GitHub radar
Flint: a local AI model that fixes bugs and writes tests
Flint is Recursive's coding-agent model that runs on your own computer: it debugs code, adds features, and writes tests without sending your code to outside servers.
A local AI model that reads your code on its own, finds and fixes bugs, adds features, and writes tests.
When it helps
Useful if you want a fully local agent that keeps your code off outside servers, or for routine tasks and sensitive code. For the strongest agent, better options exist — like Opus and Fable in Claude Code.
Pros
- Solid bug-fixing score — 74.2% on the SWE-bench Verified benchmark
- Runs fully locally — your code never leaves your machine
- Apache 2.0 license — free to use without restrictions
Cons
- Full version needs 80 GB and a data-center-grade GPU
- Falls short of top models like Opus and Fable on hard tasks
How to set it up — step by step
- 1Open your AI agent (Claude Code or Codex) inside your test project folder.
- 2Send the agent the text from the block below — it will install Ollama and run Flint.
- 3For the compact version, make sure you're on a Mac with at least 48 GB of memory; the full version needs a data-center-grade GPU.
- 4Give the agent a simple task — fixing a bug or writing a test — and watch how it handles it.
- 5Review the result and the code itself before applying any changes.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install the Flint model by Recursive through Ollama: ollama run flint. For the compressed version you need a Mac with 48 GB of memory or more. Give it a simple task in a test repository, such as fixing a bug or writing a test, and show me what it produced.
The full model is 80 GB and needs a datacenter-grade GPU, but there is a compressed version for Mac, and for that one you need 48 GB of Apple Silicon memory or more.
Open on Hugging Face▌ More finds