One brain for many agents,
with a door.
Kemory is multi-tenant, MCP-native memory for the AIs and agents you build. Skip the memory layer — we built it once.
- MCP-native
- 17 memory tools
- Self-host or hosted
Shared is the meeting room.
Gatekeeper is the door.
Every agent gets its own identity. They share one namespace but access is governed — deny by default, JIT consent, a rule engine that decides who reads what before anything runs.
17 memory tools,
annotated so your client knows.
Each tool is annotated — reading from writing from deleting — so your MCP client knows what's safe to run before it runs anything. No silent side effects.
A hash-chained audit trail
you can prove.
Every memory event lands in an append-only, hash-chained log. Tamper with an entry and the chain breaks from that point forward. Audit is architecture, not an afterthought.
Connect once
from any MCP client.
Claude Code, Cursor, Claude Desktop, Cline, Windsurf, Warp, Codex, Gemini CLI — one config entry each. Web AIs connect over OAuth with nothing to install.
Connect once
from any MCP client.
Claude Code, Cursor, Claude Desktop, Cline, Windsurf, Warp, Codex, Gemini CLI — one config entry each. Web AIs connect over OAuth with nothing to install.
Self-host with Docker
or use hosted.
Run Kemory on your own infrastructure with Docker, or use the hosted version. Compaction runs on open models — you choose which, and you can swap them without migrating data.
Skip the memory layer.
We built it once.
Multi-tenant, MCP-native, governed by default. Connect once and 17 memory tools go live across every agent you build — with an audit trail and a self-host path when you need it.
