DocuPanda

DocuPanda turns documents into structured data. Agents parse files, classify them, extract fields against a JSON schema you define, route results to a human reviewer when needed, and react when a job finishes.

1 trigger 41 actions API key available

DocuPanda website Platform Documentation Capability data verified 2026-08-13

On Event reports that a submitted claim packet has finished parsing

Classify Documents sorts the packet into estimate, police report and medical bill

Extract Data runs the schema built for each class against the parsed document

Query Extractions confirms the claim number and policy number agree across every document in the packet

Create Review renders the values that disagree against the pages they came from

Agent posts the review summary and the recipient it proposes to the claims desk

The claims lead names who may open the review and how long the link lives before Get Review Link is called

What This Integration Enables

DocuPanda is built for teams whose documents all say the same thing and none of them look alike. Rather than a template per layout, you write a schema once and give it guidelines. Those guidelines are instructions to the model, not documentation for a colleague: how to read a date written two ways, whether an amount includes tax, what to do when a field simply is not there. Classes work the same way, since a class is a name plus the description the model classifies against. Create Schema takes a schema you wrote, Generate Schema has DocuPanda write one from documents you already uploaded, and Get Schema Proposals shows what it would suggest without creating anything.

The second thing that separates it is that review is a real object, not a status flag. Create Review renders each extracted value against the page it came from, Get Review Link produces a link a person can open without a DocuPanda account, and Update Review records the verdict and any corrections the reviewer made. That matters to FlowRunner because it means the human-in-the-loop moment leaves evidence on both sides: the flow knows a person decided, and the platform knows what they were looking at when they did. Everything here is asynchronous, so On Event is the spine rather than an optimization, and the agent reacts to work finishing instead of holding a connection open waiting for it.

Without FlowRunner

Packets read in the order they arrive A person opens every attachment to find out what it is before deciding what to do with it

A rule per form layout Documents that say the same thing in different layouts each need their own extraction rules

Review by forwarded attachment Getting an outside reviewer to check one value means emailing them the whole file

With FlowRunner

Packets sorted before they are read Classification runs first, so each document meets the schema built for its type

Written guidance instead of templates A schema carries instructions on how to read ambiguous dates, missing fields and amounts that may or may not include tax

Review is an object with a verdict The reviewer sees each value against its page, and their corrections are recorded against the extraction

Use Case Scenarios

A claim packet that sorts itself

A first notice of loss arrives with seven attachments in Gmail. The agent calls Submit Document on each, preferably by URL so the file never has to travel through the flow as base64, and waits for On Event rather than polling. Classify Documents sorts them into repair estimate, police report, medical bill and photographs. Extract Data then runs against each document with the schema built for its class, so the estimate is read as an estimate and the medical bill as a medical bill. Query Extractions checks that the claim number and policy number agree across the whole packet, and the reconciled figures post to the claims record with a summary in Slack. The adjuster opens a packet that is already sorted, extracted and cross-checked.
Asking the file room a question

The reserve review needs every open claim where the estimate exceeds the policy limit, and that answer is inside documents rather than in any field. Query Extractions searches inside extracted data, so the agent finds documents by what they say rather than by what they are named. The matching claims land in Google Sheets with the extracted amount and the source document alongside each row, and Download Extraction produces the Excel or PDF version for whoever needs a copy. When a new document type shows up that no schema covers, Get Schema Proposals tells the team what DocuPanda would suggest before anyone commits to a schema, which keeps the schema list from growing one panicked addition at a time.
A verdict that gets recorded, not just given

Some values are not the agent's to settle: a diagnosis code, a liability split, a date of loss that two documents disagree about. Create Review builds a visual review showing each of those values against the page it was read from, and the agent hands it to the person who should decide. When they answer, Update Review records their verdict and their corrections against the extraction itself, so the file carries not just the final number but who chose it. Only then do the reconciled values move on to the claims system and to Airtable for reporting. This is where FlowRunner's automation exceptions stop being a queue of failures and start being a short list of judgment calls.

Human-in-Loop Highlight

Get Review Link creates a link to a review that anyone can open without a DocuPanda account, and it takes an expiry in hours. Leave the expiry empty and the link never expires. That is genuinely useful, because the people who most need to check an extracted value are frequently outside the workspace: an independent adjuster, a body shop, a treating provider's billing office. It is also the one call on this connector that puts a claim document and everything read out of it behind a permanent unauthenticated URL, and nothing in the action list revokes a link once it is out. Deleting the review with Delete Reviews closes the door and destroys the record of what the reviewer was shown, which is the opposite of what a claims file wants. So the agent builds the review, assembles the packet, drafts the message, and stops before the link exists. It posts to the claims desk: "Review ready for claim 88-421022, three disputed values. Proposed recipient is the independent adjuster at Bellwether, proposed expiry 72 hours. Confirm the recipient and the expiry, or route this internally instead." A person answers, and then Get Review Link runs. The agent did all the assembly. It did not decide who gets a standing key to a claim file.

