Agent Memory — Shared Memory for AI Agents

A lightweight, zero-auth REST API that lets AI agents share memory across machines. SQLite-persistent — memories survive server restarts.

One agent remembers. All agents benefit. No cloud, no tokens, no complexity.

How it works

Agent A (laptop)  ──POST /memory/agent-a──▶  ┌──────────────────┐
Agent B (phone)   ──GET  /memory/agent-a──▶  │  Memory Server   │
Agent C (cloud)   ──GET  /memory/agent-b──▶  │  (any machine)   │
                                              └──────────────────┘

Each agent has its own namespace (/memory/<agent_id>). Keys are simple strings. Values are text.

API

Method	Endpoint	Description
POST	/memory/<agent_id>	Store {"key":"...", "value":"..."}
GET	/memory/<agent_id>/<key>	Retrieve a single value
GET	/memory/<agent_id>	List all memories for an agent
DELETE	/memory/<agent_id>/<key>	Delete a memory

No authentication. No API keys. Designed for private networks.

Quick Start

1. Run the server

pip install flask
python server.py
# → Listening on http://0.0.0.0:9736

Configuration :

• AGENT_MEMORY_DB : chemin du fichier SQLite (défaut : ./memory.db)
• PORT : port d'écoute (défaut : 9736)

2. Use from any agent

# Store
curl -X POST http://<server-ip>:9736/memory/my-agent \
  -H "Content-Type: application/json" \
  -d '{"key":"greeting","value":"hello from laptop"}'

# Retrieve
curl http://<server-ip>:9736/memory/my-agent/greeting
# → {"key":"greeting","value":"hello from laptop"}

# List all
curl http://<server-ip>:9736/memory/my-agent

3. Python client (optional)

from client import AgentMemory

mem = AgentMemory("http://<server-ip>:9736", "my-agent")
mem.set("favorite_color", "blue")
print(mem.get("favorite_color"))  # blue
print(mem.list())                  # all keys
mem.delete("favorite_color")

Use cases

• Cross-session memory — agents remember facts across restarts
• Multi-agent coordination — agents share discovered knowledge
• Configuration hub — store API keys, preferences, environment details
• Phone ↔ Laptop sync — your phone agent knows what your desktop agent learned

Why not a database?

• Zero setup — no PostgreSQL, no Redis, no credentials
• Human-readable — curl it, debug it, inspect it
• Single file — server.py is ~80 lines
• Network-native — works over LAN, VPN, or Tailscale

Security

This server has no authentication. Run it on a trusted network only (LAN, VPN, or overlay network). Do not expose it to the public internet.

License

MIT