Open Source · Apache 2.0
███╗   ███╗███████╗███╗   ███╗ ██████╗ ██████╗ ██╗   ██╗
████╗ ████║██╔════╝████╗ ████║██╔═══██╗██╔══██╗╚██╗ ██╔╝
██╔████╔██║█████╗  ██╔████╔██║██║   ██║██████╔╝ ╚████╔╝
██║╚██╔╝██║██╔══╝  ██║╚██╔╝██║██║   ██║██╔══██╗  ╚██╔╝
██║ ╚═╝ ██║███████╗██║ ╚═╝ ██║╚██████╔╝██║  ██║   ██║
╚═╝     ╚═╝╚══════╝╚═╝     ╚═╝ ╚═════╝ ╚═╝  ╚═╝   ╚═╝
████████╗██╗  ██╗ █████╗ ████████╗  ████████╗██╗  ██╗██╗███╗   ██╗██╗  ██╗███████╗
╚══██╔══╝██║  ██║██╔══██╗╚══██╔══╝  ╚══██╔══╝██║  ██║██║████╗  ██║██║ ██╔╝██╔════╝
   ██║   ████████║███████║   ██║        ██║   ████████║██║██╔██╗ ██║█████╔╝ ███████╗
   ██║   ██╔══██║██╔══██║   ██║        ██║   ██╔══██║██║██║╚██╗██║██╔═██╗ ╚════██║
   ██║   ██║  ██║██║  ██║   ██║        ██║   ██║  ██║██║██║ ╚████║██║  ██╗███████║
   ╚═╝   ╚═╝  ╚═╝╚═╝  ╚═╝   ╚═╝        ╚═╝   ╚═╝  ╚═╝╚═╝╚═╝  ╚═══╝╚═╝  ╚═╝╚══════╝
███╗   ██╗ ██████╗ ████████╗      ██╗██╗   ██╗███████╗████████╗
████╗  ██║██╔═══██╗╚══██╔══╝      ██║██║   ██║██╔════╝╚══██╔══╝
██╔██╗ ██║██║   ██║   ██║         ██║██║   ██║███████╗   ██║
██║╚██╗██║██║   ██║   ██║    ██   ██║██║   ██║╚════██║   ██║
██║ ╚████║╚██████╔╝   ██║    ╚█████╔╝╚██████╔╝███████║   ██║
╚═╝  ╚═══╝ ╚═════╝    ╚═╝     ╚════╝  ╚═════╝ ╚══════╝   ╚═╝
███████╗████████╗ ██████╗ ██████╗ ███████╗███████╗██╗
██╔════╝╚══██╔══╝██╔═══██╗██╔══██╗██╔════╝██╔════╝██║
███████╗   ██║   ██║   ██║██████╔╝█████╗  ███████╗██║
╚════██║   ██║   ██║   ██║██╔══██╗██╔══╝  ╚════██║╚═╝
███████║   ██║   ╚██████╔╝██║  ██║███████╗███████║██╗
╚══════╝   ╚═╝    ╚═════╝ ╚═╝  ╚═╝╚══════╝╚══════╝╚═╝

Six-layer cognitive architecture for AI agents. User profiles, semantic search, temporal reasoning, contradiction detection - in a single SQLite file.

View on GitHub → See Benchmarks
$ pip install neuromem-core
91.8%
LoCoMo Accuracy
+30pp
vs Mem0
$0
Cloud Costs
30MB
Base Footprint
[01]

See How Much
You'll Save.

Neuromem runs locally. $0/month, forever. See how that compares to cloud-based alternatives as your usage scales.

Tier
Neuromem
$0/month
Always free. Runs locally.
Mem0
$9.30/month
Based on $0.0031/query
Supermemory
$19/month
Pro tier minimum
Annual savings with Neuromem $339
Start Saving →

Free forever. No credit card. No cloud account.

[02]

Not Just Cheaper.
Better at Everything.

Neuromem doesn't just save you money. It outperforms every alternative on accuracy, features, and privacy.

