About this starter
Wrenfield handles the plumbing for AI products that read documents: invoices, contracts, forms. Uploads go to S3, Celery workers run OCR and then a language model with a JSON schema per document type, and results land in a review queue where people can correct fields before export. It includes teams, per-page Stripe billing, webhooks for customers and a CSV and API export.
The subscription covers new extractors and model updates as providers change. It is opinionated that a human review step belongs in every extraction product, and the queue UI is the best part of the kit. The limitation: OCR runs on a hosted service by default, and moving to self-hosted OCR for privacy-sensitive customers is documented but slow on CPU-only servers.
Highlights
- 01Review queue to correct fields before export
- 02JSON schema per document type
- 03Customer webhooks when a document is ready