Using AlbertCode
The short version of everything. Each section links to its full guide.
Three interfaces, one agent
| Interface | Start it | Good for | Guide |
|---|---|---|---|
| Terminal | albertcode |
Working where you already are; scripts and CI | Terminal |
| Browser | albertcode --ui |
Plans and diffs side by side; several sessions at once | Browser interface |
| VS Code | The AlbertCode panel | Staying in the editor; right-click a folder → Open Here | VS Code |
All three talk to the same local service, so a task started in one can be followed in another. See How AlbertCode works.
Modes
| Mode | What happens | Use it when |
|---|---|---|
| Governed (default) | You approve a plan, it works in an isolated copy, runs your tests and reviews the result, then you approve the diff | Changes you care about |
| Direct | It works in your real files and asks before each write (or, with auto, doesn't ask). Needs Git | Pairing on something you're watching |
| Ask only | It answers questions and changes nothing; a change is offered for your go | Understanding a codebase |
Switch with /mode, or start one request in a given mode with /plan or /do. See
Chatting with Albert.
Approvals
In governed mode there are two:
- The plan. The goal, the files it expects to change, the steps, the test plan, the risk and how to roll back. Approve it, ask for changes, or reject it.
- The result. The diff, the test and check results, and the reviews. Accept it, ask for changes, or discard it. Acceptance is refused if a file it changes was edited since the task started.
Accepted changes are ordinary edits in your working tree: review them with git diff, commit them,
and undo them with Git.
Everyday commands
| Command | |
|---|---|
/connect, /model |
Connect a provider, choose a model; albertcode models test checks it (Models) |
/plan <request>, /do <request> |
Start a governed or a direct change |
/mode [governed|direct|auto|ask] |
How plain requests are handled |
/new |
A fresh conversation |
/attach <paths> |
Give it files or screenshots |
/tasks, /ledger |
Recent tasks; the evidence ledger (Tasks and evidence) |
/commands, /agents, /plugins, /tools |
Your extensions (Extending) |
/mcp, /connectors |
MCP servers (MCP servers) |
/contract |
The repository's boundary (The contract) |
/help |
Every command |
The terminal reference lists all of them, with examples. In the browser, Ctrl/⌘ K opens the command palette.
The pages of the browser interface
| Page | What it is for |
|---|---|
| Architecture | What the repository is made of, read from its files |
| Requirements | What the app must do, as a checklist with evidence |
| Preview and browser tests | Run the app beside the chat, edit it by clicking, and test it in a browser |
| Tasks and the evidence ledger | Every task, and the record of what each did |
| Files and Terminal | The repository's files, and a real shell, beside the chat |
Scripting
Every step is also a subcommand, so AlbertCode fits a script or CI job:
TASK=$(albertcode create "Add tests for the billing module" --workspace ./repo | jq -r .id)
albertcode show "$TASK" # the task, as JSON
albertcode approve "$TASK" # approve the plan
albertcode diff "$TASK" # the change
albertcode accept "$TASK" # apply italbertcode reject discards a task, and albertcode --help lists every subcommand. See
Terminal.
Attachments
Paste or drop screenshots, logs and documents into the browser or VS Code, or use /attach in the
terminal. Images reach the model as images, if the model accepts them. If it doesn't, AlbertCode
tells you. See Attachments.
Checking your setup
albertcode doctorThis shows which build tools AlbertCode can see, and where it found them.
From docs/usage.md in huynvic/albertcode-swe-agent at a8cef99 (15 Oct 2026), shown unchanged except for links. AlbertCode SWE Agent documentation, CC BY 4.0. Latest on GitHub.