nanocode

A small terminal coding agent

nanocode works in your terminal: you describe a task, and it reads code, edits files, runs commands, and reports back. It keeps stable prompt prefixes so supported providers can reuse work, maintains a searchable code index, runs background jobs, tracks its own working notes, and resumes where you left off.

Editing code and running tools in one interactive session.

Use at your own risk

nanocode edits files and runs shell commands in the directory where you start it. It has no sandbox of its own. Run it inside a container, VM, or another isolated environment when you need isolation. See Safety.

Install and run

uv tool install nanocode-cli
nanocode --init-config          # write ~/.nanocode/config.toml
# add your provider's url, key, and model to that file
nanocode

Full walkthrough: Getting started.

What it does

Working through a repository task in an interactive session.

Area	In short
Interaction	Follow-ups, commands, keys — how you drive the agent.
Tools	Read, search, navigate code; edit files; run commands; background jobs.
Sessions	Your work is saved and resumable with -c or --resume.
MCP	Connect external Model Context Protocol servers and use their tools.
Skills	Load reusable instruction packs on demand.
Providers	Any OpenAI-compatible API, plus Anthropic.