Prello
Ship the whole app.
Prello runs AI agents on your keys and your machines. Follow the line through the three pillars: Builder, Design and Code, and the loops that keep them honest. It does not let go until everything is green.
Describe it once.
Get a working app.
Builder turns one prompt into a scaffolded, gated, working product. It reads your repo before it writes, posts a plan you can veto, and pulls in Design and Code the moment it needs them.
The prototype is real.
The system drives it.
The Design lens is your built site, stitched and navigable page by page. One toggle flips to the design system behind it: tokens and components derived from your real schema.
Real diffs. Hard gates.
Honest spend.
A real IDE over the same files. Every iteration is a clean diff on a branch, judged by your gates, priced by the Pareto router.
They call each other.
Three lenses over one living project, not three tools. When one pillar needs another, it just calls it: same files, same plan, same gates.
Scouts go ahead. The orchestrator builds.
Sub-agents are read only. They fan out to map the code, research and trace causes, and each reports back a few pointers in a small footprint. The orchestrator does the real work, with its context intact.
Until it's actually done.
A failed gate is not a dead end. The agent reads the failure and goes again, and it only stops when every gate is green.
One prompt to a shipped app.
No cuts.
The ceiling becomes your provider.
Connect OpenRouter, Anthropic or any OpenAI compatible provider and pay them directly with no token markup. No keys? Credits work out of the box, on any plan.
Frontier models only for the hard calls.
You set a budget per provider and the router never crosses it. 349+ models reachable through one router, picked by cost against your gates.
Work lives in sessions.
A session runs on one device at a time: your Mac, the web app, or your phone. Host it in Prello Cloud and it keeps working while you step away, pushing every result onto your git project.
It asks before it guesses.
When a real decision comes up, Prello pings you where you already are. Tap an answer and the loop carries on. Nothing merges without you unless you allow it.
It only touches what it declares.
Every skill ships a manifest of exactly what it may reach: which network hosts, which secrets, which files, which commands. Anything not declared is denied.
Every number here,
checked.
Nothing on this page is a vibe. Each claim traces to how Prello is actually built and billed.
Pick how much you
want to run.
Five plans, one product. Every feature on every plan; the dial is weekly credits and cloud session storage. Bring your own keys on any plan with no token markup.
It ends on
your machine.
One account, two surfaces: the native macOS app and Prello Cloud. Your code and your keys stay yours, wherever you sign in.