News & Updates

Most of what regulated companies know is locked in PDFs, scans, drawings, and maps their warehouse cannot read. Here is how Pulse turns those documents into structured, confidence-scored data that lands in Snowflake as governed tables.

How Pulse extracts spreadsheet evidence for AI agents, and what it showed on Surge's GDP.xlsx benchmark.

Extraction gets the numbers out of a commercial loan package; reconciliation across the rent roll, operating statement, leases, and appraisal is what tells the underwriter whether they agree.

Pulse is now a connector in Claude's directory. Point Claude at any document by URL and get back clean, structured data without leaving the conversation.

Basin screening matrices invert the usual table shape, with rows as evaluation criteria and columns as assets, and most cells carry qualified language rather than clean values. Preserving that structure is what keeps a screening decision auditable years later.

Invoice capture stalls short of straight-through processing when line items break and routing data sits off the invoice. Calibrated confidence opens the bypass lane.

The documents utilities generate, collect, and process every day were never built for downstream systems, which is why generic extraction fails on them.

Pulse now returns a file's native metadata, including authorship, revision history, timestamps, hidden worksheets, EXIF fields, and hashes, alongside the extracted content in the same Extract call.

A chart stores most of its meaning in the height of a bar or the slope of a line rather than in any character on the page. Pulse Agentic Chart Reconstruction recovers the values, scales, series, and visual encodings inside it.

Financial spreading resists automation because statement meaning lives in visual hierarchy that OCR flattens. Structure-preserving extraction turns transcription into review.

Classify identifies document types before extraction and routes each file into the right pipeline automatically, cutting cost and removing manual triage from intake.

Point an agent at our API with no account, no API key, and no signup, and it can pay for extraction on its own, one call at a time.

The Pulse CLI brings Pulse's full document extraction pipeline to your terminal, letting you convert PDFs, folders, and URLs into clean Markdown and structured JSON with a single command. Built for developers, it integrates seamlessly with shell scripts, CI/CD, and automation workflows, making document processing fast, scriptable, and agent-ready.

Pulse is launching Selection Mark, a purpose-built model that reads the checkmarks, circles, underlines, and strikethroughs people make on forms, where frontier VLMs reliably fail. It adds selection state to the standard extraction API in under 20 milliseconds per page with no extra calls.

The Pulse MCP server is live today. Point any MCP client at Pulse, whether that's Claude, Codex, VS Code, or one you've built yourself, and your agent can call extraction, schema generation, splitting, and table parsing directly as tools, deciding for itself which steps a given document needs.

Surge's new benchmark shows why frontier models need structured, page-grounded evidence to answer questions over professional PDFs.

Mistral released OCR 4 - the obvious question on our end working in the space was a narrow but vital one: how does it handle tables? Table extraction is one of the toughest parts of document understanding, so we ran OCR 4 through the full PulseBench-Tab document set the same day it shipped.

Operators digitized their well logs years ago, but a scan only preserves the image of a document, not the depth tracks, curve scales, and spatial relationships that make the subsurface record usable as data. This piece shows why the hardest records to extract are also the most valuable, and how Pulse's vision-language models finally read well logs, schematics, and maps as structured, traceable data across even the faded and handwritten archive.

Field tickets are the financial source records connecting oilfield work to the ledger, yet their endless vendor-by-vendor variation, handwriting, and degraded scans break conventional extraction and quietly corrupt cost reconciliation, which is why Pulse uses vision-language models with normalized schemas and source-level provenance to turn them into clean, traceable, audit-ready data.

Oil and gas is an industry that runs on documents that were never meant for machines, and the capital allocation decisions that matter at an integrated operator are made on the basis of a document stack that spans every era of the business.

The documents that drive upstream decisions, from well schematics to stratigraphic charts, hold their meaning in depth columns and spatial layouts that off-the-shelf AI simply can't follow. Pulse reads them the way your engineers do, and delivers structured data ready for the workflows downstream.

Excel extraction is widely treated as the easy cousin of PDF extraction. Spreadsheets are structured, the reasoning goes, so a parser should be able to walk the grid, read the values, and emit clean tabular output

Pulse Form Fill builds on top of our state-of-the-art extraction models, leveraging the same visual document understanding that powers Pulse's extraction pipeline as the foundation for accurately placing data back into any PDF.

Pulse turns the highest-value documents in regulated industries, from handwritten geotechnical boring logs to polished institutional research, into structured, audit-grade data with every field grounded back to its location on the source page.