Build a recipe manager web app in ~10 minutes — full walkthrough
Real-time recipe for your first end-to-end factory build. Same path the Recipe Manager Web case study took (first verified prod pass on 2026-05-18). No credit card, no local toolchain — just the operator and your keyboard.
Before you start
- ·Modern browser (Chrome 100+, Safari 15+, Firefox 100+)
- ·A free OOretz Factory account — sign up at /factory/signup
Steps
- 01~1 min
Open the launchpad and pick a starter prompt
Visit /factory/launchpad. Pick any prompt with the "Web" badge — these are the surface we verify end-to-end today (desktop and mobile are roadmap). The operator opens with the prompt pre-filled in the interview chat.
- 02~2 min
Answer 2-3 interview questions
The AI will ask follow-ups to refine your manifest. Default answers are fine if you just want a demo. Common questions: which entity fields do you need, are there workflows you want to expose, what's the primary list view.
💡 TipType "default" or "all" for any question you don't care about. The AI will pick sensible defaults.
- 03~1 min
Approve the plan
You'll see an AppManifest review screen with the entities, fields, and workflows the AI extracted. Scroll through to verify. Click "Approve & build" when ready.
- 04~7 min
Watch the build stream
The operator now shows live build progress over SSE. Phase indicators flip from blue → green: Scaffold → Entities → Workflows → Build → Verify → Package. The right side shows the latest event line including AI call cost and token usage.
🪤 GotchaWeb builds run on the Linux bridge. If /factory/status shows it offline, the job queues. Desktop and mobile bridges are roadmap — picking a non-web prompt today produces code but does not complete the launch + smoke step.
- 05~1 min
Inspect the smoke result + download the workspace
When the build finishes, the Delivery panel renders. You see: functional smoke pass/fail (POST + GET round-trip against the running app), AI cost summary, errors if any, and a download link to the workspace .tar.gz. Click the link to get a 200+ line README with quick-start, file tree, deployment recipes, and a working curl smoke test.
✓ Expected outcome
You have a downloadable Next.js + libSQL web app with full CRUD per entity, real API routes, and a server-verified POST + GET round-trip on the running app. Open the tar.gz, run `npm install && npm run dev`, and the same smoke that gated the pass on prod runs locally. Typical AI cost: $0.05-$0.40 depending on model and entity count.
Troubleshooting
Build hangs at "scaffold" for more than 2 minutes+
Check /factory/status. If the Linux bridge is offline, the job queues. You can cancel from the operator and re-launch later. Desktop / mobile prompts will not complete end-to-end — those pipelines are roadmap.
Compile errors after the repair cap+
The factory cap is 6 build attempts with a 15-cycle total repair budget (lib/factory/verification/verification-loop.ts). If you hit it, the build log shows which file failed. Most cases are prompt ambiguity — sharpen the prompt and re-launch.
Verification fails — functional smoke didn't pass+
Open /factory/proof and find your job. The functionalSmoke.diagnostic field shows what the POST + GET round-trip returned. Common causes: missing required fields on the first entity, table-name collision, or @libsql/client install failure. The error message is the literal stderr from the bridge.