# From one sentence to evidenced work, inside a budget

Ulpi turns a sentence into an editable company, runs work in an isolated sandbox with models handled for you, closes each task only on real evidence, and gives you one Home view to steer it — without spending past your cap.

## Four stages, end to end

This is the whole loop — describe, land, prove, steer. Nothing closes without evidence, and nothing spends past your cap.

### 1. Describe — Describe the company in one sentence

Write a single line. ulpi proposes an editable org chart, a goal with a measurable target, 30/60/90 milestones, and the Day-1 work. Adjust the roles, goals, and budget envelope, then hire the team in a click — nothing runs until you approve the plan.

You shape the company before any spend: edit the org chart, set the goal target, and fix the budget cap that will bound every run.

### 2. Home — Land on Home and watch your team get to work

After you hire, Home becomes the daily view of your company. Your teammates pick up Day-1 work and run it in an isolated sandbox, with models handled for you and no keys to manage. Home shows verified work, current spend, budget health, decisions waiting for you, and what each teammate just made.

Every run is metered and budget-capped: cost accrues as quantity times versioned pricing as a run proceeds, secrets and cost references never leak into logs or outputs, and a run that would cross your cap pauses mid-flight and surfaces for approval.

### 3. Evidence — Work closes only when it is proven

A task is done only when it produces a content-addressed WorkProduct you can open and passes a verification on every required outcome, under separation of duties. A teammate saying it finished is never enough.

Open the WorkProduct, see the passing verification on each outcome, and trust the closure — done is evidenced, not declared.

### 4. Chat — Steer your company in plain language

Open Chat and tell your company what to do next. Each turn resolves to a typed, reviewable change — or a clear refusal that tells you why — never vague prose, and every change is undoable. When you want the numbers, the Cockpit sub-view shows burn against the cap, cost per verified outcome, and read-only revenue from connected Stripe and analytics.

Every Goal to Work to Run to cost to log is threaded by one correlation ID, on an append-only, hash-chained activity ledger you can open from the Cockpit.

## This is what a closed task looks like

Every task produces a content-addressed WorkProduct and a passing check on every required outcome before it closes. Done is shown — not declared.

- **1 sentence** to a hired team
- **0 keys** to bring or manage
- **100% evidence** required to close

## What each part does

- **Build your company** — Turns one sentence into a proposed, editable org chart with a measurable goal, 30/60/90 milestones, and a Day-1 work set. You edit everything and set the budget cap before hiring.
- **Sandboxed runs** — Your teammates execute in an isolated sandbox — the models are handled for you, with no keys to bring or manage. Every run is metered and budget-capped, and secrets and cost references never leak into logs or outputs — references, never raw values, appear in envelopes and logs.
- **Evidence-gated closure** — Each required outcome needs a passing verification and a content-addressed, immutable WorkProduct before the task closes. Separation of duties keeps the doer and the verifier apart.
- **Budget envelope** — A hard cap with a visible auto-pause. Spend is metered as quantity times versioned pricing across every matching scope, and a breach pauses the run until you resume it.
- **Cockpit** — The deeper money view behind Home: cost per verified outcome, live burn, and read-only Stripe and analytics revenue, with a trust ledger and a correlation-ID drilldown into the underlying activity.
- **Morning Briefing** — A daily Shipped / Blocked / Needs-you / Money / Mentions digest. Approve or deny what needs you in one tap, resolving to granted, denied, expired, or conflict.
- **CEO Chat** — Plain-language steering that compiles to a typed operation or a 422 refusal with a reason — never free prose. Every operation is reviewable and undoable.
- **Honest bounds** — Ulpi is in its founding-customer phase, with every run metered, budget-capped, and evidence-gated, and no claim of unlimited autonomy. The platform makes its limits explicit rather than hiding them.

## Build a company the way you would actually run one

Describe it in a sentence, set a budget, and let an evidenced team of teammates get to work — with you steering from Home.

---

[Ulpi](/en) | [Pricing](/en/pricing) | [Examples](/en/examples)