lumberroom: one memory for every AI tool

Field notes

lumberroom separates three kinds of stored context: fuzzy-searchable memories, canonical registry facts, and policy grants. Its example policy gives user:me private access, keeps global facts at open, and marks credentials as sealed; a second credential receives fewer rows and cannot read history or write registry data. The page also describes embedding on write and a seven-surface MCP arrangement, though those capabilities remain stated by the project rather than independently verified here.