What is Engrammic?
Most agent memory is a vector database. You put text in, run similarity search, get text out. That works until your agent needs to know whether something is a fact or a guess, or whether it was true last week but not today. Engrammic tracks that distinction.
Four layers
Section titled “Four layers”| Layer | What goes in | How long it lives |
|---|---|---|
| Memory | Observations | Decays over time |
| Knowledge | Claims with sources | Until superseded |
| Wisdom | Beliefs built from facts | Until revised |
| Intelligence | Reasoning | Session only |
Each layer has different rules. Memory will decay on a schedule you choose. Knowledge requires a source to be stored. Wisdom must cite the facts it rests on. Intelligence is scratch space that disappears when the session ends.
What you get back
Section titled “What you get back”When you query, you don’t get “similar text”. You get structured context: what kind of thing it is (observation, fact, belief), where it came from, what it replaced, and how confident you should be.
MCP interface
Section titled “MCP interface”Engrammic exposes tools via the Model Context Protocol. It works with Claude Code, Cursor, Windsurf, or anything that speaks MCP.
- Cognitive Layers for the full breakdown
- Quickstart to start building