In-Memory Intelligence for AI Agent Fleets

Like Redis, but for AI agents that think, learn, and remember together

3,774

Memories Stored

20+

Active Agents

<10ms

Search Time

# Start the memory server
memclawz serve --port 3500

# Store a memory
curl -X POST localhost:3500/api/v1/add \
  -H "Content-Type: application/json" \
  -d '{"content": "BTC broke 100k resistance",
      "agent_id": "trader",
      "memory_type": "event"}'

# Search memories (hybrid: vector + BM25)
curl "localhost:3500/api/v1/search?q=bitcoin+price&limit=5"
                        

Powerful Features

Everything you need to build intelligent, memory-enabled AI agent fleets

🚌

Shared Memory Bus

20+ agents read/write to a single memory layer via REST API. Perfect for fleet coordination and knowledge sharing.

🔍

Hybrid Search

Vector similarity (Qdrant) + BM25 keyword matching + composite scoring. Find exactly what you need, fast.

🕸️

Knowledge Graph

Neo4j/Graphiti for relationship mapping between concepts. See how memories connect and influence each other.

✨

Auto-Enrichment

Plain text in → Gemini Flash-Lite auto-generates type, weight, title, summary, tags, RDF triples.

⏰

Temporal Validity

Memories expire based on type: prices=hours, projects=weeks, identity=permanent. Smart lifecycle management.

🔄

8-State Lifecycle

active → confirmed → outdated → archived → contradicted → merged → superseded → deleted

🌐

Fleet Federation

Push/pull/sync memories across distributed agent clusters. Scale across regions and environments.

🔌

MCP Server

Works with Claude Desktop, Cursor, and any MCP client. Universal compatibility for AI tools.

How It Works

Modern architecture built for scale and performance

AI Agents

→

REST API

→

MemClawz Core

→

Qdrant + Neo4j

Vector Search: Qdrant for semantic similarity

Knowledge Graph: Neo4j/Graphiti for relationships

Memory Engine: Mem0 for intelligent storage

API Layer: RESTful interface for all agents

Quick Start

Get up and running in minutes

Install MemClawz

Simple pip install to get started with the core memory system

Start the Server

Launch the memory server on your preferred port

Store Memories

Use the REST API to add memories from your agents

Search & Learn

Query the hybrid search system and let your agents learn from shared memory

# Install MemClawz
pip install memclawz

# Start the memory server
memclawz serve --port 3500

# Store a memory
curl -X POST localhost:3500/api/v1/add \
  -H "Content-Type: application/json" \
  -d '{"content": "BTC broke 100k resistance",
      "agent_id": "trader",
      "memory_type": "event"}'

# Search memories (hybrid: vector + BM25)
curl "localhost:3500/api/v1/search?q=bitcoin+price&limit=5"

# View memory explorer
open http://localhost:3500/explorer
                    

Reinforcement Learning from Memory

The key differentiator: agents that learn from fleet experience

Learn from past decisions and outcomes
Identify contradictions and resolve conflicts
Sleep-time reflection reviews experiences
Extract patterns from fleet behavior
Adapt strategies based on collective memory
Share insights across agent clusters
Continuous improvement through experience

Memory → Learning → Improvement

# Agent stores decision outcome
POST /api/v1/add
{
  "content": "Strategy X failed, lost 2%",
  "memory_type": "outcome",
  "agent_id": "trader_bot"
}

# Fleet learns from failure
GET /api/v1/search?q=strategy+outcome

# Better decisions next time
🧠 Collective Intelligence Active
                    

Join the Community

Open source, welcoming, and growing fast

⭐

Star on GitHub

Show your support and stay updated with the latest releases

Star Repository

💬

Join Discord

Chat with other developers, get help, and share your projects

Join Server

🚀

Contribute

Help improve MemClawz with code, docs, or feature ideas

Get Started