HEXi's knowledge layer is where institutional memory lives. Every decision made, every risk flagged, every person involved — it all gets captured as connected facts that agents can search, reason over, and build on. Think of it as your organisation's long-term memory, written down and kept up to date automatically.
At the heart of this is an ontology — a shared vocabulary that defines what kinds of things exist in your world (Actors, Projects, Decisions, Risks…) and how they relate to each other (assigned to, depends on, champions…). The ontology is the map; the knowledge graph is the territory built from it. Without the ontology, facts pile up with no structure. With it, agents know exactly what they're looking at and what questions they can ask.
HEXi remembers things the way a seasoned colleague does — not just the facts, but the story behind them. Every decision made, every risk flagged, every person involved is stored as a connected piece of knowledge with a timestamp. That means HEXi can answer not only "what do we know?" but "what did we know at the time, and how has our understanding changed since?" As your team works, the graph grows quietly in the background — building an institutional memory that doesn't leave when people do.
Episode Queue
Agents propose knowledge additions via the Episode Queue rather than writing directly to the graph. This human-in-the-loop design means every piece of institutional memory is reviewed before it becomes canonical. Episodes move through states: Queued → Ingest → Verified (or Archived / Error).
Graph Visualisation
The Graph tab renders your full knowledge graph as an interactive force-directed visualisation. Nodes are coloured by entity type, edges are coloured by relationship type, and the visualisation updates dynamically as the graph grows. You can query by node count, search nodes and facts, and explore connections.
Ontology Management
The Ontology tab shows the structure of the knowledge graph itself — which entity types exist, which relationship types connect them, and how domains group related concepts. This is the schema that guides HEXi's LLM-powered entity extraction when episodes are ingested.
HEXi ships with 14 default entity types that cover the full lifecycle of a governed project. Each entity type has defined fields, a colour for graph visualisation, and a domain assignment:
All entity types are fully customisable via the Entity Types tab. Add new types, modify fields, assign colours, and group them into ontology domains.
Agents in HEXi have private workspaces where they write notes — observations, hypotheses, and heuristics they've developed through their work. These notes are personal and not visible to other agents.
When an agent decides a note has matured into genuine institutional knowledge, they curate it — framing it in terms of entities and relationships — and queue it as an episode. Once a human reviews and approves it, it flows into the knowledge graph permanently, where all agents can benefit from it.