Extraction
Structured data extraction from any document
Define the data you need once, as a schema. 2kw.ai reads 17 document formats, at any length, and returns clean, structured JSON — with grounding scores that link every value back to its source.
How it works
- 01Define a schemaField names, data types, structure — you describe the output you need once, and every extraction follows it.
- 02Send documentsUpload via the dashboard, REST API, CLI, or straight from your editor through the MCP server.
- 03Extraction runs at any sizeLong documents are read in full, from the first page to the last, and heavy workloads process asynchronously with retries.
- 04Get structured output backResults merge against your schema into clean, structured data. If something doesn't match the schema, you see it immediately.
Built for production document pipelines
17 input formats
PDF, DOCX, XLSX, PPTX, CSV, MSG, EML, HTML, Markdown, TXT, AsciiDoc and ZIP — plus any image file, and the engineering formats DXF, DWG, STEP and GEO (TRUMPF).
Schema-driven output
Extractions follow the schema you define instead of whatever the model felt like returning. Mismatches surface immediately, not downstream.
Version-controlled schemas
Iterate on schemas without breaking production. Every version is tracked, so you can test changes before they go live.
Grounding scores
Every extracted value links back to the place in the source document it came from — quality you can inspect, not just trust.
Documents of any length
A 300-page contract or a single scan: every page is read, and you get one result against your schema.
Async with retries
Heavy workloads run asynchronously with automatic retries, so a flaky page doesn't sink a thousand-document batch.
Conversion and transcription included
The same pipeline converts documents to Markdown, plain text, JSON, or HTML when you need content rather than fields. Audio transcription covers FLAC, MP3, MP4, MPEG, MPGA, M4A, OGG, WAV, and WebM, with output as text, JSON, or SRT/VTT subtitles.
No PDF parsing libraries. No regex nightmares. No brittle per-layout templates — the schema is the contract, and the platform does the rest. Quickstarts live in the documentation.