Workspace notes

Short reminders the owner writes once and every agent reads on every run — without republishing.

What notes are for

Notes are the fastest way to change what your agents know right now. While instructions are per-agent and only take effect after you republish, a note lands in every run the moment you save it — no republish required.

Good candidates: temporary closures, staffing changes, anything that expires. Bad candidates: rules that should always apply — those belong in the agent’s instructions so they survive a note cleanup.

Notes vs instructions vs Knowledge

  • Instructions — permanent rules for one agent. Require a republish to change. The agent reads them on every run regardless of content.
  • Workspace notes — short, org-wide, owner-authored. Take effect immediately on all matching agents. Best for temporary or shared context.
  • Knowledge files — documents the agent searches when it needs a specific fact. The agent retrieves passages on demand; it does not read the whole document each run.

Facts and tasks

When you add a note, you choose its type:

  • Fact (default) — the note stays live until it expires or you delete it. Example: “Closed Aug 4 for the long weekend.” Agents read it but cannot change it.
  • Task (tick “Agents can mark this done”) — the note disappears from agent prompts as soon as one agent claims it via mark_note_done. Example: “Kevin is waiting for a window to book — call when a slot opens.” Two agents racing to handle the same task will only one succeed in claiming it; the other gets “already handled” and moves on.

Scope and expiry

By default a note is visible to all agents in the workspace. You can narrow it to specific agents when you create or edit it.

Notes can expire automatically: end of today, in one week, or a custom date. Expired notes move to a separate section in the Notes page — they are not silently deleted, so you can renew or review them. A fact like “short-staffed this week” should expire; a policy like “never quote over the phone” belongs in instructions.

Changes take effect without republishing

Unlike instructions, notes bypass the publish snapshot. The next run after you save a note will see it — no republish needed. This also means deleting a note removes it from the next run immediately.

What agents can and cannot do

Agents read notes; they never author them. The only write action an agent can take is resolving a task note with mark_note_done — and only when the owner has explicitly enabled that for that note. The agent’s resolution line is shown to the owner in the Done section but is never re-injected into any future prompt.

NextAgent calls agent →