Quickstart

This page takes you from nothing to your first Tau session. It should take a few minutes.

1. Install Tau

Tau is a Python tool. The easiest way to install it is with uv:

uv tool install tau-ai

Check it worked:

tau --version

2. Connect a model

Tau needs an AI model to talk to. A provider is the service that hosts the model (OpenAI, Anthropic, …). Start Tau and use /login to connect one:

tau

Then run one of these inside Tau:

/login              # choose a provider
/login openai       # save an OpenAI API key
/login openai-codex # authenticate a Codex/ChatGPT subscription

Tau ships with built-in entries for OpenAI, Anthropic, OpenAI Codex, OpenRouter, and Hugging Face. See Providers & models for switching models or adding a custom/local OpenAI-compatible endpoint.

3. Start a session

Run Tau from inside the project you want to work on:

cd my-project
tau

This opens the interactive terminal UI. Type a request and press Enter:

explain what this project does

Tau streams its response, and when it needs to, it reads files and runs commands to answer you. Try something that changes code:

add a docstring to every function in src/utils.py

You’ll see each tool call (read, edit, bash) as it happens.

4. Come back later

Tau saves every session. List them:

tau sessions

Resume the most recent one for this directory, or pick from a list:

tau --resume <session-id>

…or open the picker inside the TUI with /resume. See Sessions for resuming, branching, and exporting.

One-shot mode

Don’t need the UI? Run a single prompt and get the result on stdout — handy for scripts and pipes:

tau -p "summarize the changes in the last commit"

Where to go next

Core concepts — understand what’s actually happening.
The interactive session — get fluent in the TUI.
Providers & models — switch models, add providers, use local models.