Get started
Knowledge graph
Companies, people, metrics, and the page they were taken from.
What is on the graph
Documents stay documents. The graph is what we pull out of them so you can ask a question without rereading a 120-page CIM. Each fact keeps the file, the deal (if it has one), and the excerpt it came from.
Alongside that sits firm memory: claims, decisions, and investment criteria the firm has recorded about itself. Those are your statements, not the target’s.
- Entities and relationships, with the source document and page.
- Metric history, when the same figure shows up across versions of a file.
- Themes and risk clusters, when they have been extracted.
- Firm memory: criteria, convictions, and decisions you have written down.
How it gets there
Two paths. Upload a file (or import it from Drive, SharePoint, or OneDrive) and the worker chunks it, extracts it, and loads the graph. A connector such as Attio copies CRM records on a schedule after you finish the wizard. Approving access in Attio is not the import. Confirming the wizard is.
Gmail and Outlook are tools for Chat and agents. They do not dump the mailbox onto the graph. If you want a thread in the library, save it as a document.
Attio’s copy is one way, and enrichment from Attio is stored as a hint, not as a Horison fact. Suggested deals and projects wait until someone at the firm accepts them.
How you read it
Inside the app: the Knowledge graph item on the rail, and the graph tab on a deal or project. From Claude or ChatGPT: the hosted MCP server, which can list deals, search the graph, read a document, and fetch evidence. When you quote a number, keep the source with it: the document name, the deal, the page.