Skip to content

Local-first quickstart

Install ONE with Python 3.12 or newer:

Terminal window
python3 -m pip install one-agent
one --version

Start an Ollama server separately, with a model available, then choose its actual model name in place of MODEL_NAME:

Terminal window
one --provider ollama --model MODEL_NAME --ollama-url http://localhost:11434/v1

Ask it to explain a small repository. Alternatively, use a local llama.cpp server with --provider llama.cpp --model MODEL_NAME --llama-cpp-url http://127.0.0.1:8080. These local providers do not require an API key; they still require a running, compatible server and model. ONE can execute tools with your permissions: use --preset readonly --no-extensions --no-mcp for a more restricted inspection, not as an OS or network sandbox.

Local state is stored under ~/.config/one by default, not necessarily inside the current repository. See configuration, CLI reference, and presets.