Collective Brain
Your company's memory — organised, permissioned, and answerable. A multi-tenant knowledge-graph app for SME teams.
Overview
A multi-tenant knowledge-graph app for small and mid-sized teams. Pour in scattered docs, chats, and meeting notes; an LLM distils them into atomic, typed knowledge nodes linked into a graph. Anyone on the team can then ask questions in plain English and get answers grounded only in what the company actually knows — every answer cited back to its source node.
Problem
Every team's knowledge lives in a dozen places: Slack threads, meeting transcripts, PDFs, and half-remembered decisions in people's heads. When someone leaves, or simply forgets, the company loses part of its memory. Generic chatbots don't fix this — they answer from the internet, not from what your company knows, and they can't respect who is allowed to see what.
Solution
Ingestion turns raw material into a clean graph of typed, linked knowledge. Question-answering retrieves only from that graph and cites its sources, so answers are checkable rather than plausible. And permissions aren't an app-layer afterthought: they're enforced in Postgres with Row-Level Security before any content reaches a model — two organisations can't see each other's anything, and that guarantee is proven by tests.
Lessons learned
The hard part of "AI for company knowledge" isn't the AI — it's trust. Citations and database-enforced permissions are what turn a demo into something a business would actually put its internal knowledge into.
Future improvements
Connectors for the places knowledge already lives (Slack, email, drive folders), and lightweight review workflows so the graph stays accurate as the company changes.