Documentation

Mini-claude in a nutshell

Mini-claude is a fast, private terminal coding agent for self-hosted models. It reads your files, edits them, runs commands and browses the web, all from your machine. It speaks the OpenAI-compatible HTTP API and ships as a single Go binary. Nothing leaves your machine.

Quickstart

You need three things:

An inference server running locally (we recommend Ollama).
A tool-calling model pulled.
The Mini-claude binary or the source.

ollama serve &
ollama pull qwen2.5-coder:7b

git clone https://github.com/hugostarte/Mini-Claude.git
cd Mini-Claude
go run ./cmd/tui

That’s it. The welcome screen appears, you describe a task, and the agent calls whatever tools it needs to get it done.

Next steps

Install

From source, prebuilt binary, or Docker Compose.

The 4 modes

auto, explain, guided, review: from ‘just do it’ to mentoring.

Tools

Files, bash, search and the browser, from your machine.

Permissions & safety

Diff before acting, auto-commit, /undo.

Configure

Environment variables, system prompts, model defaults.

Slash commands

Modes, model, workspace, audit, undo.

Models

Native vs weak tool calling, what to install.

Project memory

MINICLAUDE.md / CLAUDE.md, your conventions.

Recipes

Setups by mode: review, learning, multi-machine.

Release notes

What’s done, the limits, what’s next.