Extraction is rarely the interesting part of a system — it is the part that quietly ruins the interesting part. These are the places it tends to bite.
Chunking a raw HTML dump means embedding the nav bar, the cookie notice and the footer on every page — and then watching them surface as answers. Extractor hands the pipeline the document only, with headings intact so your chunker has real boundaries to cut on.
An agent that fetches a URL gets markup. Give it the MCP tool instead and it gets the page as prose it can reason over — and because the tool reports what a document cost, the agent can budget its own reading.
Pass a schema — parties, effective date, term, notice period, governing law — and get those fields back for each document, with nulls where the contract is silent. The nulls are the useful part: they tell you which files need a human.
A PDF with a text layer is parsed directly; a photographed receipt goes through OCR at the same price. Either way you get totals, dates and line items in the shape you asked for, and a scan that turns out to be blank costs nothing.
Pricing pages, changelogs, docs sites. Extract on a schedule, diff the plain text, and you see what actually changed instead of a wall of markup churn from a redeploy.
PowerPoint and Excel come back as markdown — slides as sections with their speaker notes, sheets as tables. That is enough structure to index, and enough to read.
You describe the shape you want. Extractor fills it only from text that is actually in the document, and tells you which fields the document never mentioned — because a null you can see beats a value that looks plausible and isn't there.
2 credits, whatever the schema.
curl -X POST https://extract.ounie.com/api/extract \
-H "Authorization: Bearer xtr_live_..." \
-F file=@msa.pdf \
-F format=json \
-F schema='{
"parties": ["string"],
"effective_date": "string",
"term_months": "number",
"notice_period_days": "number",
"governing_law": "string"
}'
→ { "json": { "parties": ["Acme Inc.", "Globex Ltd."],
"effective_date": "2026-04-01",
"term_months": 24,
"notice_period_days": null,
"governing_law": "Delaware" },
"missing_fields": ["notice_period_days"] }A URL, a PDF, a deck someone emailed you. Two credits, a couple of seconds, and text you can actually use.
No subscription — pay in Ounie credits