Skip to content
AlbertCode SWE

mini-AlbertCode is a second agent on the AlbertCode engine. Our main agent is AlbertCode SWE Agent, which adds a VS Code panel and documents the repository contract, the evidence ledger, requirements, architecture and browser tests. Compare them

v0.1.11 — latest release on GitHubWindows · macOS · LinuxRuns locally

mini-AlbertCode SWE Agent

One coding agent, in your terminal and your browser.

mini-AlbertCode carries the AlbertCode engine — planning, the isolated copy, the checks, the reviews and the approval gates — and puts two interfaces on it: a full-screen terminal and a browser workspace, over one local service, so a session is live in both at once.

Or plan a first run in the run planner.

The mini-AlbertCode browser workspace at the second gate: two files changed, checks 2 of 2 passed, the security and quality reviews approved, and buttons to reject, discard, ask for changes, view the diff or accept it.The mini-AlbertCode browser workspace at the second gate: two files changed, checks 2 of 2 passed, the security and quality reviews approved, and buttons to reject, discard, ask for changes, view the diff or accept it.
Gate 2 — checks passed, reviews approved, waiting for you.

Overview

What it does for you.

  • Two interfaces, one live session

    The terminal and the browser show the same sessions at the same moment. An answer given in one is seen at once in the other.

  • Approval built in

    Two decision points, at the plan and at the result, each answered with a key — the same key in both interfaces.

  • Proven, not just written

    The AlbertCode engine runs your project's own tests and checks, and reviews the change for security and quality.

  • Four modes

    Governed, Direct · ask each write, Direct · auto-approve, and Ask only. Tab cycles them in the terminal.

  • Any model

    OpenRouter, OpenAI, Anthropic, Kimi, Hugging Face, any OpenAI-compatible service, or a local model server.

  • Private by default

    Runs on your machine, with no account and no telemetry. The service needs this installation's access token.

In the product

Captures of the mini-AlbertCode SWE Agent interface from its README, following one task in a demo project. Each is shown in your current theme.

A new session in the mini-AlbertCode browser workspace: the navigator with the project and its sessions, a box with Governed, Direct and Ask only, the provider and the model, and tiles for sessions to pick up, the two human gates and ways to start.A new session in the mini-AlbertCode browser workspace: the navigator with the project and its sessions, a box with Governed, Direct and Ask only, the provider and the model, and tiles for sessions to pick up, the two human gates and ways to start.
A new session opens on a launchpad, with the two human gates spelled out.

Technical capabilities

Interfaces, modes and commands.

Interfaces

  • Terminalalbert-mini

    A full-screen conversation where you already work. --mini keeps it inline in the lower part of the terminal.

  • Browseralbert-mini web

    Reading plans and diffs side by side; every project's sessions in one list; Git and worktrees from the menus.

Modes

  • GoverneddefaultChanges you care about

    You approve a plan; it works in an isolated copy, runs your tests and reviews the result; you approve the diff.

  • Direct · ask each writeQuick changes you are watching

    Works in your files, and every write waits for your OK. Git is the undo.

  • Direct · auto-approveSmall, safe changes

    Works in your files and carries the change out without asking. Git is the undo.

  • Ask onlyUnderstanding a codebase

    Answers, and offers a change; nothing starts until you say go.

Everyday commands

/connect
Connect a model provider: its key, or the address of a local model
/model
Choose a model
/plan <request>
Start a governed change
/do <request>
Change your files directly, asking before each write
/sessions
Open, rename, pin or delete sessions
/diff
This session's change, or your working tree's
/web
Open this session in the browser
/help
Every command

From a script or CI

albert-mini run "what does this project do?"
albert-mini run "add a health check endpoint" --approve plan,result
albert-mini run "add a test for cart totals" --approve all --json
Exit codes of albert-mini run
Exit codeMeaning
0Done: answered, or the change was applied
1Failed, or the change was rejected
2Refused: a wrong option, or nothing that could run it
3Stopped at a gate --approve did not cover: answer it in either interface
124Timed out

Source: docs/commands.md#run

Architecture

One local service. Everything else is a view of it.

mini-AlbertCode is one service on your machine, listening on 127.0.0.1:8778. It starts by itself the first time an interface needs it and keeps running, so closing a window never stops a task. albert-mini status says whether it is running; albert-mini stop stops it.

Both interfaces show the same sessions at the same moment. Sessions belong to the service, not to a window: rename, pin, fork, stop, archive, export or import them from either interface, and albert-mini -c continues the latest one in a folder.

It carries the AlbertCode engine and keeps its own command, port, folders, keys and sessions, so it can be installed beside AlbertCode SWE Agent with nothing shared. To use a service running elsewhere, albert-mini attach <url> opens the terminal on it; a service started with serve --host requires authentication.

From docs/how-it-works.md and docs/commands.md.

Your computer
TerminalFull-screen conversation, gates and lists
Browser workspacePlans, diffs, files and your running app
Local service on 127.0.0.1Starts on first use and keeps running between windows. Answers only requests carrying this installation's access token.
Isolated copy of your repositoryWhere the change is built, your tests run and the reviews read it.
Your working treeChanges land only when you accept. Commit or undo them with Git.
The model you connectA hosted provider with your key, or a local model server — then nothing leaves your machine. No telemetry, no account.
Drawn from the agents' documentation (How it works, Security model). Both run only on your machine; neither is a hosted service.

