Glossary
Purpose: Align vocabulary for internal teams, prospects, and AI prompts.
| Term | Definition |
|---|---|
| MCP (Model Context Protocol) | Streamable HTTP interface that lets AI tools read/write memories via read_memories, write_memory, browse_memories, and the other verbs. |
| allowed_vendors | JSON array field that scopes visibility. ["*"] = all tools, otherwise the tool must be explicitly listed to read the memory. |
| shared memory | Team-level pool returned by read_team_memories, shared via share_memory only when the author opts in. |
| supersession | A memory that has been superseded by a newer version. The graph helpers expose these relationships so agents know what is stale. |
| session briefing | Structured snapshot of the user's memory state delivered to the model on connect—contains identity, topic clusters, open threads, and behavioral guidance. |
| deployment modes | hosted, isolated-hosted, self-host with tailored network, egress, and telemetry controls. |
| question bank | content/diligence/_source/question-bank.yaml – the transcript-derived dataset that feeds the FAQ, docs, and prompt copy boxes. |
| graph tools | Future MCP helpers (get_open_tickets, get_unresolved_threads, get_recent_decisions, get_superseded_decisions, get_current_state(topic)) that synthesize operational state. |
| audit trail | Logging of every read/write/auth event stored in usage_events, powered by the same SQLite/Postgres backend. |
| async diligence | The end-to-end experience (markdown + prompts + PDFs) that allows prospects to deploy their own AI to evaluate Reflect. |