Unsiloed AI, a seven-person San Francisco startup that turns complex business documents into data AI agents can use, has raised $7.5 million in seed funding led by Felicis, TechCrunch reported on Wednesday. Y Combinator, Orange Collective and Uncorrelated Ventures also took part.
Our number: on the startup’s own olmOCR-Bench test, Unsiloed scored 88.0 against 66.0 for Reducto’s agentic system, a 22-point gap by our arithmetic. The results come from tests the startup ran, and TechCrunch says its customer accuracy figures have not been independently verified.
Business Pill · Least Privilege
A Business Pill on least privilege: giving software, including AI agents, access only to the data and actions a task needs.
The key insight: Unsiloed AI raised $7.5 million led by Felicis with seven staff to turn documents into data agents can use; on its own olmOCR-Bench test it scored 88.0 against Reducto’s 66.0, a 22-point gap by our arithmetic.
What Unsiloed Raised and Sells
Per TechCrunch, angel investors Aarthi Ramamurthy and Beth Turner also joined the round. CEO Aman Mishra co-founded the company with Adnan Abbas, a former classmate at IIT Kharagpur, and the two joined Y Combinator’s Fall 2025 batch.
Its software breaks down documents, extracts information from tables, charts and other elements, and turns it into structured data that agents can use, while showing where each extracted item came from in the original document. Mishra said pricing starts at one cent per page, which works out to $10 per 1,000 pages by our arithmetic.
Mishra said: “The problem with LLMs is they’re inherently probabilistic”.

The Benchmarks It Cites
TechCrunch reports that in a test of 542 financial documents supplied by an enterprise customer, Unsiloed achieved 99.78% accuracy across roughly 84,000 extracted data points, per a technical brief the startup shared. At that rate, about 185 of the 84,000 data points would still be wrong, by our arithmetic.
On olmOCR-Bench, a public benchmark developed by Allen AI covering 1,403 PDF pages, Unsiloed says it scored 88.0, against 66.0 for Reducto’s agentic system and 39.9 for Unstructured, in tests conducted by the startup.
Customers and Competition
Mishra told TechCrunch that one bank uses the technology to process documents and a law firm runs it on its own servers for contracts and old case files, without naming either; he named voice AI startup Bolna AI as a customer. TechCrunch notes the market already includes Reducto and Unstructured, as well as offerings from Microsoft and Amazon.
The round comes as agent infrastructure draws money and scrutiny this week: Workday agreed to buy Honeydew so agents stop improvising numbers, and Reuters tested Meta’s Muse agent.
The Structural Read
As we read TechCrunch’s account, the pitch is error cost, not parsing: the startup’s own 99.78% figure still leaves about 185 wrong data points out of roughly 84,000, by our arithmetic.
As we read the pricing, one cent per page, or $10 per 1,000 pages by our arithmetic, positions document parsing as a metered input that agents call on.
Aman Mishra, co-founder and CEO, Unsiloed AI (to TechCrunch)
“There is going to be some hallucination involved, and that has a massive negative impact downstream”
Three Implications
FOR ENTERPRISES TechCrunch reports Unsiloed shows where each extracted item came from in the original document, so results can be checked.
FOR DOCUMENT-AI RIVALS TechCrunch names Reducto, Unstructured, Microsoft and Amazon among the competitors.
FOR AGENT BUILDERS Mishra named voice AI startup Bolna AI as a customer that uses Unsiloed to ingest knowledge bases.
The Business Engineer Lens
This story maps onto the Business Engineer framework The Agentic Harness War.
The analysis puts it this way: “The app stops being a destination and becomes a capability the harness invokes.”
As we read this round, Unsiloed is building one such capability: document parsing that agents call per page. The analysis predates this round and does not discuss Unsiloed.
What Is Not Established
The 99.78% accuracy figure and the olmOCR-Bench scores come from the startup’s own tests and technical brief; TechCrunch says the accuracy results have not been independently verified. The bank and law firm customers are unnamed. We did not contact Unsiloed.
The Bottom Line
Unsiloed raised $7.5 million, led by Felicis, with a team of seven, to sell document parsing for AI agents from one cent per page. Its own olmOCR-Bench test puts it 22 points ahead of Reducto’s agentic system, by our arithmetic.
97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The round, investors, founders, customers, pricing, benchmark results and competitors are from TechCrunch’s report of 18 November 2026, read in full; the olmOCR-Bench link is the startup’s own post as cited by TechCrunch. The 22-point gap, the $10 per 1,000 pages and the roughly 185 wrong data points are our arithmetic. We did not contact Unsiloed.
Sources: TechCrunch, 18 Nov 2026 · Unsiloed AI blog, olmOCR-Bench









