Turn messy input into typed data.
AI Ingest is the extraction layer from GamePlanLab — screenshots, PDFs, CSV, and plain text into schema-validated structured data with one call. You bring a Zod schema; you get back data that matches it, or a clear error. Instant access to the private repo the moment you buy.
One call. Typed, validated data.
The parts of AI extraction that are fiddly to get right — the prompt, the JSON parsing, the validation, the retry — done once and tested.
One call, typed result
extract(input, schema) — hand it an input and a Zod schema, get back data that matches, fully typed.
Any input
Screenshots and photos, PDFs, CSV, or plain text — same call, same clean output.
Schema to prompt, automatically
You write Zod. It describes the shape to the model for you — no hand-written "return JSON like…".
Self-repairing
If the model's answer doesn't validate, it's shown its own error and asked once more before giving up.
A client that won't fall over
Retry, backoff, and a circuit breaker built in — no SDK, no surprise failures on a rough API day.
Provider-agnostic
The model call is one injectable function, so you can swap models or test with no network at all.
Pulled from a shipped app.
This is the extraction pattern running in GamePlanLab, a coaching app that's live on the App Store — it turns opponent stat screenshots into structured data. AI Ingest is that pattern, cleaned of anything baseball, so you can point it at receipts, resumes, invoices, forms, or whatever you're building.
↗ See GamePlanLab on the App Store- Ships with a Next.js example + React dropzone
- Three ready-made schemas to copy
- Source-available, one-time price
- Bring your own model — Claude by default
Stop hand-parsing messy input.
One purchase, source-available, yours to build on — instant access to the private GitHub repo when you buy. All sales final.