Agent Magnet

Self-learning memory infrastructure for AI products. It learns from what users do — not what they say.

Installation

pip install agent-magnet

Two Integration Modes

1. MCP Server (Free, Self-Hosted)

Run on your own infrastructure. You control your data.

Add to your MCP config (Claude Desktop / Cursor / any MCP client):

{
  "mcpServers": {
    "agent-magnet": {
      "command": "agent-magnet-mcp",
      "env": {
        "MAGNET_REDIS_URL": "your_redis_url",
        "MAGNET_OPENAI_KEY": "your_openai_key"
      }
    }
  }
}

Tools available:

• get_profile — get learned memory profile for a user
• inject_memory — get memory injection string for system prompt
• add_signal — record a behavioral signal
• get_cold_start — get onboarding profile for new users
• get_team_profile — get shared team memory (requires Redis)
• get_merged_injection — merged user + team + org memory injection
• get_project_memory — per-user breakdown of what was learned in a project
• share_to_team — explicitly share a personal preference to team memory
• forget_team — remove a preference from team memory
• add_team_signal — record a signal directly to team scope

2. Proxy (Hosted, Dashboard included)

Change one line. We handle the infrastructure.

from openai import OpenAI

client = OpenAI(
    api_key="mg_sk_...",
    base_url="https://magnet-gateway.onrender.com/v1",
    default_headers={"x-session-id": "user_123"}
)

Get your API key: agentmagnet.app

Team Memory

Memory works for individual users out of the box. To share memory across a team, add Redis and a shared MAGNET_TEAM_ID.

Solo (local SQLite, no Redis needed):

agent-magnet init
# Use local storage? Y
# Team ID: (press Enter to skip)

Each person's preferences are stored privately on their machine.

Team (shared Redis):

agent-magnet init
# Use local storage? N
# Redis URL: redis://your-redis-host:6379
# Team ID: acme-eng

All team members point to the same Redis. Memory is scoped per-user but team-level insights are available.

What team memory gives you:

get_project_memory(project_id="acme-app")
# →
# {
#   "contributors": {
#     "ahmet": {"prefers": ["short responses", "Turkish"], "watch_out": ["never use em-dashes"]},
#     "ayse":  {"prefers": ["detailed explanations"], "dislikes": ["bullet lists"]}
#   },
#   "team_shared": {
#     "prefers": ["short responses"],   ← promoted because 2+ users share it
#     "watch_out": []
#   }
# }

Explicitly share a preference to team:

share_to_team(user_id="ahmet", fact_or_subject="short responses", team_id="acme-eng")

Team memory requires Redis. If you try to use team tools in local mode you'll get:

Team memory requires shared storage. Set MAGNET_REDIS_URL for all team members to use the same Redis instance.

How It Learns

Magnet observes behavioral signals — corrections, rejections, implicit patterns — and builds a living profile per user. No configuration required.

Three memory layers:

• Behavioral (Redis) — real-time, every request
• Episodic (Qdrant) — semantic recall when relevant
• Knowledge (Neo4j) — long-term entity relationships

Free vs Premium

Agent Magnet is fully usable without an account. The free tier is not a demo — it's the real thing.

Free (no account needed)

• Local SQLite memory — no Redis, no external services
• Single-user behavioral memory (preferences, corrections, forgetting)
• Cross-tool identity — same memory across Claude Code, Cursor, any MCP client
• Context compression (compress_context, retrieve_original)
• Bring your own OpenAI key (BYOK)

Premium (API key from agentmagnet.app)

• Team memory — share learned preferences across a team with a shared Redis
• Hosted storage — Magnet-managed Redis, no infra to run
• Compression analytics (compression_stats)
• Priority support

To enable premium features, set MAGNET_API_KEY=mg_sk_... in your environment or pass it during agent-magnet init.

License

MIT