Capture
Conversations, files, and external sources flow in via MCP, REST API, or auto-sync.
MCP-native LLM Knowledge Base Hebbian
Shared memory &
knowledge base for AI.
Claude, Cursor, Gemini CLI — every MCP-capable client shares the same memory. Kagura Memory Cloud turns conversations into lasting knowledge.
One line to connect (Claude Code)
$ claude mcp add --transport http kagura-memory https://memory.kagura-ai.com/mcp Cursor and Gemini CLI use the same MCP config. Self-hosted? Just swap the URL.
Self-hosted
Open-source
Secure & private
Sound familiar?
Explaining the same context to Claude — again.
Searching chat history for an answer you know you got — somewhere.
Knowledge trapped in one person's head, invisible to the team.
You don't want your AI conversations to go to waste.
Memory Cloud's answer
Turn AI conversations into memory. What you've said once, every AI remembers.
Auto-memory
Conversations automatically become searchable knowledge. No manual saving.
Beyond RAG
Inspired by Karpathy's LLM Wiki — your AI compiles, indexes, and refines knowledge as you use it. Not just retrieval — a living knowledge base.
Cross-platform
Memories shared across Claude, Cursor, Gemini CLI, and any MCP-capable client, plus REST API.
Team sharing
Personal insights become team assets. Shared contexts, role-based access.
Memory turns AI into an agent.
And memory that compounds becomes a knowledge base.
An AI without memory is a disposable tool.
An AI with memory is a growing partner.
Living Knowledge Base
How knowledge grows
Four steps from raw input to a compounding knowledge base. Inspired by Karpathy's LLM Wiki pattern.
-
-
Compile
Your AI structures each fact into a 3-layer schema — summary, context, content. Sleep Maintenance consolidates nightly.
-
Index & Query
Triple index — keyword (BM25), semantic (vector), and relational (graph). Hybrid Search with AI Reranker.
-
Compound
Hebbian learning strengthens connections every time you recall — your knowledge base gets smarter without LLM cost.
…and every recall compounds — connections strengthen with use, the cycle continues.
Built for how you work
Developers
Before Re-explaining your codebase to Claude every session
After Claude remembers your architecture, patterns, and past debugging sessions
Researchers
Before Paper summaries and notes scattered across tools
After Every insight builds on the last — connections emerge automatically
Teams
Before Knowledge trapped in individual chat histories
After Shared context means every team member's AI is up to speed
Up and running in minutes
Cloud
- 1 Request beta access (email or sponsor)
- 2 Once approved, sign in at memory.kagura-ai.com
- 3 Connect your AI client via MCP
Self-hosted (Docker)
- 1
git clone kagura-ai/memory-cloud - 2
docker compose up -d - 3
Connect and start