# opencodex

> A local proxy that lets you run OpenAI Codex, Claude Code, Claude Desktop, and Grok Build on any LLM provider by translating wire formats with streaming, tool calls, and image support.

opencodex is an open-source, community-maintained local proxy that routes OpenAI Codex, Claude Code, Claude Desktop, and Grok Build through any LLM provider — including Anthropic, Google Gemini, xAI, DeepSeek, Ollama, and 40+ others — without patching the client. It is published under the MIT License and available on GitHub at lidge-jun/opencodex, with documentation at opencodex.me. The project is not affiliated with or endorsed by OpenAI, Anthropic, or any other provider.

## What It Is

opencodex is a lightweight local proxy written in TypeScript that intercepts the OpenAI Responses API format used by Codex and translates it bidirectionally to whatever protocol a target provider speaks. This means streaming completions, tool calls, reasoning tokens, and images all flow correctly in both directions without any modification to the Codex CLI, App, SDK, or Claude Code client. The proxy runs on `localhost:10100` by default and exposes a web dashboard for configuration and live request monitoring.

## How the Routing Works

The proxy uses five adapters to cover the major wire formats:
- **Anthropic Messages** — for Claude models
- **Google Gemini** — for Gemini models
- **Azure OpenAI** — for Azure-hosted models
- **OpenAI Responses passthrough** — for native OpenAI endpoints
- **OpenAI-compatible Chat Completions** — for any compatible endpoint (DeepSeek, Groq, OpenRouter, Ollama, Together, Fireworks, Cerebras, Mistral, Hugging Face, NVIDIA NIM, and more)

Models are addressed with a `provider/model` syntax (e.g., `anthropic/claude-opus-5` or `google/gemini-3-pro`). A "Combos" feature lets users define a virtual model ID backed by failover or weighted round-robin across multiple providers.

## Platform and Deployment Model

opencodex runs as a local process or background service on macOS (arm64/x64), Linux (x64/arm64), and Windows (x64). It requires Node 18+; the Bun runtime is bundled automatically on install so no separate Bun installation is needed. Key deployment options include:

- **Foreground proxy** — `ocx start`
- **Background service** — `ocx service` (launchd on macOS, systemd on Linux, Task Scheduler or WinSW on Windows)
- **On-demand shim** — `ocx codex-shim install` starts the proxy whenever `codex` launches

Authentication supports OAuth (xAI, Anthropic, Kimi with auto-refresh), API keys, environment variable references, and a ChatGPT account pool with thread affinity and quota-aware auto-switching.

## Key Capabilities

- **Sub-agent surface** — pin up to five routed or native models for Codex `spawn_agent`, with v1/v2 surface control and fallback chains
- **Claude Code integration** — `ocx claude` launches Claude Code against the same port with every routed model available in the `/model` picker and auto-context up to 1M tokens
- **Search and vision sidecars** — non-OpenAI models gain real web search and image understanding through a GPT sidecar over a ChatGPT login
- **Image and video bridges** — documented guides for routing image and video inputs through the proxy
- **Management API** — a REST API for programmatic control alongside the CLI and web dashboard
- **Clean uninstall** — `ocx stop` restores Codex to its original configuration with zero residue

## Update: v2.11.0

The latest release is v2.11.0, published on August 8, 2026, with the repository last pushed the same day. The project was created in June 2026 and has accumulated over 8,500 GitHub stars and 650 forks in under two months, according to GitHub metadata. The repository is actively maintained with 110 open issues and documentation available in English, Korean, Simplified Chinese, Russian, and Japanese.

## Features
- Local proxy for Codex, Claude Code, Claude Desktop, and Grok Build
- 40+ built-in LLM providers
- Bidirectional streaming, tool calls, reasoning tokens, and image support
- Five protocol adapters: Anthropic Messages, Google Gemini, Azure, OpenAI Responses, OpenAI-compatible Chat Completions
- Model routing with provider/model syntax
- Combos: virtual model IDs with failover or weighted round-robin
- ChatGPT account pool with thread affinity and quota-aware auto-switching
- OAuth support for xAI, Anthropic, and Kimi with auto-refresh
- Sub-agent surface with v1/v2 control and fallback chains
- Claude Code integration with up to 1M auto-context
- Search and vision sidecars for non-OpenAI models
- Image and video bridges
- Web dashboard with live request log and cache token counts
- Management API for programmatic control
- Background service mode (launchd, systemd, Task Scheduler/WinSW)
- On-demand codex-shim for lightweight startup
- Clean uninstall restoring native Codex configuration
- Remote access with bearer token authentication

## Integrations
OpenAI Codex, Claude Code, Claude Desktop, Grok Build, Anthropic, Google Gemini, xAI / Grok, Kimi, Azure OpenAI, Ollama, DeepSeek, Groq, OpenRouter, Together AI, Fireworks, Cerebras, Mistral, Hugging Face, NVIDIA NIM, MiniMax, Qwen Cloud, SiliconFlow, Cursor (experimental), ChatGPT

## Platforms
CLI, WEB, API, WINDOWS, MACOS, LINUX

## Pricing
Open Source

## Version
v2.11.0

## Links
- Website: https://opencodex.me/
- Documentation: https://opencodex.me/getting-started/installation/
- Repository: https://github.com/lidge-jun/opencodex
- EveryDev.ai: https://www.everydev.ai/tools/opencodex