Quickstart

This takes about five minutes: install AlbertCode SWE Agent, connect a model, run a first task. Using mini-AlbertCode SWE Agent instead? The steps are the same apart from the command's name, albert-mini in place of albertcode; each step shows both.

1. Install

AlbertCode SWE Agent

macOS and Linux:

curl -fsSL https://raw.githubusercontent.com/huynvic/albertcode-swe-agent/main/installer/install.sh | sh

Windows (PowerShell):

irm https://raw.githubusercontent.com/huynvic/albertcode-swe-agent/main/installer/install.ps1 | iex

Or: mini-AlbertCode SWE Agent

macOS and Linux:

curl -fsSL https://raw.githubusercontent.com/huynvic/mini-albertcode-swe-agent/main/installer/install.sh | sh

Windows (PowerShell):

irm https://raw.githubusercontent.com/huynvic/mini-albertcode-swe-agent/main/installer/install.ps1 | iex

The installer shows what it found and what it would do, then asks Proceed? [Y/N]. Nothing changes before you answer. It installs everything the agent needs, including Python, for your user only. When it finishes, open a new terminal so it picks up the command.

2. Check it

albertcode --version     # or: albert-mini --version
albertcode doctor        # or: albert-mini doctor

doctor shows which build tools the agent can find (AlbertCode SWE Agent), or checks the installation and its service and says what to do about any problem (mini-AlbertCode).

3. Start it in a repository

cd your-project
albertcode               # or: albert-mini

The first command starts a small local service in the background. It keeps running between sessions, so closing a window never stops a task. Prefer the browser? Run albertcode web or albert-mini web instead: it opens the same sessions, signed in.

4. Connect a model

The first time, with no model connected, the connect step opens by itself:

1. Pick a provider, then paste its API key — or give the address of a model server running on your machine.
2. The key or address is checked before it is saved.
3. Choose a model from the list that opens next.

Skipped it? Type /connect at the agent's prompt any time, and /model to switch. The model must support tool calling; albertcode models test (or albert-mini models test) checks one. See Configuration.

5. Ask for a change

Type what you want, in plain words:

› The /login endpoint returns 500 when the password is empty. Find out why, fix it, and add a test.

In Governed mode, the default, the agent reads the repository and comes back with a plan: what it will change, which files it expects to touch, how it will test the change, and how risky it is.

6. Approve the plan

Read the plan. Approve it to start the work, ask for changes, or reject it. Nothing has been written yet.

7. Review the result, then accept or discard

The agent makes the change in an isolated copy of your repository, runs your tests and checks, and reviews the change. Then it shows you the diff and what passed. Your own files are still untouched.

Accept to apply the change to your repository, or discard it. Accepted changes are ordinary edits: commit them with Git as you normally would — and undo them with Git, too.

Next

• Installation — options, manual downloads, updating and uninstalling.
• Configuration — models, keys, the repository contract, commands, agents and MCP servers.
• Agent execution — modes, gates, and running a request from a script or CI.
• Troubleshooting — fixes for common problems.