Now with structured extraction
Turn documents into data your systems can actually use.
Upload a contract, get typed JSON back with a confidence score on every field and a citation pointing at the page it came from.
Built for the boring requirements
The things that stop a pilot becoming production.
Auditable
Citations on every field
Each extracted value links to the page and bounding box it came from. Reviewers verify in seconds.
Auditable
Zero retention
Your data stays yours
Nothing is stored after the response returns, and nothing is ever used for training.
Zero retention
Typed output
Deterministic schemas
Define the shape once. Output validates against it or fails loudly — never silently.
Typed output
0
Days of data retention
99.2%
Field accuracy, benchmarked
SOC 2
Type II, audited
1.8s
Median document latency
Integrations
Connects to what you already run
REST and gRPC
The same schema either way, with generated SDKs for six languages.
S3 and GCS
Point us at a bucket and we process on write, with signed callbacks.
Zapier
For teams who want extraction without writing an integration.
Salesforce
Write extracted fields onto the record with citations attached.
SharePoint
Index existing document libraries without moving the files.
Webhook callbacks
Signed, retried with backoff, and replayable for thirty days.
Pricing
Priced so you can work it out yourself
Sandbox
Free
500 pages
- ✓
500 pages a month
- ✓
All extraction schemas
- ✓
Zero data retention
- ✓
Community support
Growth
$0.04
per page
- ✓
Volume pricing from 10k pages
- ✓
Custom schemas
- ✓
99.9% uptime SLA
- ✓
SOC 2 Type II report
- ✓
Support within one business day
Enterprise
Custom
annual
- ✓
Committed volume discount
- ✓
VPC or on-premise deployment
- ✓
Custom model tuning
- ✓
DPA and subprocessor review
- ✓
Named solutions engineer
“The citations are what got it past our compliance team. Reviewers can check any field in two clicks instead of rereading the contract.”
Priya Raghavan
Head of Operations
Questions people ask first
Is my data used for training?
No. Documents are processed in memory and discarded when the response returns. Nothing is retained and nothing is used for training, on any plan including the free sandbox.
What does the confidence score mean?
It is the model probability for that specific field, calibrated against our benchmark set. Below 0.85 we recommend routing to a human reviewer, and the API can flag that for you.
Which file types work?
PDF, DOCX, PNG, JPEG and TIFF, including scans. OCR runs automatically when a document has no text layer.
How do citations work?
Every extracted value returns a page number and a bounding box, so your reviewer interface can jump straight to the source.
Test it on your own documents.
Upload twenty files in the sandbox. Nothing is retained, and you keep the output.