80% accuracy on documents is functionally 0% when a human rechecks every result. Almost every important decision a business makes starts inside a document: a loan file, a patient record, an insurance claim, a contract. Most of that information is still read, checked, and typed in by hand. In this video, our CEO Dan Maloney explains: ▸ Why large language models still stumble on real documents ▸ Why 80% to 90% accuracy is functionally 0% when a person rechecks every result ▸ How the next generation of Agentic Document Extraction (ADE), powered by our new DPT-3 models, closes that gap ADE is autonomous and auditable. It does the tedious reading at scale, and every answer traces back to the exact spot on the page it came from. It recognizes layout, hierarchy, tables, signatures, figures, and handwritten notes before reasoning over the content, routing each block to the model best equipped to understand it. Deploy in our cloud, in your own VPC, on AWS, Azure, or Google Cloud, natively in Snowflake, or fully on-premises for air-gapped environments. Agentic Document Extraction is available today. Watch Dan walk through it below.

Onboarding one business client means reading 300 pages across a dozen documents. The hard part is agreement. Names, numbers and owners must match across all of them before the account opens. Then risk schedules a periodic review, and the file gets read again. Fenergo found only about a third of these reviews run automated. Agentic Document Extraction (ADE) turns every document into structured data, with no templates to build. A new form or issuer works on the first call. ADE Parse reads the full pack, scans and long filings included. ADE Split separates it into passports, formation documents and statements. ADE Extract returns names, identifiers and ownership percentages as JSON. At one global tier 1 bank, manual review time dropped 40 to 60 percent. The team saves hundreds of analyst hours every week. Every value traces back to the page it came from. So analysts defend each decision when an auditor asks. The full KYC and KYB breakdown is in the comments.

A parser can read every word of a contract and still miss whether it was signed. Traditional OCR treats a stamp or a signature as noise, or as an unlabeled image blob. So a system can transcribe a permit or a certificate of origin end to end and never register whether it was signed, stamped, or sealed. For teams processing regulated paperwork, that is the one answer that matters. Agentic Document Extraction (ADE) closes the gap with attestation detection. At the parse stage, it finds these marks, assigns each a type, transcribes the text inside, and records where it sits on the page. Every mark comes back as one of four labels: SIGNED, E-SIGNED, STAMPED, or SEALED. Labels stack when a mark carries more than one, so an official filing stamped and signed in the same spot returns as [STAMPED][SIGNED]. The result is a queryable field, with grounding coordinates down to the line. "Is this document signed?" becomes a value your code can read, which lets you gate straight-through processing and route missing marks to a human automatically. One note for builders: attestation detection runs on DPT-3 Pro. Full breakdown, plus two real document examples, in the comments.

Building an immigration financial audit pipeline on Amazon Web Services (AWS). Auditing 6 months of bank statements for one visa applicant takes 2 to 4 hours. Openomi, an audit platform for immigration consultants, does it in 60 seconds. It checks proof of funds against the rules of 10 distinct immigration programs. All of them run under Canada's IRCC. The hard part is program drift. Federal Skilled Worker requires CAD $15,263 across 6 consecutive months. Quebec Skilled Worker requires CAD $3,462 across 3. Canadian Experience Class requires no proof of funds at all. Mixing those rulesets produces false rejections. A Streamlit UI drops statements into S3. A Lambda orchestrator then calls two services in sequence. Agentic Document Extraction (ADE) runs first. ADE Parse converts each statement to markdown, holding multi-column layouts, handwritten notes and branch stamps. ADE Extract returns a typed schema: account holder, balances, currency, and every transaction. An AWS Bedrock Agent takes that JSON and runs the compliance and fraud checks. The core design decision was keeping program rules out of the system prompt. They live in a RAG knowledge base the agent queries before any document gets read. Full write-up, architecture, and source code in the comments.

Building an immigration financial audit pipeline on Amazon Web Services (AWS). Auditing 6 months of bank statements for one visa applicant takes 2 to 4 hours. Openomi, an audit platform for immigration consultants, does it in 60 seconds. It checks proof of funds against the rules of 10 distinct immigration programs. All of them run under Canada's IRCC. The hard part is program drift. Federal Skilled Worker requires CAD $15,263 across 6 consecutive months. Quebec Skilled Worker requires CAD $3,462 across 3. Canadian Experience Class requires no proof of funds at all. Mixing those rulesets produces false rejections. A Streamlit UI drops statements into S3. A Lambda orchestrator then calls two services in sequence. Agentic Document Extraction (ADE) runs first. ADE Parse converts each statement to markdown, holding multi-column layouts, handwritten notes and branch stamps. ADE Extract returns a typed schema: account holder, balances, currency, and every transaction. An AWS Bedrock Agent takes that JSON and runs the compliance and fraud checks. The core design decision was keeping program rules out of the system prompt. They live in a RAG knowledge base the agent queries before any document gets read. Full write-up, architecture, and source code in the comments.

A parser can read every word of a contract and still miss whether it was signed. Traditional OCR treats a stamp or a signature as noise, or as an unlabeled image blob. So a system can transcribe a permit or a certificate of origin end to end and never register whether it was signed, stamped, or sealed. For teams processing regulated paperwork, that is the one answer that matters. Agentic Document Extraction (ADE) closes the gap with attestation detection. At the parse stage, it finds these marks, assigns each a type, transcribes the text inside, and records where it sits on the page. Every mark comes back as one of four labels: SIGNED, E-SIGNED, STAMPED, or SEALED. Labels stack when a mark carries more than one, so an official filing stamped and signed in the same spot returns as [STAMPED][SIGNED]. The result is a queryable field, with grounding coordinates down to the line. "Is this document signed?" becomes a value your code can read, which lets you gate straight-through processing and route missing marks to a human automatically. One note for builders: attestation detection runs on DPT-3 Pro. Full breakdown, plus two real document examples, in the comments.

