Knowledge Intelligence

Knowledge Intelligence is a multi-tenant knowledge platform. Organisations connect documents, spreadsheets, databases and web or code sources; members ask questions in a chat and get answers grounded in the sources they are allowed to see, with citations down to the page, sheet row or database query.

Around the chat sit Text-to-SQL for connected databases, agents with tools and approval gates, a live map of every source and model, an audit log, analytics and administration.

Access is enforced in two places: row-level security on every tenant table in PostgreSQL, and permission checks inside each retrieval query, with roles, grants, deny rules and expiry. A member who cannot open a document never retrieves it, so it never reaches the model.

Retrieval combines pgvector embeddings with keyword search, then reranks. Every citation is checked against the passage it points to, and when the sources do not answer the question, the assistant says so instead of guessing.

Database questions become SQL that is parsed and checked against the selected tables, columns and row filters before it runs read-only. Agents work with the asking person’s permissions; a tool marked as sensitive waits until someone else approves it, and the run then picks up where it stopped. It is built with FastAPI, PostgreSQL with pgvector, sqlglot and Next.js, and answers with Claude, OpenAI or other model providers.