supermemory

Freemium

supermemory is an AI Agents tool that gives your AI a memory. It remembers what you said last week. It pulls facts from your files and feeds them back when you need them.

You plug it into Claude Code, Cursor, or your own app. It stores what it learns in a graph. It forgets stale facts. It keeps what matters.

You can run it in the cloud or on your own machine. The local version keeps everything offline.

Open it if you build AI agents that need to remember users across sessions. Skip it if you want a finished product with no setup.

It fits when memory is the missing piece.

• Developers building AI agents who need memory infrastructure: You get a system that learns, merges, and forgets. You stop hacking together your own vector pipeline.
• Users wanting persistent memory across Claude Code, Cursor, Codex conversations: Your coding assistant remembers your project context. You stop re-explaining your stack.
• Teams requiring self-hosted memory for compliance or privacy: Data stays on your machine. You avoid sending memory to another company's servers.
• AI product builders who want to skip vector DB configuration: You get ingestion, embedding, and retrieval in one API. You ship faster.
• Teams building support copilots or onboarding assistants with returning users: The bot recalls past tickets and preferences. Users feel recognized.
• Products where user preferences and projects evolve across sessions: Memory updates as habits change. Context stays current.
• Developers wanting to replace self-managed vector DB + embedding pipeline: You rip out your DIY setup. You get one maintained tool.
• Teams with well-curated, approved business material willing to maintain inspection loops: You feed it clean docs. You still verify outputs before they reach customers.
• Organizations not handling sensitive customer data without deeper privacy review: You test with internal content first. You add safeguards before expanding.
• Cursor users wanting a /command-accessible memory skill for Composer: You install with one CLI command. Memory shows up where you already work.

These are what you notice after wiring it up.

• Zero-boilerplate SDK setup: You npm install, add your key, and start indexing. No weeks lost on vector DB tuning.
• Local self-host with one binary: You download, run, and point your agent at localhost. Your data never leaves your machine.
• Hybrid search across documents and extracted memories: You find content whether it sits in a PDF or lives as a learned fact. Results blend both.
• Knowledge graph that updates and relates: Memories connect automatically. Old facts fade. New ones strengthen.
• Connectors for Google Drive, Notion, OneDrive, Gmail: You sync sources you already use. Context lives in one place.
• Drop-in framework wrappers: You add memory to Vercel AI SDK, LangChain, or LangGraph without rewriting your stack.

The setup is minimal for what it does.

1. Export your API key from the dashboard.
2. Install the SDK with npm or pip and add your key to your env file.
3. Wrap your agent calls with the memory layer.
4. Query with natural language and watch context inject into your prompts.

It is not the right pick for every case.

• Self-reported benchmarks: The speed and accuracy claims come from the makers. You have no third-party verification to trust.
• Local version drops features: Hosted connectors and MCP do not work offline. You trade convenience for privacy.
• Inspection still required: Wrong documents rank high. Old facts persist. You cannot ship outputs without human review.
• Embedding changes mean re-ingestion: Switch models and you rebuild your index from scratch. It is easy to misconfigure.
• Soft delete is not verified erasure: If you need proof of deletion for compliance, you have extra work to do.
• No governance for source dates or access control: You cannot easily see which version of a doc was retrieved or who can see what.