AlissaOne task, one branch, one PR
The agent works on a branch named after the task and opens a pull request you can read, diff and reject. The moment the PR opens, its link is attached to the task as evidence, so the task page and the repo always agree.
✦ Free · Works with Claude Code
Your coding agent is brilliant. Give it a first day.
Direction · Memory · Accountability
Alissa Code is a small command-line tool. You type alissa code instead of launching your agent bare, and it starts already onboarded: signed in to your Alissa account, carrying your team's written conventions, bound to the right repos, and registered as a session you can follow in Alissa Studio. The agent itself does not change. What changes is that its work now lands somewhere you can find it later.
Free · macOS and Linux · Claude Code today
Free
to use
to use
Works with
Claude Code
Claude Code
60-second
install
install
Tasks are
the record
the record
One session
or a fleet
or a fleet
The problem
A great agent with no badge.
Launch a coding agent bare and you get a brilliant new hire on their first day with no badge, no handbook and no HR record. It writes good code, and then the session closes, and everything about how the work happened closes with it: which task it was for, what it decided along the way, what it checked before calling it done. The next session starts from zero, and so does your team's memory of the last one.
Alissa Code fixes the first day, not the agent. Before your agent starts, one command does what good onboarding does for a person: it checks who they are, hands them the house rules, seats them in the right place and writes their name on the board.
What happens when you run it
One command, four checks, then your agent.
Here is exactly what alissa code does before it hands off, in the order it does it. Each check can fix itself with your permission, so the first run takes a minute and every run after that takes a second.
The four checks
Badge, handbook, desk, name on the board.
- Token · who is this?It confirms the terminal is signed in to your Alissa account, so everything the agent records is attributed to you. You sign in once with
alissa auth login; the token stays on your machine. - Skills · what are the house rules?A skill is a written procedure the agent reads before it acts, such as your git rules or your review protocol. Alissa Code installs the skills your workspace requires from skills.alissa.app and keeps them current, so every agent follows the same conventions.
- Workspace · where does it sit?An Alissa Code Workspace is a directory that binds one or more repos, each laid out so that one branch is one folder and the main checkout stays clean. The session is bound to it, so the agent starts in the right place with the layout explained.
- Session · whose name goes on the board?It registers the run as a session in Alissa Studio, with a name. From that moment you can see it on your Sessions page, send it a message while it works, and read its transcript afterwards. Then it launches Claude Code with your arguments untouched.
What the agent can do now
Things a bare launch cannot.
- Take the next task from your queue, the list of committed tasks lined up for your sessions, and make it the session's focus.
- Write evidence and the pull request link onto the task, so the work is recorded where the task lives, not only in the terminal.
- Keep notes in session memory that survive the agent's own context window, so a long job does not forget its own plan.
- Receive a message from you mid-run and acknowledge it, without you interrupting the terminal.
- Hand a task to another session, or pick one up that another session left behind.
Where the work lands
Tasks are the record.
Everything in Alissa starts from a task: one piece of work with a definition of done and the criteria to check it. When your agent works through Alissa Code, it anchors to a task the same way a person on your team does, and that is what makes its work legible to everyone else.
The task
The session
A card you can open
Every run has a session card in Alissa Studio: its name, the task it is focused on, the messages you exchanged and the full transcript. Six months later you can still answer "why was this done this way?"
The review
A second pair of eyes, on request
With the review skill installed, a fresh session with none of the first agent's assumptions can read the pull request and post its findings on it, round after round, before the work reaches you.
Because the record lives on the task and not in the terminal, you can close the laptop and nothing about the work is lost.
✦ alissa tmux
One session or many
From one session to a fleet, without five tabs.
Once a session has a name, you can have several. Alissa Code runs agents inside managed terminal sessions you can list, message, attach to and revive from any shell. Each one has a queue of work to pull from, so you manage a shift board instead of babysitting windows.
- Start a named session with one command and hand it a task
- Attach when you want to watch, detach when you don't
- Revive a dead session from its registration after a restart
The next step
When you would rather not be at the keyboard at all.
