Eengramma
FeaturesHow It WorksPricingBlog
Join Waitlist
Now in Private Beta

Memory that thinks
for your AI agents

Vector databases retrieve. Engramma composes, generalizes, and reasons. Give your AI systems the memory they deserve.

Join the WaitlistView on GitHub
$ pip install engramma-memory

The Problem

Your AI has amnesia

Current memory solutions are just fancy search engines. They find the nearest vector — that's it. No composition, no reasoning, no adaptation.

Vector DBs

Find nearest neighbor. Return it. Done. Can't blend, can't reason, can't adapt.

RAG Pipelines

Retrieve chunks, stuff them in a prompt. No composition, no generalization.

Engramma

Native composition via multi-head attention. Soft generalization. Causal reasoning. Real memory.

Features

Everything a memory engine should be

Built on neuroscience-inspired architecture. Three specialized pathways working together.

Native Composition

Blend multiple memories into one coherent answer. Multi-head attention attends to different patterns simultaneously — something vector DBs physically cannot do.

Smart Routing

Confidence-based router learns which pathway handles which query. Exact recall, soft generalization, or composition — automatically selected.

Causal Reasoning

Discover causal structure in your data. Predict interventions. Block confounded compositions. Real reasoning, not pattern matching.

Safety Regimes

Three operating regimes with anomaly detection. Auto-blocks risky operations on out-of-distribution data. Sleep/Wake consolidation.

Text Interface

Store and query with natural language — no external embedding model needed. Built-in HDC tokenizer handles it all.

Explainability

Understand WHY a result was returned. Full attention maps, pathway traces, confidence breakdowns, and XAI dashboards.

Architecture

Three pathways. One answer.

Inspired by complementary learning systems in the brain.

1

Exact Memory

Perfect recall via kNN with importance-based eviction

2

Energy Memory

Soft generalization via Hopfield network dynamics

3

Multi-Head Attention

Native composition — each head attends to different patterns

Confidence Router

One line to production

main.py
from engramma_memory import EngrammaMemory

# Local: free, open source, instant
mem = EngrammaMemory(dim=256, backend="local")

# Cloud: unlimited, persistent, intelligent
mem = EngrammaMemory(dim=256, backend="cloud", api_key="nx_live_...")

# Same API. Same code. No migration needed.
mem.store(key=embedding, value=data)
result = mem.compose([key_a, key_b])  # Native composition

Comparison

Engramma vs. the rest

FeatureVector DBsEngramma LocalEngramma Cloud
Exact recall✓✓✓
Native composition—✓✓
Soft generalization—✓✓
Causal reasoning——✓
Anomaly detection——✓
Temporal prediction——✓
Explainability (XAI)——✓
Text interface——✓
Zero dependencies—✓✓

Pricing

Start free. Scale when ready.

Local is open source forever. Cloud unlocks the full potential.

Local

Free

Open source, MIT license

  • ✓Up to 1,000 patterns
  • ✓3-pathway architecture
  • ✓Native composition (equal weights)
  • ✓Sub-ms latency
  • ✓Zero dependencies (numpy)
pip install engramma-memory
BETA

Cloud

From $29/mo

Everything in Local, plus:

  • ✓Unlimited patterns
  • ✓43 premium capabilities
  • ✓Causal reasoning & safety regimes
  • ✓Text interface (HDC tokenizer)
  • ✓Full XAI & explainability
  • ✓Tiered storage (hot/warm/cold)
  • ✓Async support + 99.9% SLA
Join Waitlist

Be first in line

Join the private beta. Limited spots. We'll notify you when it's your turn.

No spam. Unsubscribe anytime.