The AI that reads your documents and tells you how sure it is
Sonar is the engine under every Blue-IQ product. It reads like a careful person, scores every value it extracts, and flags what it cannot read instead of guessing.
Sonar at a glance
- Reads
- PDFs, Word, scans, photos, exports
- Returns
- Schema-validated JSON
- Confidence
- A score on every field
- Powers
- Capture, Spend, Govern, editions
Four steps, the same in every product
- 01ReadOCR, layout, tables, stamps and handwriting, read in context.
- 02ExtractThe fields your schema asks for, from any layout, no templates.
- 03ScoreA confidence score on every value, not one per file.
- 04RouteAbove your line to your systems; below it, to a person.
Built to be trusted, value by value
- A score on every field
- Trust each value on its own: route the sure ones, review the rest.
- Flags, never guesses
- When Sonar cannot read a value with confidence, it says so.
- No templates
- New layouts are read on first sight; nothing to maintain.
- Your schema
- Every response validated against the fields you define.
- Create and Ask
- Draft documents from plain language, and ask questions across everything you hold.
- Built for volume
- Up to 200 documents per API call, with a sub-second median read.
Five fields from one contract
Four go straight to your systems. One is flagged for a person, because Sonar could not read the signature with confidence.
| Field | Value | Confidence |
|---|---|---|
| counterparty | Ardent Freight Ltd | Accepted 0.99 |
| effective_date | 2026-03-01 | Accepted 0.97 |
| total_value | 482,500.00 USD | Accepted 0.95 |
| payment_terms | Net 45 | Accepted 0.93 |
| signatory | illegible | Review 0.58 |
Sonar, answered
What is Sonar?
Sonar is the AI inside every Blue-IQ product. It reads a document the way a person would, extracts each field, scores its confidence in every value, and flags anything it cannot read instead of guessing.
What happens when the engine cannot read a field?
It flags it. Every value comes back with a confidence score, and anything the engine is unsure of is sent for review rather than guessed. A wrong value that looks confident costs far more than a field marked for a person to check.
What document types can it handle?
PDFs, DOCX, images and exports, including scans and pages photographed on a phone, which go through OCR first. If your team can read it, the engine is built to read it.
How does it handle volume?
Up to 200 documents per API call, with throughput that holds steady across batches.