Alissa Code is how one person runs one or several agents deliberately. If you want your backlog worked through the night with nobody watching, that is Alissa Dark Factory: three autonomous seats that plan, build and review your GitHub issues through Alissa Code sessions on your own machine or cloud, while you only stand at the gates and merge what is good. It is a paid activation on the Ultra plan; Alissa Code itself stays free.
Visit the Dark Factory →
Get started
Install in 60 seconds.
Three commands. The first installs the CLI without needing npm, the second signs you in once, and the third is the one you will use from then on instead of launching your agent bare.
- Install the CLIOne command, no npm required. If you prefer npm,
npm i -g @alissa/cliworks too. - Sign in oncePaste your Alissa token when asked. It is stored on your machine and never leaves it except to talk to Alissa.
- Hand off
alissa coderuns the four checks and launches Claude Code, onboarded. Anything after the command is passed through to the agent.
New to Alissa? Create your free account at studio.alissa.app. It is where your tasks, sessions and transcripts live, and Alissa Code needs nothing more than the free plan.
$ curl -fsSL https://share.alissa.app/install | bash$ alissa auth login # paste your alissa_… token
$ alissa code # four checks → hand off to Claude Code
From the blog
The thinking behind it, in order.
We wrote the system up as we built it. If you want the reasoning and not only the commands, this is the series, in the order it was meant to be read.
Start here
Introducing Alissa Code: Your AI Coworker's First Day
A bare agent launch is a brilliant intern with no badge, no handbook, and no HR record. `alissa code` onboards the agent instead.
Read the post › The CLIThe Most Durable Productivity Tool Is a Terminal
Productivity systems die one alt-tab away from the work, so we moved Alissa into the terminal. Install in 60 seconds.
Read the post › WorkspacesGive Your Coding Agent a Campus, Not a Cubicle
How forty hidden worktrees taught us that environments must explain themselves: manifest, skill, CLI, and a governed launch.
Read the post › FleetIntroducing Alissa TMUX: Management, Not Babysitting
Five agents in five tabs makes you a babysitter. Alissa TMUX gives the fleet a shift board: sessions, queues, and a boring reconciler.
Read the post › OrchestrationWhen the Agent Becomes the Manager
When a coding agent should orchestrate other agent sessions instead of native sub-agents: the durability rubric and six patterns.
Read the post › ThesisDon't Reinvent the Wheel (Invent the Meta-Wheel)
Why we orchestrate unmodified agent harnesses from the outside instead of building our own: the non-intrusive meta-harness thesis.
Read the post › CapstoneA Match Made in Clouds: Alissa & Claude
Claude Code is the workhorse (capability), Alissa brings the direction (strategy and accountability). Neither is enough alone.
Read the post ›
Questions
Before you install.
Does it change Claude Code?
No. Alissa Code runs its checks and then launches the Claude Code you already have, with your arguments passed through untouched. The one thing it adds is a session header in your project's MCP configuration, so the agent's Alissa tools know which session they belong to; you can turn that off with a flag.
Does it cost anything?
No. The CLI, sessions, tasks, skills and transcripts are free on every Alissa plan, including the free Starter plan. Credits are spent only when you ask Alissa's own intelligence to do something, such as answering in chat or running one of its tools, and Alissa Code never does that on its own.
Which agents does it support?
The hand-off target today is Claude Code. Managed terminal sessions can also host other agents or a plain shell, and because Alissa never modifies the agent it launches, adding another hand-off target is a small step rather than a rewrite.
Do I need Alissa Studio open while the agent works?
No. The session registers itself and keeps reporting in the background. Studio is where you read the record afterwards, send a message to a running session, or check what is queued; you can leave it closed while the work happens.
Can it run without me?
A single session still needs someone to start it and to merge what it produces. Running your backlog unattended, with autonomous seats that plan, build and review while you only merge, is what Alissa Dark Factory adds on top of Alissa Code.
Capability from your agent. Direction from Alissa.
Alignment, with deadly precision.
Install in 60 seconds →Free · macOS and Linux · Claude Code today