Key points
- Unsiloed AI raised $7.5 million in seed funding led by Felicis, with Y Combinator, Orange Collective and Uncorrelated Ventures also participating.
- Its software turns tables, charts and complex layouts into structured data for AI agents and shows where each value came from in the source document.
- In a startup-supplied test of 542 financial documents, it reported 99.78% accuracy on about 84,000 data points; the result is not independently verified.
- Pricing starts at one cent per page; the company has seven employees and customers in finance, legal and healthcare.
Why it matters for marketers: Agents that pull answers from contracts, reports and knowledge bases are only as reliable as the extraction layer, so source tracing and accuracy claims deserve scrutiny before you buy.
Unsiloed AI, a San Francisco startup, has raised $7.5 million in seed funding led by Felicis. Y Combinator, Orange Collective, Uncorrelated Ventures and two angel investors also joined. The company builds software that breaks down complex business documents and converts their contents into structured data that AI agents can use.
The pitch centers on trust. Documents full of tables, charts and unusual layouts can trip up AI systems, and errors hurt most in fields like banking, healthcare and law. Unsiloed says its tool links each extracted value back to its place in the original document so teams can check results. CEO Aman Mishra said language models are probabilistic, so some deviation and hallucination is expected.
The company reports 99.78% accuracy across roughly 84,000 data points in a test of 542 financial documents from a customer. It also says it scored 88.0 on the olmOCR-Bench public benchmark, ahead of Reducto and Unstructured. Both sets of results come from the startup, and the customer test has not been independently verified.
For marketers and content teams, the relevance is in any workflow where agents read PDFs, contracts, reports or knowledge bases. Voice AI startup Bolna AI, for example, uses Unsiloed to ingest knowledge bases for its voice agents. The field is crowded, with Reducto, Unstructured, Microsoft, Amazon and general-purpose models all competing.
If you are evaluating document extraction for agent work, ask for tests on your own files and check whether outputs can be traced to the source. Unsiloed has seven employees and plans to hire researchers and engineers and expand sales.
Sources
This brief is our own summary. Read the full reporting at the source.
