The schema library for Engrammic, a structured memory system for AI agents.
Most AI agents treat context like a scratchpad. Engrammic treats it like cognition: observations become claims, claims become facts, facts become beliefs. This library defines the types and rules that make that work.
Library for integrators. If you just want to use Engrammic memory in your agent, see engrammic-mcp (hosted) or engrammic-engine (local).
pip install engrammic-primitivesSchema types for four cognitive layers:
| Layer | What it holds | Example |
|---|---|---|
| Memory | Raw observations | "User mentioned they prefer dark mode" |
| Knowledge | Claims with evidence | "User prefers dark mode" (based on 3 mentions) |
| Wisdom | Synthesized beliefs | "Optimize for low-light viewing in this user's sessions" |
| Intelligence | Reasoning chains | Step-by-step derivation of a conclusion |
Scoring functions for promotion decisions:
from primitives.eag import combined_confidence, should_promote_r1
# When should a claim become a fact?
decision = should_promote_r1(confidence=0.8, corroboration_count=3)Transition predicates for enforcing layer rules (e.g., Knowledge requires evidence).
Protocols for storage backends (implement these to build your own store).
You're building something that stores and retrieves structured agent context, and you want compatibility with the Engrammic ecosystem.
If you just want to use Engrammic:
- engrammic-mcp for hosted or self-hosted service
- engrammic-engine for local-only (SQLite)
| Module | Purpose |
|---|---|
primitives.schema |
Node and edge type definitions |
primitives.eag |
Confidence, promotion, decay logic |
primitives.eag.transitions |
Layer transition predicates and constraints |
primitives.protocols |
Storage and lifecycle interfaces |
primitives.scoring |
Freshness and relevance formulas |
- EAG Paradigm for the full cognitive architecture spec
Apache 2.0