lidge-jun/opencodex
Developer ToolsA local proxy that routes Codex, Claude Code, and other coding tools to any LLM provider.
What it does
opencodex is a local gateway between AI coding clients and model providers. It accepts OpenAI Responses, Chat Completions and Anthropic Messages requests, translates them into the provider's native format, and forwards streaming, tool calls, reasoning output and images where the client, model and provider support them. Built-in providers include OpenAI, Anthropic, Gemini, xAI, Kimi, DeepSeek, Groq, OpenRouter, Ollama and any OpenAI-compatible endpoint.
The main targets are Codex, Claude Code, Claude Desktop and Grok Build, which keep their own interfaces but can run on a routed model. Other clients (OpenCode, Cline, Kilo, Qoder and similar) are connected through managed configurations or a custom base URL. A web dashboard on port 10100 handles provider accounts, model selection and request logs.
Quick start
Requirements: Node.js 18+ for the CLI install. Install the package globally and start the proxy:
npm install -g --allow-scripts=bun @bitkyc08/opencodex
ocx start
The Bun runtime is bundled by the install script. Then open http://localhost:10100, add a provider and select models in the dashboard. For Claude Code, run ocx claude; for Codex, enable its switch on the Connect page or run ocx init. To call the proxy from any other tool, set the base URL to http://localhost:10100/v1 for OpenAI-compatible clients or http://localhost:10100 for Anthropic-style clients. The client support page lists the integration path and protocol for each tool.
Desktop builds for macOS, Windows and Linux are available from the latest release; they bundle the proxy and do not require Node or Bun.
Routing and advanced options
Select a provider with provider/model, e.g. codex -m "anthropic/claude-opus-5" or ollama/llama3. Combos expose one virtual model ID with failover or weighted round-robin across providers. A ChatGPT account pool can spread Codex work across multiple logins with quota-aware switching.
Constraints
- The CLI needs Node 18+; source builds require Bun on
PATH. - The proxy binds to
127.0.0.1by default. Binding to other interfaces requiresOPENCODEX_API_AUTH_TOKEN, and clients must send it asx-opencodex-api-key. - Windows desktop builds are still rolling out Authenticode signing; until signed, SmartScreen may require the one-time "Run anyway" step. Verify
.sha256files next to each release asset. - Some providers, especially Anthropic, may suspend accounts that route traffic through third-party proxies. Check each provider's terms before enabling it. The project states it is not affiliated with OpenAI or Anthropic.
- Anonymous daily usage telemetry is sent after the user has seen the notice and not declined; it can be disabled via
ocx telemetry disable,OCX_TELEMETRY=0orDO_NOT_TRACK=1. - License: MIT.
HotRepo's take
On 2026-11-13 the repository recorded 55 stars in the TypeScript daily ranking, with a best position of 17 on that list, out of 18,468 stars and 1,469 forks. That is a single snapshot rather than a trend, but it places the project among the more active TypeScript repos that day. The README's breadth — protocol translation, 40+ providers, combos, sub-agent routing, account pools and a desktop app — suggests a tool aimed at developers who already use multiple coding agents and want one local routing layer rather than a new client.
Written by AI from the project's README, updated . The project's own docs are the source of truth.
Hacker News
You might also like
- decolua/9routerA local router that points AI coding CLIs and IDEs at multiple providers, with quota-aware fallback and token compression.
- KenKaiii/gg-frameworkA desktop coding agent and TypeScript framework that connects to existing AI subscriptions and adds local code search.
- asgeirtj/system_prompts_leaksArchived system prompts from ChatGPT, Claude, Gemini, and dozens of other AI products for prompt researchers and builders.
- michael-denyer/pstack-claudeA port of Cursor's opinionated skill stack that turns any coding agent into a verified, workflow-driven worker.
- QuantumNous/new-apiA self-hosted gateway that puts many model providers behind one OpenAI-, Claude- or Gemini-compatible API for teams and agents.