

What's Planned
Every AI assistant now keeps its own memory of you, locked inside one vendor's cloud. Switch tools and you start from zero; ask what it knows and you get a partial answer. Engram flips that around: a personal memory server that runs locally, stores what you choose to remember on your own machine, and exposes it to any assistant over the Model Context Protocol (MCP). One memory, owned by the user, portable across clients.
Planned Stack
- - Model Context Protocol (MCP)
- - Local-first storage
- - Web UI for reviewing and editing

Status
v1 design locked
Target
2026 – 2027
Role
Solo Developer
Why I'm Building It
Ownership: memory about a person should be data that person holds, can read in full, and can delete, not a feature of someone else's product.
Portability: MCP gives a single open interface, so the same memory works with whichever assistant or editor is in use that day.
Privacy by default: local-first means nothing leaves the machine unless the user decides it should, the same principle behind Talon.
