Why Engrammic?
Memory that does not rot, does not hallucinate, and can show its work
The Problem
AI agents are stateless. Every conversation starts from zero. Your agent learns your preferences, discovers project patterns, makes decisions, then forgets everything when the session ends.
You can stuff context into system prompts, but that does not scale. You can use RAG, but retrieval without structure returns noise. What you need is memory that does not rot between sessions, does not hallucinate its own history, and can show its work when asked.
How Engrammic Works
Engrammic is built on a simple idea: memory should know why it believes things.
Every piece of knowledge tracks its provenance:
- Where did this come from? (a file, a conversation, another belief)
- How confident are we? (observation vs. verified fact vs. conclusion)
- What depends on this? (beliefs built on top of this knowledge)
When your agent recalls something, it can trace the evidence chain. When evidence changes, dependent beliefs can be reconsidered. That is the difference: not just storage, but structured, traceable knowledge.
The Layers
Engrammic organizes knowledge into four cognitive layers:
Memory: Raw observations. "User said they prefer TypeScript." No evidence required, low commitment.
Knowledge: Verified facts with evidence. "This project uses pnpm workspaces (evidence: package.json, pnpm-workspace.yaml)." Higher confidence, traceable.
Wisdom: Conclusions drawn from knowledge. "This team prefers monorepo patterns (based on: workspace structure, shared configs, team discussion)." Built on other nodes.
Intelligence: Active reasoning. Hypotheses, chains of thought, decisions in progress. Temporary until committed.
Why This Matters
Most memory systems are bags of vectors. Engrammic is a knowledge graph with structure and accountability.
- Traceability: Ask "why do you believe that?" and get an answer
- Consistency: Contradictions surface because beliefs track their foundations
- Collaboration: Multiple agents share structured knowledge, not raw text
- Evolution: Knowledge updates propagate to dependent beliefs