This repository provides a proxy service to reroute requests from AI coding assistants like GitHub Copilot to various backend LLM providers, including OpenAI, DeepSeek, Siliconflow, and local Ollama instances. It targets developers seeking flexibility in their AI coding tools, enabling them to leverage different models and APIs without vendor lock-in.

How It Works

The project acts as a local HTTP server that intercepts requests from IDE plugins (VSCode, JetBrains) and forwards them to configured LLM APIs. It supports both chat completions and code generation (via codex endpoints), allowing for customization of API endpoints, models, and authentication. The configuration is managed via a config.json file or environment variables, offering granular control over API usage and model selection.

Quick Start & Requirements

• Install/Run: Configure IDEs (VSCode, JetBrains) with the local proxy address (e.g., http://127.0.0.1:8181). Run the provided scripts for IDE patching (VSCode) or follow configuration steps for JetBrains.
• Prerequisites: Python environment for running the proxy service. Specific LLM API keys and potentially local model setup (Ollama).
• Resources: Requires a running proxy service and configured IDE plugins. Local LLM usage depends on model size and hardware.
• Docs: Configuration details are within the README.

Highlighted Details

• Supports multiple LLM providers (OpenAI, DeepSeek, Siliconflow, Ollama).
• Allows configuration of chat and code generation models independently.
• Includes scripts to patch VSCode's Copilot extension for compatibility.
• Offers rate-limiting mitigation for Siliconflow by supporting multiple API keys.

Maintenance & Community

• Project is released under the MIT license.
• Discussion and contributions are welcomed on the linux.do forum.

Licensing & Compatibility

• MIT License. Allows modification and commercial use, provided original author attribution is retained.

Limitations & Caveats

• The codex_max_tokens functionality is noted as not working perfectly and has been removed.
• VSCode requires re-patching after each Copilot extension update.