OwnMem

OwnMem starts from a different question than most agent-memory systems: who owns project knowledge, who may change it, and how a bad memory entry gets stopped before it changes an agent's actions. Its memory lives as Markdown in a repository folder, travels through clone, review and rollback with the code, and is read the same way by every agent working in that repository.

Its default recall makes no model or network call, so the same query, configuration and snapshot always produce the same ranking, and delivery is graded: the memory quoted, up to three pointers, or an explicit abstention naming the gate that refused, rather than presenting unproven content as fact. A hard entry-count cap only ratchets down, so adding a new memory to a full corpus requires retiring another one in the same change. It does not replace instruction files such as CLAUDE.md or AGENTS.md, which are read in full every turn; OwnMem instead decides which of the project's hard-won lessons are worth surfacing for the current task, including deciding that none of them are.