← agnt

Run a coding agent on local models

agnt works with Ollama and LM Studio, so your code can stay on your machine. It also supports hosted providers when you want a bigger model.

Set up

  1. Install agnt and start your local server (Ollama or LM Studio).
  2. Run /model and pick a local model.
  3. Use /provider to add any other OpenAI-compatible endpoint.

Mix local and hosted

Use /fallback to name models to switch to when the current one errors or is rate-limited. Supported hosted providers: Anthropic, OpenAI, OpenRouter, Gemini, DeepSeek, Groq, xAI and Mistral.

Keep control

Use plan mode to have the agent read and propose before anything changes, /undo and /rewind to roll back, and /worktree to work on an isolated branch.

Scripts and CI

The same agent runs headless inside a script or CI step. See the docs for every command, or install agnt.