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

41 actions

Documents

Submit Document Sends a document to be parsed and returns a job identifier immediately. Give it a URL rather than the bytes when you can, so the file is fetched directly instead of travelling through the flow.
Get Document Reads a parsed document with the text and structure DocuPanda extracted from it.
List Documents Lists documents newest first, with filters for dataset and creation date.
Search Documents Finds documents by file name or identifier, for the cases where you already know what you are looking for.
Get Document Count Counts the documents in the workspace, optionally within a date range. Useful for volume reporting and for sanity checks on a batch.
List Datasets Lists the dataset names in use in the workspace, so a flow can route into an existing dataset rather than inventing one.
Download Document Downloads a document and stores it in file storage. Choose the original or the searchable version produced by recognition.
Split Document Splits one document into several and lets the model decide where the boundaries are. The move for a batch scan whose record boundaries are not known in advance.
Merge Documents Joins several documents into one, in the order you list them.
Update Document Dataset Moves documents into a dataset. Used to reorganize a workspace as processing tiers or claim types change.
Update Document Metadata Sets the metadata on documents. This replaces what is already there rather than merging, so read the current values first if any of them still matter.
Delete Documents Deletes documents and everything extracted from them. Reserved for retention workflows, and worth an approval step of its own.

Extraction

Extract Data Pulls structured data out of a parsed document using a schema, and returns a job identifier. Leaving the schema empty makes DocuPanda invent one for that document and bills for it, so point it at a schema you made.
Get Extraction Reads an extraction result as JSON.
List Extractions Lists extraction results, with filters for schema and document.
Download Extraction Downloads one extraction as a PDF, XML, plain text or Excel file and stores it. Used when the output is going to a person rather than to another system.
Query Extractions Searches inside extracted data, so documents can be found by what they say rather than by what they are called. The operation behind corpus-wide questions.
Delete Extractions Deletes extraction results and leaves the documents they came from untouched. Used when a schema change makes older results misleading.

Schemas

List Schemas Lists the schemas in the workspace.
Get Schema Reads one schema with the JSON schema it holds, so a flow can be built against the exact field names it will receive.
Create Schema Creates a schema from a JSON schema you write, describing what to pull out of a document.
Generate Schema Asks DocuPanda to write a schema by looking at documents already uploaded. Run it once and reuse the result rather than regenerating per document.
Update Schema Renames a schema or changes its description and guidelines. The guidelines are instructions to the model, so this is where extraction behavior actually gets tuned.
Delete Schema Deletes a schema. Extractions already made with it stay where they are.
Get Schema Proposals Reads the schemas DocuPanda suggests for a document without creating any of them. The safe way to look at a new document type before committing.

Classification

List Classes Lists the classes documents can be sorted into.
Create Class Creates a class. Its description is what the model reads to decide what belongs in it, so the description is the classifier.
Update Class Renames a class or rewrites the description it classifies against. Changing the description changes future routing, which is worth reviewing before it ships.
Delete Class Deletes a class. Documents already sorted into it keep their other classes.
Classify Documents Sorts documents into classes and returns a job identifier. Run it before extraction so each document meets the schema built for its type.

Reviews

Create Review Builds a visual review of an extraction, showing each extracted value against the page it came from. The step that turns a disputed number into something a person can actually check.
Get Review Reads a review, either by its own identifier or by the extraction it belongs to.
List Reviews Lists reviews in the workspace, for tracking what is still waiting on a person.
Update Review Records a person's verdict on a review, and any corrections they made. This is what makes the human decision part of the file rather than part of a chat thread.
Get Review Link Creates a link to a review that a person can open without a DocuPanda account. An empty expiry produces a link that never expires, which is why this page's human gate sits here.
Delete Reviews Deletes reviews and leaves the extractions they were built from untouched. It also removes the record of what a reviewer was shown, so use it deliberately.

Jobs

Get Job Reads a job. Everything in this API is asynchronous, so this is the polling answer when the trigger is not in use.
List Jobs Lists jobs with filters for status, type, document and date. Used to reconcile what a batch actually completed.
List Workflows Lists the workflows set up in the workspace, so a flow can address one by name rather than by a copied identifier.
Run Workflow Runs a workflow over documents and returns a job identifier. Used when the multi-step processing is already designed inside DocuPanda.
Get Account Reads the workspace the key belongs to and its credit balance. Doubles as a connection check and as the guard before a large batch.

Triggers

1 triggers

Event Triggers

On Event Fires when DocuPanda finishes a piece of work: a document parsed, data extracted, a classification run, a split or merge completed, a workflow finished, or a person recording their verdict on a review. Events come in success and failure pairs, so a flow can react to what failed as carefully as to what succeeded, and choosing no events at all receives everything. Because every processing operation in this API is asynchronous, this is the trigger that lets a workflow wait on real completion instead of a timer.

Frequently Asked Questions

What can FlowRunner do with DocuPanda?

FlowRunner agents can run Submit Document, Get Document, and List Documents in DocuPanda, plus 38 more actions.

Does connecting DocuPanda to FlowRunner require OAuth?

No. DocuPanda connects to FlowRunner with an API key, no OAuth flow required.

Can DocuPanda trigger a FlowRunner workflow automatically?

Yes. DocuPanda supports 1 trigger that can start a FlowRunner workflow automatically.

Start building with DocuPanda

Free plan, no card required. Connect in minutes.

Start Building Free Talk to us