Rossum

Connect AI agents to Rossum intelligent document processing. Agents upload documents, wait for extraction, confirm or reject annotations, export queues, and manage schemas, workspaces, and hooks across 41 actions.

41 actions Session available

Rossum website ↗ Platform Documentation ↗ Capability data verified 2026-07-27

A vendor invoice PDF arrives in the accounts payable mailbox

Upload Document sends the file to the invoices queue with source metadata attached

Wait for Document Extraction polls until the AI Engine returns the annotation in a reviewable state

Get Annotation Content reads the invoice number, totals, and line items with per-field confidence scores

The agent matches the extracted total and PO reference against the open purchase order

An AP specialist reviews every annotation with a low-confidence field or a failed PO match before anything is confirmed

Confirm Annotation validates the clean and the approved documents, firing the export hooks toward the ERP

The finance channel gets the day's digest of confirmed, held, and rejected documents

What This Integration Enables

Rossum's design premise matches FlowRunner's: extraction is a lifecycle, not an OCR call. Every document becomes an annotation that moves through states, importing, to review, confirmed, exported, and the platform tells you the confidence of every extracted field. That gives an orchestration agent something most document tools never provide: an honest signal about when it should not proceed alone. FlowRunner agents run the full lifecycle across 41 actions: upload documents from files or URLs, wait for extraction, read the datapoint tree, confirm or reject, export queues in bulk, and manage the configuration surface of queues, schemas, workspaces, and hooks. - Upload-and-extract in one branch: Upload Document plus Wait for Document Extraction returns structured fields inside 90 seconds of polling - Confidence-routed review: Get Annotation Content exposes per-field confidence, the routing signal for [human-in-the-loop](/concepts/human-in-the-loop) gates - Bulk movement: Export Queue delivers confirmed data as JSON, XML, CSV, or XLSX for reconciliation and archival - Configuration as workflow: agents create queues, bind schemas, and wire hooks when a new entity or document type onboards

Without FlowRunner

Keying invoices is a job description Someone reads PDFs and types numbers into the ERP, and accuracy depends on their afternoon

Every document gets equal attention The clean invoice and the ambiguous one cost the same review minutes

Extraction results die in a silo Data gets extracted, then someone still has to move it where the ledger lives

With FlowRunner

Documents flow through states Upload, extract, review, confirm, export: each document's position is queryable at any moment

Attention follows confidence Clean extractions confirm and export on their own; humans see only the fields the AI was unsure about

Confirmation is the integration Confirming an annotation fires the hooks that carry the data into accounting, no re-keying step left

Use Case Scenarios

AP intake that ends in the ledger, not a folder

Invoices arrive by email and land in FlowRunner as files. The agent runs Upload Document into the AP queue, Wait for Document Extraction, then Get Annotation Content for vendor, invoice number, date, totals, and line items. Extractions where every field clears the confidence bar and the PO matches are confirmed and flow through export hooks; the parsed fields also post to [QuickBooks](/integrations/quickbooks-online) as a draft bill, or to [Invoice Ninja](/integrations/invoice-ninja) for teams that run there. Everything else queues for a person with the uncertain fields highlighted. The AP team stops typing and starts adjudicating.

The reconciliation file that builds itself

Every evening the agent calls Export Queue on the day's confirmed annotations in JSON, reshapes the datapoint tree into rows, and appends them to the reconciliation workbook in [Google Sheets](/integrations/google-sheets): invoice number, vendor, total, currency, queue, and confirmation timestamp. A companion pass over List Annotations counts documents by status, and the morning digest in [Slack](/integrations/slack) reports what confirmed, what is waiting on review, and what got rejected with reasons. Month-end reconciliation starts from a complete file instead of a shoebox.

Onboarding a new entity without a services engagement

The company acquires a subsidiary with its own vendors and document formats. The agent runs Create Workspace for the entity, Create Schema with the field tree finance signed off on, Create Queue binding the two, and Create Hook so every export event notifies the integration that posts to the subsidiary's ledger. Get Queue hands back the queue's dedicated inbox address, which goes to the subsidiary's AP contacts. When the schema needs a new field a month later, Update Schema adds it without touching anything else. The destructive end of this surface, Delete Queue, Delete Schema, Delete Workspace, stays behind human confirmation, because each permanently unwinds configuration that live documents depend on.

