Mindee

Extract structured data from invoices, receipts, IDs, and custom document types with the Mindee V2 API. Runs OCR and parsing jobs and returns clean fields ready for downstream systems.

4 actions available

Receipt image submitted for expense reimbursement

Agent sends the receipt to Mindee for extraction

Agent reads the merchant, date, amount, and tax fields

Agent checks the confidence on each field and the total math

Agent matches the expense against the employee policy

Agent creates the reimbursement record

A low-confidence field or an out-of-policy amount routes to finance

What This Integration Enables

Mindee reads documents into clean, structured fields. Prebuilt models handle invoices, receipts, and IDs out of the box, and custom models parse formats specific to your business. For asynchronous work it manages extraction jobs, and every field comes back with a confidence value. Extraction is only trustworthy when you know which fields to trust. Mindee's per-field confidence is what lets a flow post the clean ones automatically and hold the uncertain ones for a person. An orchestration layer draws that line and routes the exceptions, and FlowRunner is built for that layer, so document intake scales without writing a misread number into your books.

Without FlowRunner

Receipts keyed by hand Employees or finance type every field from an image

No confidence signal A misread amount enters the system unnoticed

Policy checked manually Someone compares each expense against the rules

With FlowRunner

Fields extracted automatically Merchant, date, amount, and tax pulled from the image

Confidence per field Uncertain fields flagged before they post

Policy checked in the flow Out-of-policy amounts routed to finance

Use Case Scenarios

Expense Receipt Processing

An employee submits a receipt for reimbursement. The agent extracts the merchant, date, amount, and tax, checks the confidence on each field and the total math, and matches the expense against policy. Clean, in-policy expenses create a reimbursement record automatically. Low-confidence fields or out-of-policy amounts route to finance. Finance handles exceptions instead of keying every receipt.

Invoice Intake

Vendor invoices arrive in mixed formats. The agent parses each with the appropriate Mindee model, extracts the header and line items, and validates the totals. Matched, high-confidence invoices post automatically; anything uncertain or mismatched is held for AP, so the ledger only ever gets verified data.

Custom Document Extraction

A business-specific document type needs parsing that no prebuilt model covers. The agent uses a custom Mindee model to extract the fields defined for that format and maps them to a record. Confident extractions flow through; the edge cases route to the person who owns that document, so a new format is automated without losing accuracy.

Human-in-Loop Highlight

Document extraction fails quietly: a confident-looking wrong value is worse than an obvious error. Mindee returns confidence per field, and FlowRunner uses it to draw the line. When a field falls below the threshold, or an extracted amount breaks policy or fails a math check, the agent routes the document through a [human-in-the-loop](/concepts/human-in-the-loop/) step: it pauses, shows the person the original image beside the extracted value, and waits. They correct or confirm. Clean documents post themselves; the uncertain ones get a human read first.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

4 actions

Extraction

Extract Document Runs a document through a Mindee V2 extraction model and waits for the result. Provide the model ID (a UUID you create in the Mindee platform from the model catalog, e.g. the prebuilt Invoice, Receipt, Passport, ID, Resume, or US Driver License model, or your own custom model) and a document URL.
Enqueue Inference Enqueues a document for extraction without waiting for the result, returning the created job (with its id and polling/result URLs).

Jobs

Get Job Status Retrieves the current status of an asynchronous inference job by its id. Returns the job object, whose `status` is one of Processing, Processed, or Failed; once processed, `result_url` points to the completed inference.
Get Inference Result Fetches a completed extraction inference by its id and returns a flattened `fields` object (simple values, nested objects, and lists) alongside the complete raw inference.

Start building with Mindee

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us