Taggun

Read receipts and invoices with Taggun OCR and get back structured totals, tax, dates, merchant details, and line items, each with a confidence score. Agents post the confident ones and route the rest to a person.

10 actions API key available

Taggun website ↗ Platform Documentation ↗ Capability data verified 2026-08-01

A claimant submits a rebate form with a photographed receipt attached

Agent calls Validate Receipt against the campaign's stored rule set

Agent reads the pass or fail flag together with the list of rules that passed and failed

Agent separates rule failures about the purchase from rule failures about the image itself

Agent approves and records the payout in [Airtable](/integrations/airtable) where every rule passed

Claimant is told the outcome and, on a rejection, which purchase rule the receipt missed

Any failure on a fraud rule goes to a reviewer with the image before anything is said to the claimant

What This Integration Enables

Taggun does two jobs and it is worth being clear that they are different. The first is receipt OCR: give it an image or a PDF and it returns totals, tax, discounts, dates, merchant name and address, payment type and product line items, each wrapped with a confidence score and the source text it was read from. The verbose form gives you everything; the simple form gives you total, tax, date and merchant with none of the envelope, which is the right choice when a flow is reconciling amounts and nothing else.

The second job is campaign validation, and it is what separates Taggun from a general document parser. You store a rule set under a campaign ID covering a purchase date window, allowed merchants, required products with quantities and price ranges, expected product codes, a balance-owing range, free-form questions answered by a model, and four fraud checks for similarity, tampering, digitally generated images and handwriting. Validate Receipt then scans a submission and returns a pass or fail flag with the specific rules that passed and failed. Everything is synchronous, so a submission is judged inside the request rather than in a queue. What FlowRunner adds is the discipline of treating those two kinds of failure differently, because a receipt that missed a date window and a receipt Taggun thinks was generated by software are not the same event and must not produce the same message.

Without FlowRunner

Every submission is read by hand Someone opens each photograph and checks it against the campaign rules

Rules live in a document The campaign terms are prose that each reviewer interprets slightly differently

Suspicion has no process A reviewer who thinks a receipt looks wrong has nowhere to put that except an email

With FlowRunner

Rules are stored and applied The campaign rule set is configuration Taggun checks every submission against

Clean submissions clear themselves A receipt passing every rule is approved without a person reading the image

Suspicion is routed, not acted on A fraud rule failure becomes a reviewed decision, never an automatic rejection

Use Case Scenarios

A rebate campaign that judges itself

Before the submission window opens, the agent stores the campaign rules with Create Campaign Settings: the eligible purchase dates, the participating retailers, the required product with a minimum quantity, and the fraud checks the programme wants enabled. Submissions arrive with a photographed receipt. The agent calls Validate Receipt against the campaign ID and reads the result. A receipt passing every rule is approved, recorded in Airtable, and the claimant is told immediately. A receipt failing a purchase rule gets a specific and checkable reason back: the purchase date falls outside the window, or the required product is not on the receipt.

Expense claims reconciled on totals alone

Employees post expense photographs into a channel. The agent runs Scan Receipt Simple, which returns total, tax, date and merchant details with confidence scores and none of the verbose payload, because the reconciliation only needs the amounts. It matches each receipt against the corresponding card transaction, writes the matched pair into Expensify, and leaves unmatched receipts in a short list that a person clears at the end of the week. Where a claim needs categorising rather than just matching, the agent switches to Scan Receipt and uses the product line items rather than guessing from the merchant name.

Files that are already in the flow

A receipt arrives as an email attachment behind authentication that Taggun could not fetch itself, and Taggun only accepts HTTPS URLs, rejecting HTTP, localhost and private-network targets. The agent uses Scan Receipt From Base64 instead, supplying the bytes it already holds along with the filename and content type so Taggun picks the right decoder. Nothing has to be published to a public URL first, which keeps a claimant's receipt out of any location that did not need to hold it.

