Skip to content
BranchesTags

Repository files navigation

OmniCode Proxy — Local AI Server Extension

Expose your IDE's internal language models as a local OpenAI/Ollama-compatible HTTP API.
Like Ollama, but powered by the premium high-quality Gemini models already active and authenticated inside your Google Antigravity or vanilla VS Code workspace.

📖 Table of Contents

✨ Key Features

  • 🔄 Smart Dual-Pipeline Engine — Automatically detects the environment. Uses vanilla VS Code vscode.lm APIs when available, and seamlessly switches to the high-performance direct Connect RPC harvester inside Google Antigravity.
  • 🔌 OpenAI-Compatible Endpoints — Standard /v1/chat/completions and /v1/models APIs out of the box. Compatible with client tools like Cline, Continue, Cursor, or your own Python/Node.js scripts.
  • 🦙 Ollama-Compatible Endpoints — Drop-in /api/chat and /api/tags support for immediate CLI integration.
  • 🎛️ Premium Webview Control Panel — Rich, modern dashboard interface to monitor server status, view logs in real-time, inspect discovered models, and manage secure API keys.
  • 🔑 Secure Token Authentication — Native, encrypted bearer token management utilizing the IDE's secure storage (SecretStorage).
  • 📡 Status Bar Widget — Real-time server status, binding address, active port, and live request counting.
  • 📊 Real-time Logging — Log console output channel in the control panel + dedicated output console channel for easy API diagnostics.
  • 🧱 Graceful Sandbox Fallback — Built-in sandbox mode that lets you test integrations and endpoints seamlessly even without active logged-in models.

📋 Requirements & OS Compatibility

  • Visual Studio Code v1.93.0 or higher, OR Google Antigravity IDE v1.107.0+.
  • An active AI provider extension installed and authenticated in your IDE (e.g., Google Gemini Code Assist, GitHub Copilot, etc.).

💻 Operating System Support

Because Omni Bridge dynamically detects your environment, its features depend slightly on your OS:

  • Standard VS Code Pipeline (vscode.lm): 🟢 Fully supported on Windows, macOS, and Linux.
  • Antigravity Direct Connect Bypass: 🟢 Fully supported on Windows and Linux. On Windows, it utilizes PowerShell and native netstat commands to dynamically harvest processes and active sockets. On Linux, it uses ps aux and ss.

💡 How it Works (The Basic Explanation)

Normally, third-party extensions cannot access the built-in Gemini models inside Antigravity IDE because they run in a secure, sandboxed container and do not register themselves to public extension APIs.

OmniCode Proxy bypasses this limitation dynamically:

  1. Process Harvesting: The extension automatically scans running processes (using ps aux on Linux/macOS and PowerShell on Windows) to locate the active Antigravity Language Server binary (language_server_linux_x64 or language_server_windows_x64.exe).
  2. Secure Token Extraction: It harvests the highly secure, dynamic session token (--csrf_token) passed to the process at startup.
  3. Socket Discovery & SSL Probing: It scans listening ports (using ss -ltp on Linux and netstat -ano on Windows) belonging to that process and dynamically probes them to find the correct local SSL port.
  4. Direct Bridge Handshake: It establishes a direct TLS-bypassed HTTPS tunnel to the local server, communicating over the Connect protocol (a lightweight gRPC-compatible JSON format).
  5. Streaming Slice Decoder: When you prompt the model, the proxy converts your standard OpenAI request into a 5-byte enveloped Connect stream command, reads the incoming binary chunks frame-by-frame, extracts the words, and streams them back to your client instantly.

This translates to complete, latency-free access to internal premium Gemini models, right from your standard developer tools!

🚀 Quick Start

1. Install & Launch

  • Install the OmniCode Proxy extension (omnicode-proxy-1.2.0.vsix) inside your IDE.
  • Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and search for Omni Bridge: Open Control Panel.

2. Configure & Start Server

  • Adjust your desired binding port (default: 11434) and host interface.
  • Click Start Server or run Omni Bridge: Start Server from the Command Palette.

3. Generate secure API Tokens

  • Navigate to the API Tokens tab in the Control Panel dashboard and click + Generate Token.
  • Copy the generated bearer key (for security reasons, this is only shown once!).

