Airparser
Document ProcessingExtract structured data from emails, PDFs, scans, and spreadsheets using Airparser's large language model parsers. Agents send a document to an inbox, receive typed fields back, and act on parsed results as soon as they land.
What This Integration Enables
Airparser is built for teams whose documents arrive because somebody forwarded them. Every inbox owns an import address of the form [email protected], and a document that reaches it is parsed the same way whether it was emailed by a supplier, dropped in by a person, or pushed through the API by an agent. That single detail changes the shape of the workflow. You are not building an upload pipeline and then asking humans to use it. You are letting people keep the habit they already have and putting an agent on the other side of it.
The other half of Airparser is the extraction schema, and it is the real unit of configuration. Airparser generates the first draft from the first document you send an inbox, then you tune it into the exact field list your accounting system expects, including repeating line items and nested address objects. FlowRunner connectors give agents the whole surface: read the schema, upload against it, collect the typed result, and clone a tuned schema across the inboxes that handle the same document type. What agents do not get is unilateral authority over that schema, because it is the thing every downstream field depends on.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Supplier invoices that arrive as forwards
AP publishes the Airparser import address to its suppliers and to the internal shared mailbox rule. An invoice lands. On Document Parsed fires with the typed payload: vendor, invoice number, currency, tax breakdown, totals, and the line items as a repeating list. The agent checks that every required field is present, queries Acumatica for a bill already carrying that vendor and invoice number, then creates the bill and attaches the source document. The AP channel in Slack gets a summary with a link back to the Airparser document, so anyone reviewing later can see the original alongside what was read from it. Nobody changed how they receive invoices; the queue behind them changed.
Digital and scanned intake in the same building
A new inbox is created for warehouse paperwork. Because these documents are photographed delivery notes rather than digital PDFs, the inbox is created with the Vision engine, which analyses the rendered page image instead of the text layer. A second inbox for emailed order confirmations uses the Text engine. The agent routes each source to the inbox whose engine suits it, collects the parsed output, and writes both into the same Google Sheets reconciliation tab. When the delivery note inbox starts returning empty carton counts, the agent pulls Get Document Details with Use Original Parsed Output enabled to see whether the model or the post-processing step is at fault.
Promoting a tuned schema to production
Someone spends an afternoon on a test inbox getting a supplier's odd credit-note layout to extract correctly. The tuned schema now needs to reach the production inbox that receives that supplier's live traffic. The agent reads both schemas, produces a field-by-field diff, and posts it. Clone Extraction Schema overwrites the destination wholesale, so any field the production inbox has that the test inbox lacks disappears the moment the call is made, and every document parsed afterwards comes back missing it. The agent does not make that call on its own.
Human-in-Loop Highlight
Clone Extraction Schema and Update Extraction Schema both replace the destination schema in full rather than merging into it. On a live inbox that suppliers are actively emailing, that is a change with no warning shot: documents keep parsing, the trigger keeps firing, and the fields you quietly dropped simply arrive empty from that point forward. So the agent treats the rollout as a human-in-the-loop decision. It reads Get Inbox on both sides, posts the diff to the AP lead in Slack, and asks in plain terms: "Cloning the test schema onto the live supplier inbox will remove po_number and cost_center from every future extraction, and add credit_note_reference. Roll it out now, roll it out after month-end close, or hold?" The lead answers, the agent proceeds or waits, and either way the decision has an owner. This is the digital andon cord applied to configuration rather than to a single document: the agent can see the blast radius, and it stops the line before it pulls the lever.
Agent Capabilities
10 actionsInboxes
4- List Inboxes Returns every Airparser inbox available to the key, with the identifier the upload, listing and schema operations need. Agents use it to discover the routing table at the start of a flow rather than hard-coding inbox IDs.
- Get Inbox Retrieves one inbox with its parsing configuration and the extraction schema currently applied to it. This is how an agent inspects which fields an inbox will actually return before it sends anything into it.
- Create Inbox Creates an inbox and returns the import email address documents can be forwarded to. The engine choice matters: Text reads the extracted text layer and suits emails, spreadsheets and digital PDFs, while Vision analyses the rendered page and suits scans and photographs. No schema is required up front, because Airparser generates one from the first document.
- Delete Inbox Permanently deletes an inbox together with its schema and every document it holds, and the import address stops accepting mail immediately. Irreversible, so agents confirm the identifier before calling it and generally leave this one to a person.
Documents
4- Parse Document Downloads a file from a URL, including a FlowRunner file URL, and imports it into an inbox for extraction, returning the structured fields. With Wait For Result enabled the call is synchronous and waits roughly a minute; if parsing outruns that window the response carries the document id so the result can be collected afterwards. Disabling it returns immediately and allows ZIP archives.
- List Documents Returns a paginated list of the documents in an inbox, filterable by date range, name search and processing status. This is the usual way an agent finds what failed to parse or is still in flight.
- Get Document Retrieves a document with its extracted data in the compact shape Airparser sends to webhooks, including the credits the parse consumed. Used to collect an asynchronous result.
- Get Document Details Retrieves the extended view: owning inbox, raw text, format, page count, image preview and neighbouring document ids. Enabling Use Original Parsed Output returns the extraction exactly as the model produced it, bypassing inbox post-processing, which is what an agent reaches for when a field looks wrong and the cause is unclear.
Extraction Schema
2- Update Extraction Schema Creates or replaces an inbox's extraction schema, defining exactly which fields every document yields. Supports scalars, repeating lists such as invoice line items, nested objects such as an address, and fixed-choice fields. The whole schema is replaced on each call, so the complete field list must be sent every time.
- Clone Extraction Schema Copies one inbox's schema onto another, overwriting the destination. The intended use is promoting a schema tuned on a test inbox into production, or keeping several inboxes that handle the same document type in lockstep. Because it overwrites rather than merges, agents gate it on human approval.
Triggers
1 triggersEvent Triggers
1- On Document Parsed Fires when a document reaches one of the watched statuses in an inbox and emits it together with its extracted data. It fires regardless of how the document arrived, so a supplier forwarding an invoice, a colleague uploading one in the Airparser interface, and an agent importing one through Parse Document all produce the same event. Statuses defaults to Parsed; adding Failed turns the trigger into an [automation exceptions](/concepts/automation-exceptions/) feed that surfaces broken extractions instead of letting them settle in a queue. The first polling cycle records what is already there without firing, so existing documents are never replayed, and each cycle inspects the most recent fifty documents at an interval you choose down to thirty seconds.
Frequently Asked Questions
What can FlowRunner do with Airparser?
FlowRunner agents can run List Inboxes, Get Inbox, and Create Inbox in Airparser, plus 7 more actions.
Does connecting Airparser to FlowRunner require OAuth?
No. Airparser connects to FlowRunner with an API key, no OAuth flow required.
Can Airparser trigger a FlowRunner workflow automatically?
Yes. Airparser supports 1 trigger that can start a FlowRunner workflow automatically.
Start building with Airparser
$100 in credits. No card required. Connect in minutes.