Human-in-Loop Highlight

Four of Taggun's campaign rules are fraud checks: similarity against other submissions, signs of tampering, whether the image looks digitally generated, and handwriting. When one of those fails, Taggun has not found an error in a document. It has produced a suspicion about a named person, and the response to it lands on that person's account. An automatic rejection on a tampering flag tells a claimant, in effect, that they submitted a fake receipt. The ones that are wrong are wrong in the worst possible way: a legitimate customer whose phone camera produced an odd artefact, or an employee whose receipt was reprinted at the till.

So a FlowRunner agent splits the failure list in two. Rule failures about the purchase are factual and go straight back to the claimant, because "this receipt is dated the second of March and the campaign ran from the tenth" is checkable by the person who receives it. Rule failures on any fraud check stop the flow before a single word reaches the claimant. The agent sends the reviewer the receipt image, the extracted fields with their confidence scores, exactly which fraud rule failed, and the other submissions Taggun matched it against on a similarity flag, then asks: "Taggun flagged this submission as a possible duplicate of claim 3391 from the same account. The merchant and total match, the timestamps are eleven minutes apart. Reject, approve, or ask the claimant for the original?" A person answers, and the claimant hears an outcome that somebody stands behind. This is the digital andon cord at the point where automation stops being a productivity question and starts being about how you treat people.

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

10 actions

Extraction

Scan Receipt Reads a receipt or invoice hosted at an HTTPS URL and returns the full verbose extraction: totals, tax, discounts, dates, merchant name and address, payment type and product line items, each wrapped with a confidence score and the source text it was read from. Runs synchronously with nothing to poll. Only HTTPS URLs are accepted, and HTTP, localhost and private-network targets are rejected.
Scan Receipt From Base64 Reads a base64-encoded receipt and returns the same full verbose extraction, without the document needing to be hosted anywhere. The operation for files already held in the flow, or sources behind authentication that cannot be forwarded. Filename and content type are both required so Taggun picks the right decoder.
Scan Receipt Simple Reads a receipt at an HTTPS URL and returns only the core fields: total, tax, date, and merchant name, address, city, state, postcode, country and types, each with a confidence score. The right choice over the verbose form when a flow is reconciling amounts, because there are no line items or feature envelopes to parse.
Scan Receipt Simple From Base64 Reads a base64-encoded receipt and returns only the core fields. The lightest option available: no hosting step and no verbose payload. Filename and content type are required.
Validate Receipt Scans a receipt at an HTTPS URL and checks it against the rules stored under a campaign: date window, allowed merchants, required products and quantities, product codes, balance owing, free-form questions answered by a model, and the fraud checks enabled on the campaign. Returns a single pass or fail flag plus the list of rules that passed and failed, synchronously in the response.

Campaign Validation

Create Campaign Settings Stores the rule set Validate Receipt checks submissions against, under a campaign ID you choose. Every section is optional, so a campaign carries only the checks it needs. Rules cover a purchase date window, an allowed merchant list, required product names, quantities and price ranges, expected product codes, a balance-owing range, free-form questions, and the four fraud checks for similarity, tampering, digitally generated images and handwriting.
Update Campaign Settings Replaces a campaign's stored rule set wholesale, so any section left out is removed rather than preserved. The safe pattern is to fetch the current rules, change what needs changing, and submit the whole thing back, which is why an agent that edits live campaign rules reads before it writes.
Get Campaign Settings Returns the validation rules currently stored for a campaign. Empty rule sections may be omitted from the response, so the result is treated as the authoritative current state and round-tripped rather than reconstructed by hand.
List Campaigns Returns the campaign IDs with validation settings stored on the account. Agents use it to audit which campaigns are live before a submission window opens, then read each one's rules individually.
Delete Campaign Settings Removes a campaign's validation rules. Validation calls can no longer evaluate them afterwards, and Taggun does not document the exact failure those calls return, so this belongs after a submission window has fully closed rather than during one.

Frequently Asked Questions

What can FlowRunner do with Taggun?

FlowRunner agents can run Scan Receipt, Scan Receipt From Base64, and Scan Receipt Simple in Taggun, plus 7 more actions.

Does connecting Taggun to FlowRunner require OAuth?

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

Can Taggun trigger a FlowRunner workflow automatically?

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

Start building with Taggun

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

Start Building Free Talk to us