FROM DOCUMENTS TO A DATABASE YOU CAN TRUST
Your documents already contain a database. Talonic reads them once, returns typed rows each traced to its source line, and every system and agent you build queries that database from then on.
Response within 1 business day · Or start free: 5,000 credits/mo, no credit card
Ask the document estate
↳ ▋
| counterparty | value | renews | source · confidence |
|---|---|---|---|
| Northern Utilities | €1.24M | Sep 04 | p.12 · 0.96 |
| Municipal Heat Ltd | €640K | Sep 11 | p.08 · 0.94 |
| East Industrial Park | €2.08M | Sep 13 | p.21 · 0.97 |
3 rows · 0.38 s · every value traces to its source line
Note: Illustrative query. On your corpus the answers come from your documents, extracted once and never re-read.
TRUSTED ACROSS LOGISTICS, ENERGY, AUTOMOTIVE, AND CONTRACT OPERATIONS
01 / production
contract_value · illustrative01Five thousand documents, read once. Queried by every department since.
1M+ · source observations, live tenantThe migration needed 75 fields per contract dataset. One read of the estate holds 1,042,665 source observations and 52,937 reusable concepts, live tenant, 26 August 2026.
LIVE TENANT · 26 AUGUST 2026
Read the case study →
supplier_name · illustrative0220,000 documents, every page read, delivered in the shape Ivalua imports.
1,130 · contracts delivered, one row each, Ivalua import-readyA multilingual contract estate across thirty countries, in formats that never agreed, that purchasing could not query. One batch read once, text and image alike, and delivered as one row per contract in the shape Ivalua imports. The programme runs to 22,000 contracts.
PILOT COMPLETED · BATCH DELIVERED AND RECEIVED, ERP IMPORT-READY
Read the case study →
lane_rate · illustrative0396 percent against legacy OCR’s 78, on the same documents, scored the same way.
96% vs 78% · same documents, same labels, same scoringThe same hand labelled documents, two fields, both systems scored the same way. Replacing a six figure incumbent.
IN PRODUCTION · AUTO BILLING IN SHADOW RUN SINCE 12 AUGUST 2026
Read the case study →Automotive, research, and industrial operations. Different document estates, the same database.
metric_value · illustrativeAutomotive · Management reporting · Report intelligence
Management reports that were already approved, already read, and already filed. Nineteen months from 24 verticals, with drifting labels and formats normalised into one queryable layer. Nobody had to rebuild the systems underneath.
POC AND INSIGHT PACKAGE COMPLETED · SCALE-UP PROPOSED
Read the case study →
hours · illustrativeSocial research · Legal data collection · Regulation transcription
German school timetable law from 1996 to 2021, transcribed exactly as printed. Every row carries its source page, a confidence score, and a note on anything ambiguous.
ONE BATCH DELIVERED · THE CORPUS GATED
Read the case study →
tolerance_band · illustrativeIndustrial hydraulics · Technical product data · Curve digitisation
Performance and reliability data that existed only as printed curves on technical datasheets. Talonic finds each chart, reads its axes and curves, and delivers the X/Y series as tables an engineer can reuse: 736 charts, about 3,300 curves, each sampled to 100 points and checked back against the page.
METHOD AND PILOT ESTABLISHED · DETECTED CHARTS DELIVERED · SCALE-UP FOLLOWS APPROVAL
Read the case study →02 / one pass
Any source: file shares, email, scans, RPA feeds, system exports. 25+ formats, four languages at production quality.
PDFs · scans · handwriting · spreadsheets · emailsThe Schema Graph finds the fields, entities, and relationships on its own. Canonical fields compound across every document you add; humans review only low-confidence values.
vendor_name · contract_value · governing_lawSQL, natural language, REST, or straight into your agents. Ask a question, get typed rows with evidence.
WHERE auto_renew = true AND notice_period_d < 30
Note: The database, live. Illustrative capture, not a customer tenant. · source lines preserved
03 / put it to work
Push the fields a system needs into Dynamics, Ivalua, Salesforce, or any endpoint. Every delivery is signed, retried, and logged. The work you used to script with RPA runs on the database instead.
Approve a load for billing. Flag a contract for review. Find the entitlement nobody claimed. When a rule cannot decide alone, it asks a person, and the answer improves the rule next time.
Give an agent exactly this: these contracts, this reference table, nothing else. It queries the database, every read is logged, every answer carries its source. You decide what it sees. You can prove what it saw.
04 / live agents
Each of these three reads only what it was granted, decides by rules a person wrote in their own words, and writes every decision to a record nobody can edit.
four words. No synonyms, no adjectives
DECIDE
produces · a hold
Rules read every contract before a person does. One of them stops the contract.
RUNNING · FLAGS AND HOLDS
How Contract Review works →DECIDE
produces · a verdict
The rules a billing team only ever said out loud now decide, in writing, whether three freight documents agree.
SHADOW RUN · DECIDES BESIDE THE BILLER
How Auto Billing works →DELEGATE
produces · a question the estate can answer
Because the read was never scoped to one schema, a department outside the original scope gets its answer in an afternoon rather than a project.
ON REQUEST · A STANDING QUERY
How Money Found works →05 / receipts
Three benchmarks published, same method.
