A software team made of AI agents
Not one assistant doing everything: 10 specialists, each with one job and its own instructions, working on a board you can see — and asking you before anything that matters.
Why a team, not one assistant
Each agent does the work it is made for, and the work is checked before it reaches you.
A plan before any code
The Product Owner turns your description into a board of epics, stories and tickets, and asks you what it needs to know. You approve the plan before the team builds it.
The right specialist for each ticket
The architect sets the project up, the designer draws, developers build, QA tests and DevOps deploys — each with its own instructions and tools.
Work that is checked
QA checks every ticket against its acceptance criteria, and writes and runs the tests, browser tests included. A security reviewer reads risky changes.
No changes that collide
One specialist works in the code at a time, so one agent's change never overwrites another's.
The right model for each ticket
Each ticket is rated light, standard or deep: simple work runs on a lighter model, hard work gets more reasoning.
Watch any of them work
The dashboard shows which agents are at work, on what and for how long, and each one's session is a click away.
The team
One role per job. A Website project has one agent of its own instead, which builds the site with you in one conversation.
Product Owner
Understands your goal, plans the board and asks you what it needs to know. Never writes app code.
Architect
Decides which apps the project needs and sets up the TypeScript monorepo they live in.
Designer
Draws your app's screens as static mockups, so you approve a look before any interface is built.
Developer
Builds features end to end: interface, API and database.
Junior developer
Turns an approved design into the app's interface components.
QA
Checks each piece of work against its acceptance criteria, and writes and runs the tests.
Security reviewer
Reads risky changes and reports what it finds. It never edits code.
DevOps
Declares what your project deploys and runs each deploy once a person has pressed Deploy.
Explore
Searches the code to answer the other agents' questions. Read-only.
Project memory
Reads every commit, the agents' and yours, and records what each feature now does. Read-only.
One dial for how capable the agents are
Pick a level for each project. Higher levels use more capable models and more reasoning, so each task costs more. Smart and Intelligent come with every plan; Brilliant and Genius with Pro and Max.
Smart
Our fastest, cheapest models. Good for simple apps and quick iteration.
IntelligentDefault
A step up in reasoning. Best balance of quality and cost.
Brilliant
Stronger reasoning again. For complex architecture and harder bugs.
Genius
Our most capable models at maximum reasoning. For the hardest work.
How the team works through a project
The order a software team follows, with you at the points that matter.
You describe the app
In plain words in the wizard, with any documents you have. The agents read them before they plan.
The Product Owner plans
It writes the board and asks its questions. You approve a build plan in one click.
The architect and the designer prepare
The architect sets up the TypeScript monorepo; the designer draws the screens, and you approve a look.
Developers build, QA tests
Ticket by ticket, each checked against its acceptance criteria before it counts as done.
DevOps releases
It prepares the release. A person presses Deploy, and production waits for an owner.
Good to know
- The agents are AI, not people: they can be wrong, and you review what they produce.
- They cannot deploy without a person pressing Deploy, reach production without an owner's grant, or read the keys stored for test and production.
Ready to build?
Describe your first app. The agents take it from there, and production waits for you.