Skip to main content
Pulse is the document layer for enterprise AI. Agents, data platforms and your own applications use it to read the documents your company holds, from filings and scans to handwritten forms and workbooks. Pulse turns those documents into AI-ready data: clean Markdown, structured JSON, tables, chart data and a bounding box for every element, so each value traces back to its place on the page.

Quickstart

Parse a document and extract fields from it in Python, TypeScript or cURL.

SDKs

Install the Python or TypeScript SDK and call every step from your code.

Pricing

See what each step costs in credits, and what is free.

API reference

Look up every route, field and response.
Pulse reads PDFs, scans and photos, Word documents, PowerPoint decks, Excel workbooks, CSV files and HTML pages. See Files and limits for every format and size limit.

​
How it works

Every step except Classify reads a parse by its parse_id, so you parse a document once and then run as many steps on it as you need. Classify reads the file itself, before any parse. Each step returns its own id, and you read the step’s result with that id:

​
Endpoints

Parse

Turn a document into Markdown and elements, with figures, footnotes or chunks.

Extract

Fill a JSON Schema from a parse, with a citation for each value.

Split

Find each topic’s pages, to route them or extract each one.

Tables

Get every table as HTML, JSON or Markdown, merged across pages.

Charts

Read each chart’s series and points, and export them to Excel, CSV or LAS.

Classify

Sort a document into the categories you name, before you parse it.

​
Run at scale

Pipelines

Parse a document and run up to 10 steps on it in one call.

Batch

Parse up to 1,000 documents with one config.

Saved configs

Save an endpoint’s config once, then send its config_id.

Webhooks

Hear when your jobs change status, with signed deliveries.

​
Cookbooks

Split, then extract

Extract each section of a long report with its own small schema.

Highlight sources

Draw an extracted value’s exact words on the page.

RAG ingestion

Chunk a parse, with metadata that traces answers to pages.

Cloud intake

Parse a bucket of documents as one batch.

​
Base URL

Send every request to:
Every request needs an API key. Authentication shows how to create one and send it.

​
Need help?

Support

How to reach Pulse, and what to send with a question.

Moving from V1

Map your version 1 calls to version 2.