Getting started

A first task, in about five minutes. You need mini-AlbertCode installed (Installation) and a Git repository to work in.

1. Open it in your repository

cd your-project
albert-mini

The terminal opens on a new conversation: the name in solid letters, the version, the folder and the model, and the commands to start with. The prompt at the bottom says where you are, which model answers and which mode you are in:

your-project · gpt-4o · governed ❯ Ask anything, or say what to change; a change is planned first

Prefer the browser? albert-mini web opens the same thing there. Both show the same sessions.

2. Connect a model

The first time, nothing is connected, so Connect a provider opens by itself:

No model connected yet. Pick a provider to connect one now, or press Esc to skip and use /connect later.

   Connect a provider
   Search: ▏                          type to filter · ↑↓ · Enter · Esc
   POPULAR
 › OpenAI                             OpenAI's GPT models
   Anthropic                          Anthropic's models
   OpenRouter (recommended)           One key for most of the market, open models included
   OTHER
   ...
   ON THIS MACHINE
   Local server                       A model server on this machine: Ollama, LM Studio, llama.cpp

Type to filter, choose with the arrows and enter, and paste your API key when it asks (or give a local server's address; its https:// or http:// is already there). The key is checked with the provider before anything is saved, and it is never shown again. Then the models your key can use are listed: pick one, and it is used at once. See Models.

3. Ask a question

Questions are always just answered, in every mode:

your-project · gpt-4o · governed ❯ What does this project do, and how are its tests run?

4. Ask for a change

In the Governed mode (the default), a change starts on a plan:

your-project · gpt-4o · governed ❯ The discount in cart.py is applied twice. Fix it and add a test.

mini-AlbertCode reads the repository and proposes a plan, and waits:

Approve this plan?
Nothing has been changed yet · 2 files · risk low

❯ 1. approve   a   Approve the plan and begin isolated implementation.
  2. revise    e   Say what should change; a revised plan comes back.
  3. details   d   Read the whole plan before deciding.
  4. reject    r   Discard this plan and delete its temporary workspace.

Press a (or enter). It builds the change in an isolated copy of your repository, runs your tests, reviews the result, and stops again:

Apply reviewed changes?
Built and checked in an isolated copy; your files are untouched until you apply it

Press v to read the diff, then a to apply it. Only now do your files change. See Modes and approvals.

5. Next

• tab switches between the modes; Direct works in your files directly, each write asking first.
• /help lists every command; /keys every key.
• ctrl+x b shows the sidebar with the session's mode, model, task and changed files.
• Close the terminal at any time: the task keeps going, and albert-mini -c continues the session.

Then:

• How mini-AlbertCode works: the ideas behind every page.
• Modes and approvals: the modes, the two gates and every answer.
• The terminal and the browser: every key and command.
• Custom commands and agents and MCP servers: your own commands, roles and tools.
• All the documentation.