Open Source — MIT Licensed

Your AI has amnesia.
Fix it.

Self-hosted AI memory for OpenClaw agents. Auto-captures facts, recalls context, resolves contradictions. Runs 100% locally for ~$1/month.

Get Started → View Source

# Install in 3 commands
$ docker run -d --name qdrant -p 6333:6333 qdrant/qdrant
$ pip install mem0ai fastapi uvicorn
$ ./install.sh

🧠 Mnemonic is running. Your AI will now remember everything.

# What it captures automatically:
→ "User moved from NYC to SF" # supersedes old location
→ "Prefers Python over Java" # stored as preference
→ "Meeting tomorrow at 3pm" # temporal, auto-expires

Features

Memory that actually works

Not a vector dump. An intelligent memory layer that thinks about what to remember.

🔄

Auto-Capture

Extracts facts, preferences, and decisions from every conversation automatically. No manual tagging needed.

🔍

Auto-Recall

Before each AI turn, semantically searches memories and injects relevant context. Your AI always has the full picture.

⚡

Contradiction Resolution

"Moved to SF" automatically supersedes "Lives in NYC." Old facts get updated, not duplicated.

👤

User Profiles

Builds persistent profiles from conversations. View everything your AI knows about each user.

🏠

100% Self-Hosted

Your data never leaves your server. Qdrant + mem0 + OpenAI embeddings — all running locally.

🔧

OpenClaw Native

Drop-in plugin. Registers as memory slot, provides tools (store/recall/forget/profile), works instantly.

How It Works

Five steps, fully automatic

You say something

"My company just landed a new client in Dubai."

Auto-Recall triggers

Plugin searches mem0 for relevant context — surfaces what it already knows about your business.

AI responds with context

Your agent sees all relevant memories, responds intelligently without you repeating yourself.

Auto-Capture extracts facts

An LLM processes the conversation and extracts: "Company landed new client in Dubai."

Memory updates intelligently

New facts get added. Contradictions get resolved. Duplicates get merged. Your memory stays clean.

OpenClaw Gateway └── openclaw-mem0 plugin (TypeScript) ├── before_agent_start → search mem0 → inject context ├── agent_end → extract last turn → feed to mem0 └── tools: mem0_store, mem0_recall, mem0_forget, mem0_profile │ ▼ mem0 API Server (Python/FastAPI, localhost:8765) │ ├── LLM Extraction (OpenAI GPT — fact extraction + contradiction resolution) └── Qdrant Vector DB (Docker, localhost:6333)

Compare

Why Mnemonic?

Feature	Mnemonic	Supermemory	mem0 Cloud	LanceDB (built-in)
Self-hosted	✓	✗ Cloud only	✗ Cloud only	✓
LLM fact extraction	✓	✓	✓	✗
Contradiction resolution	✓	✓	✓	✗
User profiles	✓	✓	✓	✗
Auto-recall + capture	✓	✓	✓	✓
Monthly cost	~$1	$20+	$20+	~$0
Setup time	10 min	2 min	2 min	1 min
Data privacy	✓ Your server	✗ Their cloud	✗ Their cloud	✓ Your server

Qdrant + mem0

Self-hosted, open source

OpenAI API

~$1

Embeddings + extraction/month

Total RAM

130MB

Qdrant ~30MB + API ~100MB

Your AI has amnesia.Fix it.