4. Connect your scripts

🔹 curl

curl http://localhost:11434/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-lmb-...YOUR_TOKEN..." \
  -d '{
    "model": "gemini-3.5-flash-low",
    "stream": true,
    "messages": [{"role": "user", "content": "Tell me a joke!"}]
  }'

🔹 Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="sk-lmb-...YOUR_TOKEN..."
)

response = client.chat.completions.create(
    model="gemini-3.5-flash-low",
    messages=[{"role": "user", "content": "Explain quantum computing in one sentence."}]
)
print(response.choices[0].message.content)

🔹 Cline / VS Code Extensions Configuration

Configure OmniCode Proxy in client extensions as an OpenAI-compatible provider:

  • Base URL: http://localhost:11434/v1
  • API Key: sk-lmb-...YOUR_TOKEN...
  • Model ID: gemini-3.5-flash-low (or any model listed under the control panel's Discovered Models tab).

📡 API Endpoints Reference

Method Path Description
GET /v1/models Lists all discovered active language models (harvested or registered)
GET /v1/models/:id Returns details of a specific model
POST /v1/chat/completions OpenAI-compatible chat completion (supports real-time stream decoding)
GET /health Server health check (returns active status, host, and port)
POST /api/chat Ollama-compatible chat completions interface
GET /api/tags Ollama-compatible active models tags listing

⚙️ Configuration Properties

You can customize server behaviors either from the Control Panel UI or directly in the VS Code Settings under the omniBridge.* workspace scope:

Configuration Default Description
omniBridge.port 11434 The local port the API server binds to.
omniBridge.host 127.0.0.1 Host interface. Bind to 127.0.0.1 (localhost only) or 0.0.0.0 (all local network interfaces).
omniBridge.autoStart false Automatically start the server on IDE launch.
omniBridge.authEnabled true Enforce API Bearer token authentication header checks.
omniBridge.defaultModel "" Default fallback model name if the incoming request omits the model property.
omniBridge.corsOrigins * Custom CORS allowed origins list (comma-separated).
omniBridge.logLevel info Logs verbosity levels: none, error, info, or debug (verbose).
omniBridge.maxConcurrentRequests 5 Maximum number of parallel LLM inferences supported simultaneously.
omniBridge.requestTimeout 120000 Timeout threshold in milliseconds for model execution.

🏗️ Technical Architecture

OmniCode Proxy is built with a dual-layer abstraction framework:

┌────────────────────────────────────────────────────────┐
│ Client Apps (Cline, Cursor, Python/Node, curl, etc.)   │
└───────────────────────────┬────────────────────────────┘
                            │ (HTTP / OpenAI Protocol)
                            ▼
┌────────────────────────────────────────────────────────┐
│              OMNICODE PROXY EXTENSION                  │
│                                                        │
│  ┌──────────────────────────────────────────────────┐  │
│  │ 1. Local HTTP API Server (port: 11434)           │  │
│  │    - Routes: /v1/chat/completions, /v1/models    │  │
│  │    - Token validation & CORS filtering           │  │
│  └────────────────────────┬─────────────────────────┘  │
│                           │                            │
│                           ▼ (Dual-Pipeline Router)     │
│  ┌──────────────────────────────────────────────────┐  │
│  │ 2. Pipeline 1: Standard VS Code (vscode.lm)     │  │
│  │    - Active in vanilla VS Code                   │  │
│  │    - Queries registered third-party providers    │  │
│  └──────────────────────────────────────────────────┘  │
│  ┌──────────────────────────────────────────────────┐  │
│  │ 3. Pipeline 2: Antigravity Direct Connect RPC    │  │
│  │    - Scans process table (ps aux / PowerShell)   │  │
│  │    - Extracts dynamic CSRF session keys          │  │
│  │    - Dynamic socket probing (ss / netstat)       │  │
│  │    - TLS cert bypassed local HTTPS Handshake      │  │
│  │    - 5-byte Connect streaming binary slice parser │  │
│  │    └────────────────────┬────────────────────────┘  │
│  └──────────────────────────────────────────────────┘  │
└───────────────────────────┼────────────────────────────┘
                            │
                            ▼ (Direct Connect Tunnel)
            Antigravity Local Language Server
  (language_server_linux_x64 / language_server_windows_x64.exe)

📄 License

Licensed under the [MIT License].

About

No description, website, or topics provided.

Resources

Readme
MIT license
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages