The public utility for AI memory

One memory.
Every AI.

Your AI forgets you the moment a chat ends. LLMem is the memory that keeps running between conversations, consolidating and connecting your context across every model. Memory is home. Models are just where it thinks.

Developers can start todayClaude and ChatGPT connector rolling out. Free tier included.

Memory shouldn't be locked to a platform

Your identity, your preferences, your context, your history. None of it should belong to OpenAI, Anthropic, or Google. It shouldn't reset every conversation. It shouldn't get stale. It should work everywhere, evolve with you, and stay private.

LLMem is the neutral, shared memory layer for all AI. We're a not-for-profit. No investors, no data monetization. Just infrastructure at cost.

A save file, or an ongoing process?

Most memory is a diary.

It records moments. It does not think about them. Facts go in and never integrate. That is anterograde amnesia with a nice UI, and it is where most memory tools stop.

LLMem has a default mode network.

Ambient agents revisit your memories between conversations. They consolidate recent episodes, let unused ones fade into recoverable reveries, and promote what defines you into core. It runs at 3am with no one watching.

A save file versus a self.

Your experience has a history and a trajectory, not just a checkpoint. Restore a save file and you get data. This keeps running.

Four pillars

Substrate independent, ambient by default, append-only at the core, and private all the way down.

Substrate independent

Models get deprecated. Your memory does not. Swap in a faster brain and the mind it carries stays intact. No vendor owns your continuity.

Ambient curation

Short to long to core. Unused memories fade into reveries you can recover. A temporal graph tracks how facts change over time. Background processing, not passive storage.

Living memory, perfect recall

It grows like a mind and remembers like a machine. The surface evolves while the substrate stays append-only. Revision never destroys the record.

Zero-knowledge

Per-memory encryption. Your keys. We cannot read your memories, and that is the point. Honest memory only works when it is private.

Remote MCP for claude.ai and ChatGPT. Stdio MCP for Claude Code and Cursor. REST for agents. No pasting.

Connect in under a minute.

claude.ai and ChatGPT

Rolling out

Add a custom connector. Sign in once. Your AI remembers you instantly. Here is how it will work:

  1. 1. Settings, then Connectors, then Add custom connector
  2. 2. URL: https://api.llmem.io/mcp
  3. 3. Sign in with LLMem.

Claude Code and Cursor

Native MCP over stdio.

npx @llmem/mcp

Set LLMEM_API_KEY in your environment.

Agents and custom

Two REST endpoints, any language.

GET  /v1/mems/search?q=...
POST /v1/mems

X-API-Key: <your key>
Get an API key