Gates

Every answer has a key — the same key in both interfaces.

A gate is where a change waits for you. An answer given in the terminal is seen at once in the browser, and a task is never answered twice.

The plan

  • A approve and build
  • E ask for changes
  • D read the whole plan
  • R reject

The result

  • A accept into your files
  • V view the diff
  • E ask for changes
  • X discard
  • R reject

A write (Direct · ask each write)

  • A allow once
  • Shift A allow the rest of the task
  • N decline

A change offered (Ask only)

  • P plan it first
  • O do it directly

Side by side

How it compares with AlbertCode SWE Agent.

Facts from each repository. Where mini-AlbertCode's documentation is silent, the table says so instead of guessing. Neither is claimed to be faster, cheaper or lighter. They install side by side, with nothing shared.

AlbertCode SWE Agent compared with mini-AlbertCode SWE Agent
FeatureAlbertCode SWE Agentmini-AlbertCode SWE Agent
Getting it
Commandalbertcodealbert-mini
Latest documented release1.54.1 (15 Oct 2026)0.1.11
InstallOne-line installer, or a downloaded installer per OSOne-line installer, or a downloaded installer per OS
UpdateRun the install command againalbert-mini update (or run the install command again)
GitRecommended; Direct mode needs itNeeded; installed after asking if missing
macOSA current macOS, Apple silicon or IntelmacOS 11 or later, Apple silicon or Intel
CostNo charge for model usage: your provider account, or a free local modelNo charge for model usage: your provider account, or a free local model
Working with it
InterfacesTerminal, browser, VS CodeTerminal, browser
Shared live sessionsTerminal and browser share sessionsTerminal and browser show the same session at the same moment
ModesGoverned, Direct (ask each write or auto), Ask onlyGoverned, Direct · ask each write, Direct · auto-approve, Ask only
Two human gatesPlan and diffPlan and result
Isolated copy, your tests, security and quality reviewsYesYes — it carries the AlbertCode engine
One request from a scriptalbertcode run, plus create / approve / diff / acceptalbert-mini run, with documented exit codes
A service run elsewherealbertcode --url <service> --api-key <key>albert-mini attach <url>; albert-mini serve --host
Documented features
Repository contract (albertcode.contract.toml)YesNot in its public docs
Hash-chained evidence ledger and compliance exportYesNot in its public docs
Requirements and architecture pagesYesNot in its public docs
App preview beside the chatYes, with edit-by-clickingYes
Browser tests (Playwright)YesNot in its public docs
MCP serversYes — connectors gallery, 88 as of 1.53.0Yes — 105 connectors ready to add as of 0.1.11
Custom commands and agents (Markdown)Yes, plus plugins, saved tools and skillsYes
Published benchmarkSWE-bench Pro HARD-51: 39/51, self-reportedNone published

From each repository's README, docs and changelog. “Not in its public docs” means the documentation doesn't say — not that the feature is absent.

Install

Prerequisites and installation.

The installer checks your machine, shows its plan and asks before it changes anything. It installs for your user only, in an environment of its own.

Install mini-AlbertCode SWE Agent

for your user · asks before changing anything
Terminal
curl -fsSL https://raw.githubusercontent.com/huynvic/mini-albertcode-swe-agent/main/installer/install.sh | sh
  • Open Terminal (⌘ Space, then type Terminal), paste the command, read the plan, then type Y.
  • Open a new Terminal window when it finishes, then run albert-mini --version.

Then check it

Open a new terminal first, so it picks up the command.

albert-mini --version

Prints its version and the AlbertCode engine version it carries.

albert-mini doctor

Checks the installation and the service, and says what to do about any problem.

albert-mini models test

One tool-calling request to the chosen model.

Update

albert-mini update

Reads the latest release, checks its installer against the release's SHA-256 and runs it. Your sessions are always kept.

Uninstall

albert-mini uninstall

Asks first. --all also removes settings, saved keys, sessions and history; --check only shows the plan. Your repositories are never touched.

Prerequisites

Windows
Windows 10 or 11, x64 (the x64 build also runs on ARM PCs). PowerShell 5.1 or 7.
macOS
macOS 11 or later, on Apple silicon or Intel.
Linux
x86-64 with glibc: Ubuntu 20.04+, Debian 10+, Fedora, RHEL 8+, openSUSE, Arch. Not Alpine or other musl systems.
Disk
About 1 GB free.
Network
pypi.org and github.com while installing; your model provider while working.
Git
Needed. The installer installs it if it is missing, after asking.
Python
Not needed. The installer uses your Python 3.12 if present, otherwise downloads one for mini-AlbertCode alone.

Prefer a download?

The installers aren't code-signed yet, so macOS Gatekeeper and Windows SmartScreen warn the first time; the one-line command avoids both. Check a download against the release's SHA256SUMS.

Latest release
v0.1.11
Published
13 Oct 2026
Repository updated
13 Oct 2026
Command
albert-mini