Unsiloed AI
by Unsiloed AI
Document parsing APIs that turn PDFs, slides, scans and spreadsheets into cited JSON and Markdown for AI agents.
Unsiloed AI is a document ingestion API that converts PDFs, scans, slide decks, Word files, spreadsheets, charts and images into structured Markdown and JSON for language models and agents. It is built for engineering teams in finance, legal and healthcare who would otherwise spend months building and maintaining their own parsing pipeline.
Unsiloed AI is a document ingestion API from a San Francisco startup founded in 2025 by Aman Mishra (CEO) and Adnan Abbas (CTO), who went through Y Combinator's Fall 2025 batch. It converts PDFs, scans, slide decks, Word files, spreadsheets, tables, charts and images into Markdown and JSON that language models and agents can consume. The API has four operations. Parse returns ordered Markdown and HTML chunks with segment types, bounding boxes, OCR text and confidence scores. Extract pulls the fields you define in a JSON Schema and returns each value with a confidence score and a page citation. Classify assigns a document or page to categories you supply, and Split breaks a bundled PDF into one file per document type. Jobs run asynchronously: you submit a document URL, get a job ID back and poll for the result. The company says its models process text and image data in two separate streams, that heat map based chunking keeps related content together across pages, and that figures, numbers and charts get extra compute. In May 2026 the founders reported an 88.0 strict pass rate for Unsiloed Parser v3.1 on Allen AI's olmOCR-Bench across 1,403 PDFs. This is a self-reported result, and TechCrunch set it against 66.0 for Reducto's agentic system and 39.9 for Unstructured. On November 18, 2026, the company announced a $7.5M seed round led by Felicis. It had seven employees at the time. Named and described customers include an unnamed bank, a law firm that runs the software on its own servers, and the voice agent company Bolna AI. Integrations include a remote MCP server for Claude, Anthropic tool-use schemas and community n8n nodes. Per-page list prices are published and run from $0.02 a page on pay as you go to $0.01 a page on the Growth plan.
A platform or AI engineering lead who feeds contracts, filings or clinical documents into RAG or agent workflows and needs extracted values tied to a page and bounding box so reviewers can check them.
Schema-defined fields extracted from messy documents, each with a confidence score and a page citation, at a published per-page price of one to two cents.
At a Glance
- Category
- Data & Analytics
- Pricing
- Usage-based, Subscription, Contact for pricing
- Target Market
- CTOs, Enterprise Developers, Data Engineers, AI Platform Teams
- Deployment
- API-based, Cloud-first, Self-hosted
- Founded
- 2025
- Headquarters
- San Francisco, USA
- Team Size
- 1-10
Key Features
- ✓Parse endpoint
Returns a document as ordered Markdown and HTML chunks with segment types, bounding boxes, OCR text and confidence scores, so downstream retrieval keeps the original layout.
- ✓Schema-based Extract
Pulls only the fields you define in a JSON Schema and returns each value typed, with a confidence score and page citation you can audit.
- ✓Classify and Split
Assigns pages or documents to categories you supply and splits a merged PDF bundle into one file per document type, which removes manual sorting from intake queues.
- ✓Layout-aware chunking
Heat map based chunking keeps related content together across page breaks, and figures, numbers and charts get extra compute where extraction errors cost the most.
- ✓MCP server and Claude tool schemas
Lets you add Unsiloed to Claude.ai or Claude Desktop as a remote MCP connector over OAuth, or call it from the Claude API with drop-in tool-use schemas.
- ✓Self-hosted and air-gapped deployment
The company says the parser can run fully on premise in air-gapped environments, and one law firm customer runs it on its own servers.
Capabilities
Use Cases
- •Financial document extraction
A bank pulls figures from statements and filings into typed JSON, and reviewers check each value against its cited page and confidence score.
- •Contract and case file review
A law firm parses contracts and old case files on its own servers so privileged documents never reach a third-party cloud.
- •Knowledge base ingestion for voice agents
A voice agent company ingests customer knowledge bases into structured chunks so its agents answer from the right section of the source document.
- •Mailroom and intake automation
An operations team sends merged PDF bundles through Split and Classify, which separates invoices, IDs and forms into routed files without manual sorting.
- •Agent-driven document review
A developer pairs Unsiloed extraction with a downstream checker model so low-confidence or implausible values get flagged for human review before they reach production systems.
Ideal For
Best For
- ✓Financial services teams extracting tables and figures from filings, statements and loan packets into typed JSON
- ✓Legal teams processing contracts and archived case files that must stay on their own servers
- ✓RAG pipelines that need layout-aware chunks with bounding boxes rather than flat OCR text
- ✓Back-office workflows that receive merged PDF bundles and need them split and classified by document type
- ✓Agent builders who want document parsing available as an MCP tool inside Claude
Not Ideal For
- ✗Buyers whose procurement requires a published SOC 2 report or HIPAA BAA, since the vendor site makes no compliance claims you can verify today
- ✗Teams that want a mature vendor with a large support organisation, because the company had seven employees at its November 2026 seed round
- ✗Developers who expect official SDKs in several languages, since the docs list REST endpoints and an MCP server but no SDK packages
Deployment
Market Analysis
Pros
- ✓Published per-page pricing from $0.01 to $0.02 makes cost modelling simple before you talk to sales
- ✓Field-level confidence scores and page citations support human review in regulated workflows
- ✓On-premise and air-gapped deployment is available, and a law firm customer runs it on its own servers
- ✓Benchmark harness is open source (Apache-2.0), so you can test accuracy claims against your own documents
Cons
- ✗Headline accuracy figures are company-reported, and no independent benchmark replication was found
- ✗Onboarding was not fully self-serve as of May 2026, according to a founder reply on Hacker News
- ✗The vendor site states no SOC 2, HIPAA or ISO 27001 attestations, which will slow procurement in regulated industries
- ✗Very small team (seven people at the seed round) and no user reviews on G2 or Capterra yet
Pricing
Pay per page
$0.02/page
- ✓Parse, Extract, Classify and Split each $0.02 per page
- ✓No subscription
- ✓Minimum purchase $0.01
Credits
From $5 (prepaid credits)
- ✓$0.0125 per page across Parse, Extract, Classify and Split
- ✓Minimum purchase $5
Standard
$250/mo
- ✓20,000 pages included ($250 credit)
- ✓$0.0125 per page across all services
Growth
$750/mo
- ✓75,000 pages included ($750 credit)
- ✓$0.01 per page across all services
Enterprise / self-hosted
Contact for pricing
- ✓On-premise and air-gapped deployment described by the vendor
- ✓Terms negotiated directly
Billing is per page. One PDF page, image or slide counts as one page, and so does each 5,000 by 5,000 cell block of a spreadsheet. List prices are published in the vendor's pricing manifest: $0.02 a page pay as you go, $0.0125 on credits or Standard ($250 a month for 20,000 pages), and $0.01 on Growth ($750 a month for 75,000 pages). Parse, Extract, Classify and Split all cost the same. Self-hosted deployment is not priced publicly, and as of May 2026 the founders said onboarding was not fully self-serve.
Security & Compliance
Connect
Sources
This page was written from 6 sources, 5 on domains other than unsiloed.ai.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Nebuly
AI ROI platform that measures what employee and customer AI agents are worth to the business
Edelbyte
One engine for search, analytics, vector, graph and geospatial queries, built for AI agents that query company data constantly
Revefi
AI FinOps and observability agents that cut LLM, Snowflake, Databricks and BigQuery spend
Corvic AI
Agentic data engine that turns PDFs, images, sensor logs and tables into cited answers and pipelines