Skip to content

Engramma Memory

A composable memory engine for AI systems that learns, adapts, and reasons — not just retrieves.

Why Engramma?

Vector databases find the nearest neighbor. That's it.

Engramma is a memory engine — it retrieves, composes, generalizes, and adapts:

Feature Vector DBs Engramma Local Engramma Cloud
Exact recall Yes Yes Yes
Native composition No Yes Yes (weighted)
Soft generalization No Yes Yes
Adaptive routing No Confidence-based Active Inference
Causal reasoning No No Yes
Safety regimes No No Yes
Text interface No No Yes
XAI explainability No No Yes

Quick Example

from engramma_memory import EngrammaMemory
import numpy as np

mem = EngrammaMemory(dim=256, backend="local")

# Store embeddings
mem.store(key=embedding_a, value=embedding_a)
mem.store(key=embedding_b, value=embedding_b)

# Native composition — no manual blending
blend = mem.compose([embedding_a, embedding_b])

One Line to Production

# Switch from local to cloud — same API, unlimited power
mem = EngrammaMemory(dim=256, backend="cloud", api_key="nx_live_...")

# Now you have access to 40+ cloud-exclusive features:
mem.explain(query)                # XAI: why was this returned?
mem.consolidate()                 # Sleep/Wake memory consolidation
mem.get_causal_graph()            # Discover causal structure
mem.query_text("user preferences") # Natural language queries
mem.get_current_regime()          # Safety regime monitoring

Architecture

Query ──┬──> [Exact Memory]     ──> Perfect recall (kNN)
        ├──> [Energy Memory]    ──> Soft generalization (Hopfield)
        └──> [Multi-Head Attn]  ──> Native composition
                    │
            ConfidenceRouter ──> Best result       ← Local
            phi_B + EFE     ──> Optimal routing   ← Cloud

Next Steps