recommended
▱
Local (Docker)
Run everything on your machine.
- ✓ Full control, no external deps
- ✓ Data stays local
- ✓ Works offline
:: SEQUENCE_INIT
:: GET_STARTED
$ automem install
curl -fsSL https://automem.ai/install.sh | sh
npx @verygoodplugins/mcp-automem setup
✓ detect environment ok
✓ choose local or cloud ok
✓ provision AutoMem ok
✓ wire MCP clients ok
macOS, Linux, and Windows (WSL2)
How it fits
Drops into your stack.
Your agents connect through MCP. AutoMem stores every memory in a graph for relationships and a vector index for meaning — so recall is connected and semantic at once.
More than vector search
Vector-only memory matches text that sounds similar. AutoMem also maps how memories connect, so recall returns the answer and the thread it belongs to — not just the closest paragraph.
Speaks remote MCP
AutoMem runs behind an HTTPS endpoint and speaks remote MCP — so any client that supports it can connect, including the ChatGPT and Claude mobile apps, not just desktop CLIs.
AGENT CLIENTS
⌬
Codex
AI
Claude Code
◈
Cursor
◌
ChatGPT
MCP
Model Context Protocol
AutoMem
Persistent Memory Layer
⌘
FalkorDB
Graph Storage
◈
Qdrant
Vector Store
Graph for relationships. Vector for meaning.
Why agent memory
From forgetful to brilliant.
Out of the box, every agent starts each session from zero — re-asking your stack, your conventions, the decision you settled yesterday. AutoMem is the memory layer that fixes that. It captures what matters as you work and recalls it the moment it's relevant, so your tools stop forgetting and start compounding.
ChatABC Pro
Standard
Built on research. Designed for recall.
AutoMem combines a knowledge graph and vector search so agents can recall both facts and their context.
⌘
Graph (FalkorDB) captures entities, relationships, and time.
✣
Vector (Qdrant) captures meaning, similarity, and intent.
# Example: hybrid recall
MATCH (p:Project)-[:USES]->(t:Tool)
WHERE p.name = 'AutoMem'
RETURN p, t
ORDER BY ts DESC
LIMIT 20;
# + vector search for semantic matches
SEARCH "install script errors WSL" LIMIT 20;The result: fewer repeats, more relevant answers, and memory that compounds over time.
Read the research overview →How it works
Memories link themselves.
AutoMem runs quietly in the background, consolidating new memories into a retrieval graph. It clusters related ideas, strengthens the connections you return to, lets noise decay, and enriches recall — so the memory you get back tomorrow is sharper than the one you stored today.
0 memories 0 relations ┄ 0 auto-linked
click a node to trace its relationships · dashed gold links are added during consolidation
Works where agents already run
| Client / Platform | AutoMem MCP | macOS | Linux | Windows (WSL2) | Docker | Kubernetes |
|---|---|---|---|---|---|---|
| Codex (CLI) | MCP | ✓ | ✓ | ✓ | ✓ | ✓ |
| Claude Code | MCP | ✓ | ✓ | ✓ | ✓ | ✓ |
| Cursor | MCP | ✓ | ✓ | ✓ | ✓ | ✓ |
| ChatGPT (Code Interpreter) | MCP | ✓ | ✓ | ✓ | ✓ | ✓ |
| Custom MCP Clients | MCP | ✓ | ✓ | ✓ | ✓ | ✓ |
If your client supports MCP, AutoMem works. See full compatibility guide →
Extended compatibility: Claude Desktop, Claude Code, Cursor, GitHub Copilot, Codex, Windsurf, ChatGPT, Any MCP-compatible client.
Choose your setup
Deploy Once. Remember Everywhere.
▤
Railway
Managed cloud in minutes.
- ✓ One-command deploy
- ✓ Auto updates & backups
- ✓ Global availability
⬡
InstaPods
Self-hosted cloud in your infra.
- ✓ Your VPC, your data
- ✓ K8s native
- ✓ Enterprise ready
All options expose the same MCP endpoint, including the Existing Endpoint path. Self-hosted / Docker remains available for teams that want full control. Learn more in the install guide →
Ready to give your agents memory?
Deploy once. Remember everywhere.