Team memory, served over MCP
A durable, shared place for the decisions, conventions, and constraints an assistant should carry across conversations — read and written through one MCP endpoint by any MCP-capable client.
kumbuka
Shared, persistent memory for AI assistants working with a team — served over MCP, curated through an admin console, with a private space that stays private.
kumbuka is Swahili — the imperative “remember!”.
The context tax
Section titled “The context tax”AI assistants are stateless between sessions. A team using them pays the same toll over and over — re-explaining the things that should simply be known: which database is the system of record, that money is integer minor units, that service names are kebab-case. kumbuka turns that steering knowledge into a first-class, team-owned asset, so it is applied without being re-told.
Private stays private
Each member has a private space reachable only by them, only through their own authenticated session. No admin, console, or team API can read it — enforced structurally, not by a setting.
Curated, not accumulated
A web admin console lets the team see and edit what the assistant relies on, instead of each person growing an opaque, divergent context.
Open-core, self-hosted
The Community Edition is the free, AGPL-licensed, single-tenant memory core. Run it as one Docker Compose stack on your own infrastructure.
Where to next
Section titled “Where to next”- Overview — what kumbuka is and the personal / shared boundary.
- Quickstart — self-host the Community Edition.
- Concepts — the memory entry, scopes, and authorship.
- MCP tools — the five tools an assistant calls.