Esfis is in private beta. Read the post
One memory,
every assistant.
Esfis is a small engine that runs on your machine. It learns from what you tell Claude, ChatGPT and Cursor, and gives all of them the same memory of you.
One store, linked by people, places and time.12 memories
What it does
Every assistant keeps a private notebook about you. None of them talk to each other, none of them are yours, and all of them forget. Esfis is the layer underneath: one store you own, that every tool reads.
Learns without being asked
Facts are pulled from ordinary conversation and linked to the people, projects and dates they belong to. Nothing to tag, no “remember this”.
Follows you between tools
The store sits outside the model providers, so it carries across Claude, ChatGPT, Cursor, Perplexity and your own agents. Explain yourself once.
Stays on your machine
A single file on your disk, embeddings computed on-device, and a full history of every change. Privacy as architecture, not as a policy page.
ClaudeChatGPTCursorPerplexityRaycastAny MCP client
One message in, four memories out
This is the whole trick. Esfis reads what you already said, keeps the parts worth keeping, and dates each one.
Had a great call with Alex. He’s enjoying the new PM role at Stripe, though the payments work is intense. He moved to Seattle for the job, Capitol Hill. Wants dinner next time I’m in town.
- Alex works at Stripe as a PMsince August 2026
- Alex finds the payments work intenseas of 4 September
- Alex lives in Capitol Hill, Seattlesupersedes: Alex lives in Austin
- Alex wants dinner when I’m next in Seattleopen intent
Same question, different assistant
Nothing about the model changes. The only difference is whether it can read what you already told it.
How it works
You just talk
Meetings, deadlines, preferences and the people involved are saved on their own while you work.
Everything connects
Each fact is atomic, carries enough context to stand alone, and links to what Esfis already knew about that person or project.
Recall is instant
Four deterministic stages in under 50 ms: keyword, meaning, graph, rerank. Fast enough to sit in every request.
It keeps up with your life
When something changes, the old fact is superseded, not deleted. You can always ask what you knew last month.
Numbers
Recall, p95, on a laptop
LongMemEval, against 71% for the raw context window
Memories kept across all installs
“Tried structured memory files, custom prompts, everything. Esfis is the first thing that just works, and it doesn’t phone home.”
Founder, Chatarmin
Questions
What is Esfis, exactly?
A memory engine that runs on your machine. It learns from your conversations and gives every assistant you use the same persistent memory of you.
Can I use it today?
Not yet. Esfis is in private beta while the retrieval pipeline settles. Ask for early access and we will write to you when your turn comes up.
How will it reach my assistants?
Through MCP for Claude, Cursor and any MCP client, through browser plugins for ChatGPT and Perplexity, and through a small HTTP API for your own agents. One store, shared by every tool.
Where is my data stored?
In a SQLite file on your disk. Embeddings are computed locally. Nothing leaves your device unless you turn on sync, and enterprise deployments can run fully air-gapped.
How is this different from built-in memory?
Built-in memory is per-vendor, opaque and shallow. Esfis is one shared store you can open, diff and edit, with a full history and a retrieval pipeline benchmarked in public.
What will it cost?
Free for personal use, with no limit on memories. Paid plans add sync between devices and shared spaces for teams. See pricing.