The open source
agentic harness
Run any model. Any provider. Your terminal.
Star on GitHubInference Providers
Any provider.
Local runtimes, hosted inference, or a hybrid setup. Swap providers without changing how you work.
Why Agent Open
Not another wrapper.
A real harness.
Any Model
Local or hosted — Llama, Qwen, DeepSeek, Mistral. Your choice, always.
Any Provider
OpenRouter, Together, Ollama, vLLM. Plug into whatever you're already running.
Fully Open Source
No vendor lock-in. No telemetry. Fork it, extend it, own it.
Terminal Native
Built for how developers actually work. No browser tabs, no context switching.
Who it's for
Built for developers
who think for themselves.
Open Source Enthusiasts
Run frontier OSS models locally. Contribute back. Be part of a community that ships in the open.
Privacy-First Teams
Your code never leaves your machines. No cloud APIs required. Full control over your data.
Cost-Conscious Devs
Use cheap inference providers instead of $20/mo subscriptions. Same power, fraction of the cost.
The Thesis
"The best coding agents shouldn't require proprietary models or cloud lock-in."
Open-source models are catching up fast. Inference is becoming a commodity. The missing piece is a harness that treats every model as a first-class citizen — local or remote, large or small. That's what we're building.