AI NewsInfrastructureAnnouncement
My Free Code, a model gateway, reached 537 stars in four days
An MIT-licensed gateway that supports the Anthropic Messages API and the OpenAI Responses API, then routes to any of about 45 providers with fallbacks and local models. 537 stars in four days.
Image: GitHub
Why it mattersA coding agent that talks to a gateway instead of a provider can move to a different model with one configuration change, which is the difference between a quick switch and a long migration.
My Free Code is an open-source gateway that sits between a coding agent and a model provider. It supports the Anthropic Messages API on /v1/messages and an OpenAI Responses-compatible endpoint on /v1/responses, then routes the request onward. The repository was created on 27 August 2026 and had 537 stars four days later, under an MIT licence.
The README says plainly that it is an independent implementation and not affiliated with Anthropic. Worth repeating, because the name does not say that.
What it does
The gateway handles streaming, tool definitions and tool calls, tool results, images and reasoning metadata pass-through. It supports ordered model fallback, per-provider concurrency limits, provider health backoff and a rate window.
The provider list is long, around 45 entries, including OpenRouter, Groq, Together, Fireworks, DeepInfra, Cerebras, SambaNova, Azure OpenAI, Amazon Bedrock, Google Vertex, DeepSeek, Mistral, Cohere, Hugging Face, and the local runners Ollama, LM Studio and llama.cpp.
The project is careful here in a way that is worth noting. It says the provider list is not a claim of universal support: the common OpenAI-compatible providers share one transport, and providers with unusual authentication need a dedicated adapter. That is an honest thing to write in a README and it is the sentence to check first.
There are launcher adapters for Claude Code, Codex, OpenCode, Cline, Grok Build, DeepSeek Harness and others. A launcher sets up the local proxy environment and passes the arguments to the client you already have.
The argument for a gateway
The reason this pattern exists is that the arrangement between a model provider and a client tool can end at either side's decision, for reasons that have nothing to do with any customer of either company.
A team whose agent talks to a gateway changes one configuration file when that happens. A team whose agent talks directly to a provider changes the tool.
The same applies to less dramatic events: a price change, a rate limit, a region outage, a model deprecated with 30 days notice. Every one of those is a small annoyance with a gateway and a full project without one.
What you take on
You are adding an extra network step and a piece of software to the path between your developers and their models. When it is down, everyone is down. It is at version 0.8 and days old.
There is also a real question about what "free" means here. Routing to a provider still costs whatever that provider charges. The saving comes from reaching cheaper or local models, which is a decision about quality as much as price.
Read the fallback ordering carefully before relying on it for real work. A fallback with no warning to a weaker model is worse than an error, because the code still gets written.
Source
- hkqr/my-free-code on GitHub, MIT licence. Star and date figures read from the GitHub API on 31 August 2026
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.