Engram Alpha — documentation
Everything here is for users: what each part does, how to use it, and what to do when something is off. Start with Getting started if Engram isn't installed yet.
| Page | What it covers |
|---|---|
| Getting started | Install, wire your assistants, the per-harness support matrix, first session, updating |
| The memory model | The nine node types, seven edge verbs, durability, capture modes |
| Customization | Reshaping the ontology, presets, custom fields, tuning trust/decay, version tracking, handoff notes |
| Trust & decay | How trust is computed, what moves it, pins (and what to do when a pinned note is wrong), stale knowledge |
| The pane | The graph UI: the GPU canvas and its layouts, the timeline feed, Tasks, the session-history screen, tags and filters, editing, review, audit, System |
| Recall & capture | The session brief, search, silent writes and their verdicts, what gets redacted, code refs |
| Conflicts & Checkup | Suspected conflicts, judgments, claim checks, graph sweeps |
| Multi-project memory | The machine core, the registry, the home graph, promotion |
| Runtime architecture | The processes and their roles, discovery, MCP bridging, idle unload, lifecycle |
| Storage & TepinDB | The .tepin file, migrating from SQLite, at-rest encryption, the npx tepindb flow |
| Local models | The cortex (embeddings, reranker, NLI), choosing models, staying offline |
| Troubleshooting | doctor, status, stop, daemons and ports, locked stores, common fixes |
Security posture — threat model, measures in place, known gaps — lives in SECURITY.md at the repository root. The roadmap lives in the project's own memory graph — dogfooding is the spec.