28 SEC 10-K filings, preregistered protocol, XBRL gold labels. Structure-first answers 5 of 5 headline ranking queries; BM25 top-12 answers 0 of 5. A 39-query battery over 53 filings against hybrid RAG and a long-context baseline is on the same page, including the rows Talonic loses.
Retrieval’s cost line climbs with every question. Talonic’s structured path flattens after ingestion: $10.14 once for 53 documents, then $0.00 per 1,000 questions. Same corpus, same questions, published with the protocol, including the agent path that is not flat.
A database returns the same answer every time. Embeddings do not. Asked ten times, hybrid RAG matched itself on 2 of 39 questions. The structured path was byte-identical across ten passes of 100 questions.
06 / provenance
Most tools return a value. Talonic returns a value, the line it came from, the region of the scan that produced that line, the confidence, the phase, and the reasoning. Auditable by default. Defensible by design.
07 / why a data registry
Without a registry, adding a downstream system means re-running every extraction. Cost scales as documents times systems.
Every new agent, workflow, or integration multiplies your extraction cost. Unless the data is already in a registry.
10%
Of enterprise unstructured data is ever analyzed.
5%
Of enterprise GenAI pilots reach production.
Retrieval re-reads the same corpus on every question, re-spends the same tokens, and can return a different answer tomorrow. When the answer feeds an ERP entry or a filing, “the vector was close” is not provenance. Talonic reads each document once, resolves the schema from the documents themselves (529 document types, zero templates), and returns a database where every value traces to its source line, region, confidence, and reasoning.
See the published benchmark against RAG →DOCUMENT ONTOLOGY
From Schedule K-1 to Bill of Lading (Ocean), from Notarial Deeds to QC Inspection Forms — Talonic understands the structure of every enterprise document type out of the box.
Schedule K-1, Form 1099-MISC, Form W-8BEN, Form 1040, Form 990, and 48 more
Purchase Order, Commercial Invoice, Pro Forma Invoice, Goods Receipt Note, Three-Way Match Report, and 48 more
Bill of Lading (Ocean), Bill of Lading (Inland), Air Waybill (AWB), House Air Waybill (HAWB), Sea Waybill, and 48 more
Notarial Deed, Master Service Agreement, Statement of Work, Non-Disclosure Agreement, License Agreement, and 48 more
Commercial Register Extract, Certificate of Good Standing, Certificate of Incorporation, Annual Return, Director's Report, and 48 more
Patient Intake Form, Medical History Questionnaire, Informed Consent Form, Discharge Summary, Operative Report, and 48 more
QC Inspection Form, Certificate of Analysis (CoA), Certificate of Conformance, First Article Inspection Report, Material Test Report, and 48 more
Insurance Application, Policy Declaration Page, Insurance Policy Wording, Endorsement / Rider, Certificate of Insurance, and 48 more
Property Deed, Title Report, Land Registry Extract, Survey Plan, Zoning Certificate, and 48 more
Employment Application, Offer Letter, Employment Contract, Background Check Report, I-9 Employment Eligibility, and 47 more
DIN SPEC 91491
Talonic co-authored DIN SPEC 91491 with Fraunhofer IIS, Humboldt-Innovation, GIIC, and the German standards body. Europe's first standard for AI-ready data at the schema layer.
As the EU AI Act extends into data-layer compliance, DIN SPEC 91491 defines what schema-layer data readiness looks like. Enterprises aligning with the standard need an implementation that was built alongside it.
The standard we helped write. The implementation we ship.
REST API, Node SDK, or MCP server (Model Context Protocol). Every response carries typed fields, confidence, and provenance, so your agent knows when not to trust an answer. Free tier: 5,000 credits a month, no credit card.
npx -y @talonic/mcp@latestTalonic is the schema layer for enterprise data. It transforms unstructured documents into schema-validated structured data with per-cell provenance and full audit trail.
Talonic supports over 25 file formats including PDFs, scans, Word documents, Excel spreadsheets, images, emails, and more, classified into a 529-type document ontology.
Yes. Talonic is GDPR and HIPAA compliant, ISO 27001 and ISO 42001 aligned, and co-authored DIN SPEC 91491. All data is processed in EU-resident infrastructure in Germany West Central.
Europe's first standard for AI-ready data at the schema layer, published in 2025. Talonic co-authored the standard alongside Fraunhofer IIS, Humboldt-Innovation, and GIIC.
OCR converts pixels to text. Talonic classifies documents against a 529-type ontology, extracts schema-validated fields with confidence scores, reconciles entities into cases, and delivers typed data to systems of record with full audit trail.
Yes. Send a representative sample of your documents and Talonic runs them within five business days. You receive a field-level analysis of extractability, confidence distribution, recommended schema, and a production roadmap.
08 / your documents
Send a representative sample of contracts, scans, case files, or operational documents. We run them through Talonic and send back your extracted data within five business days: field-level coverage, confidence, and provenance. If the database doesn’t beat what you have, you’ll know exactly why, field by field.
Photography: A.Savin (FAL) · U.S. Air Force, public domain · Marcin Wichary (CC BY 2.0) · Prime Minister’s Office, India (GODL) · Coenen (CC BY-SA 3.0) · Kleuske (CC BY-SA 3.0) · Blogtrepreneur (CC BY 2.0) · Virtual-Pano (CC BY 4.0) · Zonk43, public domain