- The Brain
- The typed model of a business built from connected sources: entities, relationships, processes, governing rules, and the vocabulary the team actually uses. Called the domain model in technical docs. Everything else in BrainBox reads from or writes to it.
- Connector
- A source feeding the Brain — a database schema, a set of documents, an event stream, or another MCP server. New connectors enrich the existing model; they never rebuild it.
- Intelligence Trace
- BrainBox’s interpretation of a single event you send it — a meeting note shared from Claude or ChatGPT, a forwarded email, a Slack thread tagged @brainbox. A trace records what the event means (accounts touched, process in play, signal surfaced), not a copy of the raw content.
- The Wiki
- An auto-maintained page about an account, a recurring pattern, or a person. Written by the Brain from accumulated traces and rewritten as new ones arrive, with full change history.
- Learning
- A correction or insight surfaced from real usage, reviewed by a human, then folded back into the Brain — the mechanism by which the model sharpens over time.
- Grounding
- Answering from the Brain rather than from raw retrieval: an agent resolves entities, definitions, and rules against the model — with source attribution — before it responds.
- Multi-tenant brain
- An isolated brain provisioned per end customer, for teams embedding BrainBox in their own product. The tenant is selected at query time; boundaries are enforced by the platform.