This project provides a local proxy to enable the latest OpenAI Codex CLI and desktop applications to interface with a wide range of modern Large Language Models (LLMs). It addresses compatibility issues with newer Codex versions and offers flexibility by supporting various LLM backends, including built-in MiMo and DeepSeek models, alongside any OpenAI Chat Completions-compatible API (like Qwen, GLM, Kimi, Ollama, vLLM) or native Responses API. This allows users to leverage advanced LLMs with their existing Codex setup without code modifications.

How It Works

Mimo2codex functions as an intermediary, translating requests between the Codex client's "Responses API" format and the upstream LLM's "Chat Completions API" format on the fly. It employs a generic provider mechanism, allowing users to configure various LLM endpoints. Requests are dynamically routed based on the model field specified by the client, enabling seamless switching between different LLM providers within a single proxy instance. This approach avoids the need for client-side code changes or re-publishing the Codex application.

Quick Start & Requirements

• Installation: Primarily via npm install -g mimo2codex. Alternative methods include a curl one-liner, PowerShell script, or manual Git clone and build.
• Prerequisites: Node.js version 18 or higher is required.
• Setup: Users need to obtain API keys for their chosen LLM providers. The proxy can be started using mimo2codex init to set up environment variables, followed by mimo2codex. Configuration for Codex involves updating auth.json and config.toml files. Docker deployment is also supported via docker compose up -d.
• Documentation: Detailed setup, deployment, and provider guides are available in the doc/ directory.

Highlighted Details

• Multi-Provider Support: Integrates MiMo V2.5/DeepSeek V4 Pro natively, with extensive support for third-party OpenAI-compatible APIs (Qwen, GLM, Kimi, Ollama, vLLM, LM Studio, etc.) via a generic provider mechanism.
• Codex Compatibility: Specifically designed to resolve issues with newer Codex versions that may hard-error on older API configurations, ensuring compatibility with the latest releases.
• Tool Calling: Supports advanced features like function tools, parallel calls, local_shell, and custom tool integrations.
• Admin Web Console: Provides a local dashboard (http://127.0.0.1:8788/admin/) for monitoring token usage, viewing logs, managing models, and configuring providers.
• Codex Enable Feature: Offers one-click model switching directly from the admin UI, simplifying configuration updates for Codex.

Maintenance & Community

The project shows active development with recent releases (e.g., v0.2.17 on May 19, 2026). A detailed version history is maintained in doc/tag-log.md. No specific community channels like Discord or Slack are mentioned in the README.

Licensing & Compatibility

The project is released under the MIT License, permitting broad use, modification, and distribution, including for commercial purposes.

Limitations & Caveats

Codex's /hatch command for custom pet generation cannot be directly intercepted by the proxy due to client-side OpenAI API calls; a workaround is provided via the mimoskill scripts. Vision support is limited to specific MiMo models. Older versions of the proxy may exhibit issues with multi-turn tool calls on certain models, requiring an upgrade to version 0.2.3 or later. Proper handling of MiMo's reasoning_content is crucial for multi-turn tool-calling workflows to prevent errors or degraded performance.