Onboarding one business client means reading 300 pages across a dozen documents. The hard part is agreement. Names, numbers and owners must match across all of them before the account opens. Then risk schedules a periodic review, and the file gets read again. Fenergo found only about a third of these reviews run automated. Agentic Document Extraction (ADE) turns every document into structured data, with no templates to build. A new form or issuer works on the first call. ADE Parse reads the full pack, scans and long filings included. ADE Split separates it into passports, formation documents and statements. ADE Extract returns names, identifiers and ownership percentages as JSON. At one global tier 1 bank, manual review time dropped 40 to 60 percent. The team saves hundreds of analyst hours every week. Every value traces back to the page it came from. So analysts defend each decision when an auditor asks. The full KYC and KYB breakdown is in the comments.

A parser can read every word of a contract and still miss whether it was signed. Traditional OCR treats a stamp or a signature as noise, or as an unlabeled image blob. So a system can transcribe a permit or a certificate of origin end to end and never register whether it was signed, stamped, or sealed. For teams processing regulated paperwork, that is the one answer that matters. Agentic Document Extraction (ADE) closes the gap with attestation detection. At the parse stage, it finds these marks, assigns each a type, transcribes the text inside, and records where it sits on the page. Every mark comes back as one of four labels: SIGNED, E-SIGNED, STAMPED, or SEALED. Labels stack when a mark carries more than one, so an official filing stamped and signed in the same spot returns as [STAMPED][SIGNED]. The result is a queryable field, with grounding coordinates down to the line. "Is this document signed?" becomes a value your code can read, which lets you gate straight-through processing and route missing marks to a human automatically. One note for builders: attestation detection runs on DPT-3 Pro. Full breakdown, plus two real document examples, in the comments.

The documents worth automating are the ones you cannot afford to leak. A loan file carries income, identity, and account numbers on one page. A claims packet carries diagnosis codes. A utility contract carries pricing terms and counterparty data. Finance, insurance, healthcare, energy, and legal teams automate these workflows anyway. The volume leaves them no other option. Every vendor in that pipeline becomes another place your data lives. Security review starts there, long before anyone opens the accuracy benchmark. Agentic Document Extraction (ADE) is built for that review. Zero Data Retention, available on Team and Enterprise plans, processes your files in memory. Nothing is stored at rest. Nothing goes into model training. Your data stays in the region you pick, AWS US East or AWS EU. Encryption runs TLS 1.2 or higher in transit and AES-256 at rest. Enterprise teams can deploy ADE inside their own VPC. Every value ADE returns links to its place in the source document. That traceability is what a compliance reviewer actually asks for. Full detail on how we handle your data:

Every API call gives you a choice: wait for the result now, or collect it later. Agentic Document Extraction (ADE) works both ways when you run a Parse or Extract. The first way is synchronous. You send the request and wait on the line until the result comes back. Simple, and right when someone needs the answer immediately. The second is asynchronous. You submit the work, get a ticket, and pick up the result once it is done. Nothing sits waiting on an open line. The result is identical either way. Same Markdown from Parse, same fields from Extract. You are only choosing how long you wait. So why ever hand it off? Two reasons. Big jobs need it. A single asynchronous Parse Job takes files up to 1 GB or 6,000 pages, far more than one live request can hold. It also costs less. Waiting on the line runs at the top rate, since something is blocked on it. Hand the job off and you drop to a cheaper tier for the same result. The rule of thumb is one question. If a person or an agent is waiting on the answer, run it synchronously. If nothing is waiting, submit a job and pay less. Full breakdown, plus the service tier companion, in the comments.

Agentic Document Extraction (ADE) Parse returns three components: Markdown, structure, and metadata. Markdown is the full document as one string in reading order. Structure is a document tree of typed blocks with inline grounding. Metadata is a receipt tracking page counts, billing, and more. Five optional config options shape that output, each with a sensible default if you leave it out: page selection, table format, dropping a block from the Markdown string, per-line bounding boxes, and node-level Markdown. The video breaks down each one:

Your document parser is blind to what it is parsing! A loan packet lands as one PDF carrying pay stubs, bank statements, tax returns, and IDs. The parser reads every page the same way, so it pulls fields that fit none of them. The parser needs to know what it is reading before it reads it. The fix is classification before extraction, not after. Each page needs a label before any schema gets applied to it. Agentic Document Extraction (ADE) fixes this. ADE Classify evaluates every page concurrently and assigns a label per page. Pay stubs route to the pay stub pipeline. Bank statements route to their own. Pages that do not fit get flagged with a suggested class. ADE Extract then applies the matching Pydantic schema per type. A pay stub returns employee name, pay period, gross pay, and net pay. A bank statement returns bank name, account number, and balance. The schema follows the page, not the other way around. Every value comes back with a chunk reference and a page-level bounding box, so each number traces back to its source. Full working notebook example in the comments.

The hard part of RAG isn't finding the right chunk! When you chunk a document for RAG, each chunk lands in the index on its own, disconnected from the section it came from. So two chunks with identical text, but from completely different parts of the document, look exactly the same to a flat index. That becomes a problem the moment a question needs context from more than one section. The chunks come back, but how they relate to each other gets lost. ADE Section fixes this. It reads the parsed document, builds the actual hierarchy, and figures out where every chunk falls within it. That gets attached to the chunk before it's embedded. Once that's in place, a broad question can stay at the section level. A specific one can drop into a sub-chunk. You can scope a search to one part of a document instead of the whole thing. Citations get more accurate too. The model knows exactly which section a fact came from. Run ADE Parse, then run ADE Section.