Skip to content
Rush Commerce
AI & Automation2 min read

Unsiloed AI raises $7.5M for agent-ready document parsing

Unsiloed AI raised $7.5M to turn PDFs, tables and charts into data AI agents can use. Its 99.78% accuracy claim isn't independently verified.

Unsiloed AI raised $7.5 million to fix the step that breaks most AI agent projects: getting clean data out of documents. The seven-person San Francisco startup parses PDFs, tables and charts into structured data agents can use, and shows where each value came from, TechCrunch reports. The accuracy numbers are big. They're also the company's own.

What actually happened

Per TechCrunch:

  • The round: $7.5 million seed led by Felicis, with Y Combinator, Orange Collective, Uncorrelated Ventures and angels. Founders Aman Mishra and Adnan Abbas came through YC's Fall 2025 batch.
  • The product: it breaks complex documents into parts, pulls data from tables and charts, and links each value back to its spot in the source. That citation trail is what lets a person check the agent's work.
  • The accuracy claim: 99.78% across about 84,000 data points from 542 financial documents one customer supplied, per a technical brief from the startup. TechCrunch notes this is "not independently verified."
  • The benchmark claim: on Allen AI's public olmOCR-Bench (1,403 PDF pages), Unsiloed says it scored 88.0 against 66.0 for Reducto's agentic system and 39.9 for Unstructured. Unsiloed ran those tests itself.
  • Customers: finance, legal and healthcare, per Mishra. One law firm runs it on its own servers. Voice AI startup Bolna AI uses it to ingest knowledge bases.
  • Price: starts at one cent per page, according to Mishra.

Why it matters for your business

Most "the agent got it wrong" stories are parsing stories. The model never saw the right number because the table came out of the PDF scrambled. Invoices, packing slips, supplier price sheets, insurance forms: if your agent reads those, extraction quality sets the ceiling on everything after it.

Treat vendor accuracy numbers as a starting point. A 99.78% score on one customer's financial filings says little about your faded scanned bills of lading. The fix is cheap at a penny a page:

  • Pull 200 real documents that represent your mess, ugly scans included.
  • Hand-label the fields you actually use.
  • Run two or three parsers against that set and score them yourself.
  • Require source citations so a human can spot-check fast.

That afternoon of work tells you more than any leaderboard.

Key takeaways

  • Unsiloed AI raised a $7.5M seed led by Felicis for document parsing aimed at AI agents
  • It extracts tables and charts and cites each value back to the source
  • 99.78% accuracy and the olmOCR-Bench lead are company-run tests, not independent ones
  • Pricing starts at one cent per page, per the CEO
  • Benchmark parsers on your own documents before you build on one

Drowning in PDFs your systems can't read? We build document pipelines that test parsers against your real files, keep a source trail on every field, and send only clean data to your agents. See our work or send us a sample batch.

Sources: TechCrunch, Unsiloed AI.

  • #document-parsing
  • #ai-agents
  • #pdf-extraction
  • #ocr
  • #funding
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.