Extract data from any PDF. Get back typed JSON.

Sensible mixes LLMs (Claude, Gemini, GPT) with deterministic methods, field by field, to handle the documents that defeat either approach alone.

✓

PDFs, images or spreadsheets. Extract data from native files and scans. OCR built in, multi-page tables intact.

✓

Don't want to learn another platform? No problem. Our engineering team can partner with you and take that work off your plate.

✓

We price per document. Stop worrying about page counts and token costs.

IN PRODUCTION AT

★★★★★

4.9

ON G2

“What we've come to appreciate from Sensible is the blend of accuracy and speed. With other solutions you trade off one or the other; with Sensible we got a quick turnaround from the API while preserving very, very high accuracy.”

DANIL KOLESNIKOV · LEAD ARCHITECT

73%

less manual loss run review at Ledgebrook, after replacing an internal GPT build with Sensible's hybrid extraction.

LEDGEBROOK · INSURANCE

Why use Sensible

Same API either way. What changes is how much of it you write.

For engineers

Combine AI + deterministic methods

Sensible blends modern LLMs with deterministic methods. This means simple documents are processed without using AI, with no chance for hallucinations.

Normalization happens during extraction

Dates, currencies, and units come back in the format you specified. There is no mapping layer to write between the response and your database. Key values match your schema.

Validation runs before you get the response

Type checks, range limits and cross-field reconciliation live in the configuration. Bad data fails at the API, not three services downstream.

Built to survive a security review

SOC 2 Type II and HIPAA ready, with documented storage and deletion, and every field traceable to the page it came from.

Read the API docs

For product and operations teams

Use our prebuilt and pretested configurations

We've worked for years with specialty documents and have a built out configuration library, 150+ ready to use.

Run it without writing code

Drag files into the Extract tab and download Excel, or describe the fields you want with Sensible Instruct rather than writing config. Zapier routes output to Airtable, Google Sheets or Slack.

Hire us as your engineers

Don't want to spend dev time on this? No problem. Our team can partner with you and do all of the setup, custom code templates and build a custom integration if needed.

Priced per document, because explaining tokens sucks

Our default pricing plans price per document, which in most cases works out to an individual API call. Stop getting hit with overages when models change or you parse a large package.

Ask about managed services

The integration is one request.

Sensible is built for developers but easy enough to use for everybody. Use our templates out of the box or have our team run the entire process for you. All you need to do is set up the API and send us the documents.

REQUEST

curl -X POST \
  "https://api.sensible.so/v0/extract/loan_estimate" \
  -H "Authorization: Bearer $SENSIBLE_API_KEY" \
  -H "Content-Type: application/pdf" \
  --data-binary @loan_estimate.pdf

RESPONSE200 OK

{
  "id": "424dd912-47b0-4e17-a7c0",
  "status": "COMPLETE",
  "parsed_document": {
    "lender_name": { "value": "FICUS BANK" },
    "loan_amount": { "value": 162000, "source": "$162,000" },
    "loan_term":   { "value": "30 years" } …
  },
  "validations": [ … ]
}

Don't start from zero.

Sensible works with even the most challenging documents, many of which are already prebuilt for you to use. See anything missing from our library? Reach out and an engineer can build custom workflows based on your actual documents.

Invoices Bank statements Loss runs ACORD forms Pay stubs EOBs Closing disclosures Bills of lading Rate confirmations Utility bills Tax forms Browse the library →

More than an OCR layer

Sensible is an orchestration layer for your entire document workflow.

The Sensible editor: SenseML configuration on the left, the source document with highlighted values in the middle, and the extracted data on the right

Config on the left, the document it read in the middle, the output on the right.

Designed for accuracy

See the details of each extraction and adjust every field so you get consistent data from every document.

Normalization happens in app

Dates, currencies and units come back in the format you want. No mapping layer downstream.

Validations sit with the fields

Type checks, ranges and cross-field rules run before the response leaves the API.

Then wire it up

Python and Node SDKs, REST, and a webhook when an extraction completes.

Start Free

100 documents free

· 150+ configurations built · no credit card

Try it on the PDF you came here with.

Start Free Book A Demo

14-DAY TRIAL · NO CREDIT CARD

Why not just prompt Claude?

Sensible calls the same frontier models you would. The difference is what production asks for: accuracy you can check, an audit trail, certified handling, and a price you can quote.

ACCURACY

Guessing is not extraction.

A raw model call returns its best guess, drifts when the model updates, and can invent a value to fill your schema.

✓

Sensible validates each field against your schema and routes uncertain values to a reviewer.

AUDITABILITY

Every number has an address.

Ask a chatbot where a value came from and you get another guess.

✓

Each value carries coordinates to its spot on the source page. Auditors get the trail without asking.

SECURITY

Answers for the security review.

A DIY pipeline makes you the one filling out the security questionnaire.

✓

SOC 2 Type II certified. HIPAA ready. Storage and deletion documented.

PREDICTABLE PRICE

A price you can put in the plan.

Token bills move with page count, prompt size, and the model of the week.

✓

One flat price per document. Price the workload before you run it.

Price it before you process it.

We try to keep things simple by pricing per document so you aren't estimating token costs or getting penalized for a larger package.

Growth

For teams getting started with production document extraction.

$449

/mo billed annually

✓

750 documents / month

✓

Unlimited document layouts

✓

Automatic & selective OCR

✓

$0.57 per document over

Start Free

Security your compliance team can verify.

SOC 2 Type II certified. HIPAA ready. The full detail, including how documents are stored and deleted, lives on the security page.

Read the security details

SOC 2 TYPE II

HIPAA

GDPR

What engineers ask before they start.

Does it work on scanned PDFs?

Yes. Sensible runs OCR on scans and images, and one configuration handles native and scanned versions of the same document. Fields that come back below your confidence threshold get flagged for review instead of passed through.

Can it handle tables that continue across pages?

Yes. A multi-page table comes back as one array. Merged cells and repeated headers are handled in the configuration, not in your post-processing.

How is this different from sending the PDF to Claude or ChatGPT?

Sensible calls the same models under the hood, so the difference isn't the AI. Prompting Claude or ChatGPT directly leaves you owning the schema checks, the retry logic, the drift after model updates, and the "where did this number come from" question. With Sensible you pick the method per field, LLM prompts where layouts vary, deterministic rules where they don't, and results are validated before you see them.

Can I get CSV or Excel instead of JSON?

Yes. The API returns JSON, and any extraction can be exported as CSV or Excel. Teams pipe the JSON into systems and hand the spreadsheets to operations.

What does it cost?

Per document, not per page or per token. One extraction result is one document. The full table is on the pricing page.

Do we have to build the extraction ourselves?

No. 150+ prebuilt configurations cover common document types, and our team can build and run the extraction backend for you. Many teams do some of both.

Start with the document that breaks everything else.

Start Free

14-DAY TRIAL · NO CREDIT CARD · 150+ PREBUILT CONFIGURATIONS

Need to evaluate the integration first?

Read the API docs.

Rather have it built for you?

Ask about managed services.