canonical: https://jentic.com/apis/mindee.com/mindee

# Mindee OCR API

Jentic publishes the only available OpenAPI specification for Mindee OCR API, keeping it validated and agent-ready. Mindee is a document processing and OCR platform that extracts structured fields from invoices, receipts, identity documents, passports, and custom forms. The API exposes synchronous and asynchronous prediction endpoints, queue polling for long-running jobs, document feedback for retraining, and workflow execution for multi-step document pipelines. Predictions return parsed key-value pairs with confidence scores so agents can route documents without manual review.

## For AI agents

Submit a document image or PDF and receive structured field extractions for invoices, receipts, IDs, and custom forms. Supports synchronous predictions for fast paths and asynchronous queues for batch processing.

## Scope

Does not handle accounting ledgers, tax filing, or document storage - use for OCR field extraction from documents only.

## Capabilities

- Submit invoices and receipts to /v1/products/{account}/{product}/{version}/predict and receive parsed line items, totals, vendor, and tax fields
- Queue large document batches via predict_async and poll job status until extraction completes
- Execute multi-step document workflows with /v1/workflows/{workflow_id}/executions to chain classification, extraction, and validation
- Send corrected field values back through PUT /v1/documents/{document_id}/feedback to improve future model accuracy
- Discover available custom and off-the-shelf products with GET /v1/customers/self/products
- Retrieve a parsed document by its document_id to fetch all stored predictions and metadata

## Use cases

### Accounts Payable Invoice Capture

Automate invoice intake by submitting supplier PDFs to the Mindee invoice product and receiving structured supplier name, invoice number, line items, tax, and totals. Agents can post extracted fields directly into accounting systems without manual keying. Synchronous predictions return in seconds for low-volume flows; asynchronous queues handle thousands of invoices per hour.

Example prompt: Submit invoice.pdf to /v1/products/mindee/invoices/v4/predict and return supplier_name, invoice_number, total_amount, and line_items as JSON

### Receipt Expense Reporting

Extract merchant, date, total, and tax from receipt photos for expense management apps. The receipt product handles thermal prints, photos, and PDFs in over 60 languages, returning parsed fields with confidence scores so agents can flag low-confidence values for human review.

Example prompt: Send receipt.jpg to the receipt product predict endpoint and create an expense entry with merchant, date, and total_amount fields

### ID and Passport Verification

Capture name, document number, date of birth, and expiry from passports and national IDs as part of KYC onboarding. Mindee parses MRZ codes and visual zones in a single call, returning structured fields plus crop coordinates so agents can build verification flows without bespoke ML.

Example prompt: Submit passport.jpg to the passport product and verify that the parsed expiry_date is in the future and the surname matches the user record

### Agent-Driven Document Workflows

AI agents discover the Mindee API through Jentic by intent (parse a receipt, extract invoice fields). Jentic returns the predict endpoint with its input schema, the agent uploads the document, and parsed fields stream back. The Mindee bearer token stays in your Jentic One instance so the agent never sees the raw secret.

Example prompt: Search Jentic for 'extract invoice data', load the Mindee predict schema, and execute against three queued invoice PDFs

## Key endpoints

| Method | Path | Description |
| --- | --- | --- |
| POST | /v1/products/{account}/{product}/{version}/predict | Synchronous document prediction |
| POST | /v1/products/{account}/{product}/{version}/predict_async | Queue an asynchronous prediction |
| GET | /v1/products/{account}/{product}/{version}/documents/queue/{job_id} | Poll async job status |
| GET | /v1/products/{account}/{product}/{version}/documents/{document_id} | Retrieve a parsed document |
| PUT | /v1/documents/{document_id}/feedback | Send field-level correction feedback |
| POST | /v1/workflows/{workflow_id}/executions | Execute a document workflow |

## Key resources

- **Predictions** — Synchronous and asynchronous document parsing endpoints for invoices, receipts, IDs, and custom products
- **Documents** — Retrieve parsed documents by ID and submit field-level feedback for retraining
- **Workflows** — Execute multi-step document pipelines with classification, extraction, and validation
- **Products** — List the off-the-shelf and custom Mindee products available on the account

## Why Jentic

- **Setup:** Wiring Mindee by hand means handling its bearer auth against api.mindee.net and building the account, product, and version path segments for each predict call yourself, including the async queue polling. Through Jentic you install once, import Mindee from the API Directory, store the token once, and your agent calls it.
- **Permission scoping:** Mindee puts the account, product, version, and document id in the URL path (/v1/products/{account}/{product}/{version}/predict), so a rule can pin your agent to one product's predict operations. You choose the operations it may call, so a write like submitting feedback is not included unless you add it.
- **Credential handling:** Your Mindee bearer token is stored once, encrypted, by your own Jentic One instance and injected at execution time. It never enters the agent's prompt, logs, or context.
- **Discovery method:** Agents search Jentic by intent such as 'extract invoice data' or 'parse a receipt', and Jentic returns the matching Mindee predict operation with its input schema so the agent calls the right product slug without browsing Mindee documentation.

## Related APIs

- **Rossum** — Rossum focuses on transactional document AI for invoices and purchase orders with human-in-the-loop review built in
- **ABBYY** — ABBYY offers enterprise OCR and intelligent document processing across many document classes
- **Airtable** — Airtable stores parsed Mindee fields as structured records for downstream review and reporting

## FAQ

### Why is there no official OpenAPI spec for Mindee OCR API?

Mindee does not publish an OpenAPI specification. Jentic generates and maintains this spec so that AI agents and developers can call Mindee OCR API via structured tooling. It is validated against the live API and kept up to date. Get started with Jentic One, the self-hosted execution layer.

### What authentication does the Mindee OCR API use?

Mindee OCR API uses HTTP bearer token authentication. The token is sent in the Authorization header as `Bearer <token>`. When called through Jentic, the bearer token is stored in your Jentic One instance and injected at execution time so the agent never sees the raw secret.

### Can I extract line items from invoices with the Mindee OCR API?

Yes. Submit the invoice file to POST /v1/products/{account}/{product}/{version}/predict using the invoices product slug and Mindee returns parsed line_items, supplier_name, invoice_number, total_amount, and tax fields with per-field confidence scores.

### How do I handle large document batches without timing out?

Use POST /v1/products/{account}/{product}/{version}/predict_async to queue the document, then poll GET /v1/products/{account}/{product}/{version}/documents/queue/{job_id} until the job_id transitions to a completed state. The completed response contains the parsed prediction.

### How do I improve extraction accuracy on documents Mindee gets wrong?

Send corrected field values to PUT /v1/documents/{document_id}/feedback. Mindee uses these corrections to retrain the underlying models for custom products and surface the corrected values on subsequent predictions of the same document.

### How do I extract receipt data through Jentic?

Run `pip install jentic`, then search Jentic for 'parse a receipt'. Jentic returns the Mindee predict operation with its input schema. Load the schema, supply the receipt file as the document field, and execute. Parsed merchant, date, and total fields return as JSON.

### Can I limit what my agent is allowed to do with the Mindee OCR API?

Yes. Because you run Jentic One yourself, your own rules decide which Mindee operations and credentials the agent may use. Since Mindee encodes the account, product, and version in the URL path (/v1/products/{account}/{product}/{version}/predict), you can pin the agent to a single product's read-only predict and queue-polling operations. Writes such as submitting field feedback with PUT /v1/documents/{document_id}/feedback are not included unless you explicitly add them.