Our stack, behind your endpoint
The document processing and inference we built for our own products and our clients' systems, exposed as APIs you can call and MCP servers your agents can pick up on their own — in real time or in batch.
Two ways in
Call it yourself, or let an agent call it
- REST APIA conventional HTTP interface for application code — predictable request and response shapes, versioned, with the same semantics whether you call it once or a million times.
- MCP serverThe same capability published over the Model Context Protocol, so an AI agent can discover the tool, read its schema, and call it directly without a bespoke integration.
What it does
Capabilities, not a model zoo
Each one is something we already run in production somewhere — in one of our products, or inside a client's system.
Data processing
Documents in, structured data out
The extraction stack behind our own products, exposed as an endpoint. Send a PDF, a scan, or a pile of supplier filings and get back structured data with the document's shape intact — tables reconstructed cell by cell, merged headers preserved, multi-column layouts kept in reading order, and every value traceable to the page it came from.
- Parse documents into structured JSON
- Reconstruct tables with merged cells and spanning headers
- OCR scanned and photographed pages
- Extract named fields against a schema you define
- Return provenance for every extracted value
- REST API
- MCP server
- Batch
Model inference
Predictions in real time, or a million at a time
Run our models — and models we build for you — behind one interface. Single calls return in-line for interactive use; large jobs go to a queue that streams results back as they finish. The same model, the same weights, the same answer either way, so what you validate in batch is what you serve in production.
- Synchronous prediction for interactive requests
- Batch jobs over large datasets, with results streamed back
- Job status, progress, and retry on partial failure
- Versioned models, so a result can be reproduced later
- Confidence scores alongside every prediction
- REST API
- MCP server
- Realtime
- Batch
Agent tooling
Tools your agent can pick up on its own
Every capability is published as an MCP server, so an agent can discover it, read its schema, and call it without glue code. Point Claude, or any MCP-capable client, at the endpoint and document processing and inference become tools it already knows how to use.
- MCP servers for every platform capability
- Typed tool schemas, discovered at connect time
- Works with any MCP-capable client
- Scoped keys, so an agent only gets the tools you grant it
- MCP server
Running it
The unglamorous half that decides whether you ship
Keys you can scope and rotate, usage you can see, limits that fail predictably rather than silently, and deployment that can sit in our infrastructure or inside yours when the data cannot leave.
- Scoped API keys with rotation
- Usage and quota visibility
- Predictable rate limiting with backpressure
- Self-hosted deployment where data residency requires it
- REST API
- MCP server
Early access
Want a key?
The platform is opening up gradually, and we would rather start with a real workload than a signup form. Tell us what you are building and we will get you access.