How it works.
You describe a goal and approve the steps. Under the hood, that simple loop is a real engineering setup: your own Claude, your own cloud, a team of specialist agents, and a harness that keeps them honest. Here's the whole thing, in plain terms.
It runs on your Claude — not our markup
Connect your own Claude and it does all the thinking. Your key lives in your environment, never with us. We orchestrate the work; you're never renting tokens at a premium or bracing for a surprise invoice.
- No metered API bills from us. You bring your own Claude; we don't charge per token.
- Your key, your control. It stays in your own environment — revoke it any time.
- Predictable cost. One flat subscription to us, plus your own Claude. That's it.
We orchestrate. Your Claude does the work. Your tokens, your account.
Project interview
It starts by getting to know your project
Before a single line of code, it interviews you about what you're building, who it's for, and what matters. Your answers shape everything that follows — the setup, the structure, the plan.
- The right setup, chosen for you. Your project's shape is picked from what you describe — not a one-size template.
- A working app from minute one. You start from a running starter, not an empty repo.
- A structured plan. Your goal becomes an ordered set of steps, each handed to the right specialist.
A team of specialists,
not one do-everything bot
Your goal is split across five distinct roles, each with its own brief and judgment — the way a real product team divides the work. One model, five different jobs.
Product owner
Breaks your goal into ordered steps and assigns each to the right specialist. It's also the team's harshest critic — it reviews everyone's work and won't call a thing done until it's proven.
Front-end
Builds the screens, flows and interactions — accessible and polished by default, with the UX details handled so you don't have to spell them out.
Back-end
Handles the data, the APIs and the server-side logic — the plumbing your features actually run on, designed to fit the steps before and after it.
DevOps
Takes care of the build, the wiring and the deploy steps — so what the others write doesn't just exist, it actually runs in your environment.
QA
Writes and runs real end-to-end tests on every feature, then reports a clear pass or fail. Nothing is trusted on a promise — only on a green test.
Every project lives in its own cloud — always on, always testable
Your project gets its own private, isolated environment in the cloud, and the app it builds is live from the start. The moment a feature lands, you can open it on a real URL and try it for yourself.
- Instantly testable. A live preview URL, updated as the work lands — no local setup.
- Isolated & yours. One project, one environment — never tangled with anyone else's.
- A safe blast radius. Approved commands run in your environment, not ours.
The part that makes it actually good
A chatbot in a box would be useless. Your Claude runs inside a real harness — the scaffolding that gives it memory, holds it to a standard, and proves the work before it ever reaches you.
Smart memory
Every agent remembers your project — the decisions, the code, the history of what you've approved. They recall the right context instead of starting cold every session.
Adversarial review
Work doesn't get a free pass. A critic agent challenges each step — poking holes, catching shortcuts and weak assumptions — before anything reaches your board.
E2E tests before ship
Every feature is exercised end-to-end automatically. A pass/fail summary lands with the work, so "done" means tested — not just written.
Project knowledge base
Your goals, constraints and notes stay on hand for every agent, so the work lines up with what you actually want — not a generic best guess.
Real tools, not guesses
Agents can read your data, browse, and run your project — grounding their work in what's really there instead of hallucinating how it might behave.
One conversation, start to finish
Planning, building and your feedback all share a single thread — so a comment changes the plan in context, and nothing gets lost in a handoff.
It can't wander off and quietly burn your tokens
Left unchecked, agents drift — chasing tangents, looping, and running up a bill with nothing to show. The harness budgets each agent's context and tokens so it stays locked on the task in front of it.
- No runaway loops. Budgets and limits stop an agent before it spirals.
- Focused context. Each agent gets just what it needs for its step — not the entire history dumped in.
- Your tokens, respected. Less waste means lower cost on your own Claude.
Token budget · this step
Your team. Your Claude.
Your call.
Describe a goal, approve the steps, and watch a real dev team move your project forward — with you on the trigger the entire time.