
• src/sidebar.ts Activity Bar multi-view webview UI
• src/server.ts server bootstrap, localhost restriction, static/config routes
• src/routes/chat.ts chat/model/stream routing and client sessions
• src/llm.ts model lookup and token streaming from vscode.lm
• src/transport/* transport implementations

• Server binds to localhost only
• Non-local remote addresses are rejected
• CORS is enabled for local web app interoperability
• No authentication (intended for trusted local use)

• No models listed:
  • install/sign in to a provider extension that supports vscode.lm
• 409 on POST /chat:
  • connect stream endpoint first (/chat/stream or /chat/ws) with same clientId
• Transport endpoint errors:
  • ensure endpoint matches localAI.transport
• Port conflict:
  • change localAI.port and retry