Skip to content

Argo Proxy

PyPI version PyPI pre-release

A universal API gateway for LLM services via Argo — translates between OpenAI, Anthropic, and Google GenAI formats, routing requests to the optimal upstream ARGO endpoint automatically.

Quick Start

pip install argo-proxy        # install the package
argo-proxy serve              # run the proxy in universal mode

Key Features

  • Universal API Gateway — Serve all 4 major API formats from a single proxy:
    • OpenAI Chat Completions (/v1/chat/completions)
    • OpenAI Responses (/v1/responses)
    • Anthropic Messages (/v1/messages)
    • Google GenAI (/v1beta/models/{model}:generateContent)
  • Smart Model Routing — Automatically routes to the optimal upstream (OpenAI-compatible or native Anthropic) based on the requested model
  • Cross-Format Translation — powered by llm-rosetta, seamlessly converts between API formats when needed
  • CLI Tool Ready — Works out of the box with Claude Code, Codex CLI, Aider, Gemini CLI, OpenCode, and more (see CLI Tools Guide)
  • Native Function Calling — Full tool call support across all providers and formats
  • Vision Support — Automatic image URL download and base64 conversion
  • Live Model Refresh — Reload model list via POST /refresh without restarting
  • Self-Update — Built-in argo-proxy update check and argo-proxy update install commands

NOTICE OF USAGE

The machine or server making API calls to Argo must be connected to the Argonne internal network or through a VPN on an Argonne-managed computer if you are working off-site. Your instance of the argo proxy should always be on-premise at an Argonne machine. The software is provided "as is," without any warranties. By using this software, you accept that the authors, contributors, and affiliated organizations will not be liable for any damages or issues arising from its use. You are solely responsible for ensuring the software meets your requirements.

Citation

Argo-Proxy uses LLM-Rosetta for cross-provider API translation. If you use this translation layer in research, please cite:

@article{ding2026llm,
  title={LLM-Rosetta: A Hub-and-Spoke Intermediate Representation for Cross-Provider LLM API Translation},
  author={Ding, Peng},
  journal={arXiv preprint arXiv:2604.09360},
  year={2026}
}

Get Involved