documents and knowledge management
What you upload stops being an opaque file: it becomes citable blocks, queryable rows or nodes with relationships. Here is the full path, from raw file to a deliverable you can defend.
PDFs, scans, notes and pages land as an editable document split into blocks by section. Each block is embedded on its own, so the answer cites the exact paragraph — not “the file”.
Upload a table or connect a read-only SQL engine. The agent queries the rows, aggregates and compares — and the table is indexed as readable rows, not raw JSON, so semantic search finds it too.
When what matters is the relationship rather than the paragraph — who depends on whom, what causes what — the document is stored as nodes and edges. The agent queries by neighbourhood: it pulls the part of the graph it needs, not the whole thing.
All three types live in the same knowledge base, share its permissions and leave through the same endpoint. You never pick a format up front — you add the one you need when you need it.
the loop
This is not a chat with a PDF attached. It is the whole loop — from the raw file nobody can search to the deliverable you can actually defend.
Drop in a PDF, a scan or a table and Doctiling rebuilds it: page geometry so reading order survives, tables reassembled as tables, images pushed through OCR in the background, and cuts made at section boundaries — not every 8,000 characters. What comes out is an editable document, not an opaque attachment.
A firm loads a 60-page scanned contract their own search engine could not even open.
It comes out as a document with readable blocks and tables — and from that moment it is citable material.
Every block is embedded in 768 dimensions and indexed on its own. When you ask, the agent retrieves blocks — not whole documents — and answers with a citation to the exact one. If the answer is not in your sources, it says so.
“Which contracts renew this quarter, and at what penalty?” — across forty agreements at once.
Back comes a table where every row links to the clause it came from: verifying takes a click, not an afternoon.
Mirror view splits the screen: your document on one side, the analyses on the other. Summary, entities, contradictions, comparison table, mind map, questions, critique — plus five lenses for database documents. Each one runs on the indexed content and stacks without erasing the last.
A consultancy uploads three market reports and asks where they contradict each other before the client call.
The mirror shows every disagreement next to the clashing paragraphs — ready to take into the deck.
Save any analysis as a variant: an editable document that keeps its trail back to the sources, ready to keep working on and to share with roles. Or expose the knowledge base as an MCP endpoint with a scoped, revocable token, so it answers from the tools your team already works in.
A research group shares its literature base and queries it from their own MCP-compatible client.
One indexed corpus, many ways to work — and private analyses still belong to whoever made them.
what the agent returns
Answers render inside the document. Charts, diagrams, mind maps and comparison tables come out of the indexed content — with the trail back to the source intact — and save as an editable variant of your own.
Bars, lines and distributions computed over the real rows of your table documents, not over an invented summary.
The agent writes the diagram and it renders in the block: processes, architectures, sequences. Editable afterwards, like any other block.
A long document's structure unfolded as a navigable map, so you can see what it covers before reading all of it.
Several documents set against each other row by row, each cell linked to the block it came from: verifying takes one click.
Plus summary, entities, contradictions, questions, critique and study cards — twelve one-click lenses, no prompt writing.
the studio, live
Real captures of the studio — no filler mockups.
The studio: your knowledge bases and documents in one place, indexed and ready.
Ask, and every answer cites the exact block of your document — click the citation to jump to the source.
Mirror view: the document on one side, the agent's analysis — grounded in its content — on the other.
Ready-made analysis lenses: summary, entities, mind map, contradictions, questions and more — one click.
The same studio in your pocket — install it as an app and query your documents anywhere.
Stop planning your RAG. Start operating your knowledge bases.
Request your free access with your email — we review it and send the invitation shortly. No credit card. Already invited? Sign in and your first knowledge base is one click away.