Give every AI a
lasting memory
Synapse is a local-first memory layer exposed over MCP. Any AI client gains persistent, semantic, cross-tool recall, so it never forgets what matters to you.
The problem
Today's AI is brilliant but amnesiac
Every chat starts from zero. Context is trapped inside one tool. You repeat your preferences, your stack, your decisions, over and over. The intelligence is there; the continuity is not.
How it works
Three moves to total recall
Connect any MCP client
Point Cursor, Claude Desktop, or your own agent at Synapse over stdio or HTTP. No SDK lock-in.
It remembers as you work
Synapse stores facts, preferences, and decisions as on-device vector embeddings, kept private.
It recalls across sessions
Every future conversation starts with relevant memory recalled semantically, in any tool.
The MCP surface
Six tools your AI can call
remember()
Durably store any fact, preference, or decision.
recall()
Semantic search across everything the AI has learned.
build_context()
Synthesize a ready-to-inject context block.
get_related()
Traverse the semantic graph around a memory.
list_memories()
Browse the most recent memories, newest first.
forget()
Permanently delete a memory by id.
Why Synapse
Memory that belongs to you
Open protocol
Built on MCP, so memory is portable across every compatible client, not locked to one vendor.
Local-first & private
Runs on your machine with an embedded vector store. Your memories never have to leave your device.
Zero-key by default
On-device embeddings mean semantic recall works with no API key, no signup, no cloud cost.
See it remember, live
Teach the playground something about yourself, refresh, and watch it recall, with no setup, no key, all in your browser.