0
Neuromem Pro
vs
0
Mem0
+30.4pp more accurate
LoCoMo Accuracy91.8%61.4%
Multi-hop Reasoning89.7%37.7%
Runs Offline✓✗
Personality Profiles✓✗
Temporal Reasoning✓✗
Contradiction Detection✓✗
Open Source✓✗
Cost per Answer$0.0015$0.0031
0
Neuromem Pro
vs
0
Supermemory
+26.4pp more accurate
LoCoMo Accuracy91.8%65.4%
Multi-hop Reasoning89.7%✗
Runs Offline✓✗
Personality Profiles✓✗
Temporal Reasoning✓✗
Contradiction Detection✓✗
Open Source✓✗
Monthly Cost$0$19–$399
0
Neuromem Pro
vs
0
RAG (ChromaDB)
+5.6pp more accurate + 6 extra capabilities
LoCoMo Accuracy91.8%86.2%
Multi-hop Reasoning89.7%✗
Personality Profiles✓✗
Temporal Reasoning✓✗
Contradiction Detection✓✗
Salience Filtering✓✗
Consolidation✓✗
Predictive Coding✓✗
[03]

Six Layers.
One Chip.

Each layer is a functional component of a custom memory processor. Together they form a single orchestration engine with graceful degradation.

Neuromem 6-layer chip architecture
01 Engram Processor Core

Pre-computed entity profiles. Communication patterns, preferences, traits. The core identity processor.

~2msJSON engramsPre-computed
02 Episodic Memory Registers

FTS5 keyword search with BM25 ranking and temporal windowing. The indexed lookup table.

5-8msSQLite FTS5Trigram index
03 Semantic Tensor Core Array

256-dim Model2Vec embeddings fused with episodic via Reciprocal Rank Fusion. The vector math engine.

18-22msQwen3-EmbeddingRRF k=60
04 Salience Signal Filter

Entity overlap, recency decay, importance weighting. Many signals in, best signals out.

8-12msMulti-signalMiniLM reranker
05 Consolidation Bus Controller

Timeline construction, contradiction detection, summary generation. Memory that evolves, not accumulates.

4-6msTemporal clusteringResolved timeline
06 Predictive Output Substrate

Surprise scoring via KL-divergence. High-surprise memories boosted, redundant suppressed. Final output stage.

2-3msKL-divergenceNoise suppression
55ms 6-layer search
468ms agentic + reranker
~1.5s agentic + HyDE
[04]

We Don't Think Benchmarks
Tell the Full Story.
But We Lead Them Anyway.

LoCoMo: 1,540 questions across 10 conversations. All systems evaluated on identical pipeline. Zero errors across 12,320 total answers.

91.8%
LoCoMo Accuracy
Pro Tier
+30pp
vs Mem0
91.8% vs 61.4%
3x
Cost Efficiency
$0.0015 / correct answer
# System Accuracy Offline Multi-hop Temporal
1 Neuromem Pro 91.8% ✓ 89.7% 76.0%
2 Neuromem Base 88.2% ✓ ✓ ✓
3 RAG (ChromaDB) 86.2% ✓ ✗ ✗
4 Engram 84.5% ✗ ✗ ✗
5 BM25 Baseline 80.5% ✓ ✗ ✗
6 Supermemory 65.4% ✗ ✗ ✗
7 Mem0 61.4% ✗ 37.7% ✗
[05]

Setup in 5 Minutes.
No GPU Required.

Works with Claude Desktop, Claude Code, or any MCP-compatible client. Base tier runs on any machine with 512MB RAM.

quickstart.py 
# install - that's it, no docker, no cloud keys
# $ pip install neuromem-core

from neuromem import MemoryEngine

# one sqlite file. your entire memory system.
engine = MemoryEngine("memories.db")

# store memories - entities extracted automatically
engine.store("Josh prefers dark mode and vim keybindings")
engine.store("Meeting with Sarah about the API redesign on Tuesday")

# search - hybrid retrieval across all 6 layers
results = engine.search("what editor preferences does Josh have?")
# → "Josh prefers dark mode and vim keybindings" (score: 0.94)

# temporal queries just work
results = engine.search("what happened on Tuesday?")
# → "Meeting with Sarah about the API redesign" (score: 0.91)

# entity profiles - personality extraction
profile = engine.get_entity_profile("josh")
# → traits, preferences, communication patterns, topics
Base Tier
Model2Vec (8M params, 30MB) + MiniLM reranker. Any CPU. 512MB RAM. 88.2% accuracy.
Pro Tier
Qwen3-Embedding (600M params) + mxbai reranker. 4GB RAM. 91.8% accuracy.
■ Built With
SQLite + FTS5
Model2Vec
sqlite-vec
Cross-Encoder Reranker
MCP Protocol
Python 3.10+
Apache 2.0
SQLite + FTS5
Model2Vec
sqlite-vec
Cross-Encoder Reranker
MCP Protocol
Python 3.10+
Apache 2.0

Your Agent Needs
Its Memory.

One SQLite file · Zero cloud · Total recall

Get Started →
■ Neuromem · Open Source Memory for AI Agents