
Veryfi
Thinking Mode: AI agents that check your extracted data
114 followers
Thinking Mode: AI agents that check your extracted data
114 followers
Today we’re introducing Thinking Mode, our new agentic data extraction for receipts and invoices. Add one parameter to your API request, and a team of AI agents reviews the extraction, catches mistakes and fixes them before the results reach you. Pick light or deep per request, or leave it off and still get results in under three seconds.
This is the 2nd launch from Veryfi. View more

Veryfi Thinking Mode
Launched this week
Thinking Mode is Veryfi's agentic data extraction. Add the thinking parameter to a document request, and AI agents review the extraction, catch mistakes and correct them before the results come back. It currently works with receipts and invoices.

Launch Team

ZethosThe ledger you and your AI work from
Promoted

Hi everyone! Marianna here, Head of Marketing at Veryfi.
Every team that processes documents at scale ends up building the same thing: review queues, validation rules and exception workflows to catch the errors extraction misses. We wanted to put that second look right into the API.
Thinking Mode adds an agentic review loop to our real-time pipeline. With one parameter:
Agents read the extraction side by side with the original document
They check every field and test the math. Do the line items add up to the subtotal, and the subtotal plus tax to the total?
When something’s off, they go back to the page and fix it
Every fix gets rechecked before the result comes back
You choose per request: light, deep or off. Real-time stays the default, so your live flows don’t slow down.
The most helpful feedback would be your trickiest invoices and receipts and how Thinking Mode handled them.
Thanks for checking it out!
The recheck loop is the part that caught my eye — most extraction tools hand you a confident answer and you only find the math error later. Checking that line items add to the subtotal and the subtotal plus tax adds to the total is exactly the kind of thing I'd want a second pass for. When the agent's fix is uncertain, like a total that's cut off, does it flag it for a person instead of guessing?
How does Thinking Mode deal with a crumpled receipt where half the total is cut off? Does it flag it for a person? Having agents re-check fields before results come back makes a lot of sense.