Skip to Content

Your first app

≈ 30 min · you will have a “Purchase requests” app running in the dashboard preview: staff create purchase requests, a manager approves or rejects them.

You write a brief, review the spec the agent writes, watch the preview and ask for changes. The example is a real template an agent built on manh.run from spec to deploy.

Not connected yet? Do Connect your agent first. It takes about 2 minutes.

Steps

Write a brief for the agent

Say who uses the app, what they do and what information it keeps. You don’t need to name libraries or a file layout, because the agent reads the manh.run framework docs.

Use manh.run to build a "Purchase requests" app. Staff create a purchase request (title, description, amount, category, needed by) and follow its status. Staff can edit or cancel a request while it is pending. A manager sees an approval inbox and approves or rejects — a rejection needs a reason. Every step is recorded on a timeline. The manager sees the approved total per month.

Review the spec before any code

The agent follows the manh.run spec-first process: it writes PRD.md, ORCA.md (roles, objects, permissions) and spec/01–04 (data, lifecycles, API, screens) in your working folder. Then it asks you once about roles, permissions, status flows, the screen list and anything your brief left open.

Answer those questions. The code only translates the spec, so a change here costs less than a change later.

The agent writes code and checks it

The agent creates a project on manh.run and writes files into app/ one phase at a time: tables and business rules, then the API, then the pages. manh.run checks every write right away and returns the result to the agent; at the end of each phase the agent runs a full check with type checking.

The app/ tree of the Purchase requests template:

app/ ├── auth.ts → roles, sign-up rule (sign-in and users are generated by manh.run) ├── layout.tsx → app frame: menu per role, sign-out ├── page.tsx → page / — pending work and the monthly summary ├── login/page.tsx → page /login ├── register/page.tsx → page /register ├── shared/request.ts → request statuses and lifecycle └── requests/ ├── requests.table.ts → requests table ├── request-events.table.ts → the timeline of each request ├── repository.ts → lifecycle, permissions, history ├── page.tsx → page /requests — the list ├── get.ts · post.ts → GET · POST /api/requests ├── stats/get.ts → GET /api/requests/stats └── [id:int]/ ├── page.tsx → page /requests/12 — detail ├── get.ts · patch.ts → GET · PATCH /api/requests/12 ├── approve/post.ts → POST /api/requests/12/approve ├── reject/post.ts → POST /api/requests/12/reject └── cancel/post.ts → POST /api/requests/12/cancel

Watch the live preview

Open manh.run  and select the project. The project’s Overview page has a preview that runs the latest draft; it refreshes a few seconds after each file the agent writes. Screens that call the API show data after the first deploy. The agent deploys once the API phase is done so it can test it.

Ask for changes

Tell the agent what to change in the same conversation, for example: “Add a department column to the request list.”

To point at an exact spot on the screen, click Pick element on the preview, click the part you want changed, then click Copy prompt. The copied prompt names the file and line; paste it into the agent and add what you want changed.

Good to know

  • The spec lives in the folder your agent runs in, on your machine. The manh.run project keeps only the app/ tree.
  • Writing files does not change the live app. The live app changes only when you deploy.
  • Demo data goes in through the API after deploy, never hard-coded in the code.

Next steps

Last updated