Concepts
The memory model and the design decisions behind zero-memory.
This section explains how zero-memory represents knowledge and why it is built the way it is.
Start with the memory model: memories, kinds, entities, links, and the MCP tools that operate on them. Stored content is always canonical English. Canonical language covers why, and how other languages are handled on write, read, and in the dashboard.
Access is the next layer. Scopes and isolation describes personal, project, and team scopes, explicit membership, project-isolated reads, how a session attaches to its project (and why an unattached write is refused), and the verified portable layer. Writers have identities of their own. Agent principals explains how non-human writers are identified and what that identity enables across memories, activity, and metrics, while Identity covers prefixed domain identifiers, branded id contracts, the system-of-record rule, and how a project is identified.
The rest is lifecycle. Provenance records who wrote a memory, how that ranks it, and which client it came from. Supersede and versioning shows how facts are updated without creating duplicates, and how version history forms behind a current fact. Open loops are tasks and open questions that persist in briefings until closed, and the project board holds work that runs longer than a loop: cards whose every move carries its reason.