Documentation

Team Memories

Workspace memories are short, team-authored facts that load into every member's Claude Code session automatically. Think of them as your team's shared brain — the conventions and rules everyone's sessions should always know, managed centrally in the SuperClawd app.

What a memory is

A memory is a short, durable team fact — a convention or rule worth repeating to every session. For example:

  • "We use pnpm + turbo. Never npm."
  • "Deploy Mon–Thu only, never Fridays."
  • "The auth service is owned by the platform team — ask before changing it."
  • "Our API errors always use the AppError class."

Memories are pushed one-way into sessions. They're team context, not instructions you have to invoke — but an explicit request from you always takes precedence over a memory.

One-way and privacy-clean

Memories only flow down into sessions. SuperClawd never reads your machine, your native Claude Code memory, or your session content. Every memory is text a teammate typed into the app.

How they reach your session

When you launch superclawd, the CLI fetches your workspace's memories and injects them as a small always-on block at the start of the session — right alongside your foundational rules. The block is re-applied as the session grows so it stays relevant during long sessions. There's nothing to invoke: the facts are simply present.

Always in full, never retired

Team (human-authored) memories are always served in full and never decay — unlike AI-learned memories, which load on demand and retire when they go unused.

Authoring memories

Memories are managed entirely in the SuperClawd app — it's the single source of truth. These are team memories (shown as Team Memories in the app): members receive them read-only, and maintainers, admins, and owners can create, edit, and delete them. Each memory has a title (max 60 chars) and a short body (max 256 chars), plus an enable/disable toggle, an experimental flag, and a favorite star.

Keep memories short

The always-on block is re-injected for salience, so every byte is paid for repeatedly. Aim for one short fact per memory — bodies are capped at 256 characters, and the combined enabled set has a 4,000-character budget. The app shows a budget meter and warns as it fills. You can keep up to 1 memory on the free tier and 100 on paid.

Billing

Team memories are free. They're injected into every session at startup rather than fetched on demand, so they're never metered — and keeping memories in the app is never charged. This differs from learned memory: on paid accounts, a get_memories call that returns one or more repo or agent memories is metered at 0.10 credits per call (free on the free tier), while private memory reads are always free. See Billing & Credits for details.

Experimental memories

A memory can be flagged experimental — it's then only injected into sessions running in experimental mode, and withheld from everyone else. Use it to trial a new team fact with a power user before rolling it out to the whole team. See Experimental Mode.

Managing the list

Workspace memories live in their own paginated, searchable list with a status filter (all / enabled / disabled). From there you can search by title or body, toggle individual memories on or off, mark them experimental, star favorites, and edit or delete them. Lightweight audit fields (who created and last updated each memory, with timestamps) make "who changed our team rules" easy to answer.