Human-in-Loop Highlight

Confirm Annotation is the exact point where extracted numbers become financial facts: confirmation advances the annotation toward export and fires the configured hooks, which is how a total on a PDF becomes a payable in the ERP. Rossum itself models this as a human click, and FlowRunner keeps it that way wherever the data is not certain. The agent confirms only annotations where every material field, total, currency, vendor, PO reference, clears the confidence threshold and the PO match. Everything else lands in front of an AP specialist with the doubtful fields named: "Total extracted as 12,480.00 at 61 percent confidence; PO 4471 expects 12,840.00." The specialist corrects and confirms, or runs Reject Annotation, which can notify the sender, another message that reaches a real vendor and is worth a deliberate click. A wrong payment recalled costs a week; a review costs a minute.

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

Document Processing

Upload Document Uploads a PDF, image, or office file from a FlowRunner file or URL to a queue for extraction, with optional pre-filled values and metadata. Returns an asynchronous task.
Import Document to Queue Uploads through the legacy endpoint that returns the annotation reference synchronously, for flows that need the annotation URL immediately.
Wait for Document Extraction Polls an annotation until extraction finishes and it reaches a reviewable or terminal state, then returns it with extracted content. The second half of every upload-and-extract branch.

Queues

List Queues Lists processing queues, optionally by workspace. Queues bind a schema to a document stream.
Get Queue Retrieves a queue with its schema, hooks, per-status document counts, and dedicated inbox email address.
Create Queue Creates a queue bound to a workspace and schema, with automation settings.
Update Queue Patches a queue's name, schema, automation, or connectors.
Delete Queue Permanently deletes a queue and detaches its documents. Cannot be undone; human-gated.
Export Queue Exports a queue's annotations as JSON, XML, CSV, or XLSX, filterable by status or specific IDs. The bulk outlet for confirmed data.

Annotations

List Annotations Lists annotations across the organization, filterable by queue, status, and document, with optional sideloaded documents.
Get Annotation Retrieves one annotation's status, queue, document link, and timestamps.
Get Annotation Content Retrieves the extracted tree of sections and datapoints with values and confidence. The primary read for structured document data.
Update Annotation Patches an annotation's status, assignee, or metadata.
Start Annotation Moves an annotation into reviewing and locks it, the required step before programmatic edits.
Confirm Annotation Validates an annotation and advances it toward export, firing export hooks. The action this page's human gate exists for.
Reject Annotation Rejects an unprocessable or duplicate document, optionally notifying the sender.
Delete Annotation Moves an annotation to deleted and removes its document from processing. Cannot be undone.

Documents

List Documents Lists uploaded document objects and their metadata.
Get Document Retrieves one document's file name, MIME type, download URL, and linked annotations.
Create Document Creates a document from a file or URL without queueing it for extraction, for staging.
Get Document Content Downloads the original file as Base64 with filename and MIME type, for archiving or re-attachment.

Schemas

List Schemas Lists extraction schemas, the field trees that define what gets extracted.
Get Schema Retrieves a schema's full content tree, the map for reading annotation content.
Create Schema Creates a schema from a tree of sections and datapoints.
Update Schema Patches a schema's name or content; changes affect future extractions on bound queues.
Delete Schema Permanently deletes a schema not bound to any queue. Cannot be undone.

Workspaces

List Workspaces Lists workspaces, the containers that group queues by department or entity.
Get Workspace Retrieves a workspace with its queues.
Create Workspace Creates a workspace for a new department or entity.
Update Workspace Patches a workspace's name or properties.
Delete Workspace Permanently deletes a workspace and its queues. Cannot be undone; human-gated.

Hooks

List Hooks Lists webhook and serverless-function extensions configured on annotation lifecycle events.
Get Hook Retrieves a hook's type, attached queues, events, and configuration.
Create Hook Creates a webhook or function hook that fires on document lifecycle events, the wiring for downstream integration.
Update Hook Patches a hook's events, config, active flag, or queues.
Delete Hook Permanently deletes a hook. Cannot be undone.

Users and Organization

List Users Lists users filterable by username, for review assignment.
Get User Retrieves a user's role, groups, and queue assignments.
Get Current User Retrieves the authenticated account's profile, organization, and queues, the discovery call for new deployments.
List Organizations Lists organizations accessible to the authenticated user.
Get Organization Retrieves an organization's name, plan, settings, and usage limits.

Frequently Asked Questions

What can FlowRunner do with Rossum?

FlowRunner agents can run Upload Document, Import Document to Queue, and Wait for Document Extraction in Rossum, plus 38 more actions.

Does connecting Rossum to FlowRunner require OAuth?

No. Rossum connects to FlowRunner with session-based authentication, no OAuth flow required.

Can Rossum trigger a FlowRunner workflow automatically?

Rossum doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with Rossum

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

Start Building Free Talk to us