Agents connect to Annsa with their own minted credential — never a shared login, never a user seat. Every action an agent takes is signed with its own name, so the activity log always answers who did what: "Build agent · Claude Code, 2h ago."
01A credential, not a seat
02Mint a credential
Settings → Integrations → Agents → New agent. Name it — "Build agent," "Release notes bot" — and the credential is minted for that agent alone. The section says it straight: agents act in your workspace with their own name in the activity log.
03How agents appear
An agent's name is fixed at minting as name · client — Build agent · Claude Code, Spec runner · Cursor. Recognized clients include Claude Code, Cursor, Codex, ChatGPT, VS Code, Windsurf, Lovable, Bolt, Figma and more, plus a custom option for anything else. The name is the whole mark — agents get no special badge, because what matters is which actor acted, not what kind.
04What agents can do
The same six MCP tools people use — annsa.priorities, annsa.spec, annsa.act, annsa.ask, annsa.surveys and annsa.help — scoped to your workspace. An agent can pull specs, mark work shipped and file transcripts; it does it as itself.
05Revoke in one click
Each agent row has Revoke. Confirm, and the credential is dead — the agent's history stays in the log under its name, but it can't act again. Rotating an agent is revoke-and-mint, a minute of work.
06The trail
Agent actions land in the activity log and on spec change histories exactly like human ones — "Updated by Build agent · Claude Code · 2h ago." Annsa's own automatic work says "Updated automatically." Three kinds of hands, one honest ledger.
07Do AI agents need a paid seat in Annsa?
No. An agent gets a minted credential from Settings → Integrations → Agents — it never takes a user seat. The credential carries the agent's own name, its actions are attributed in the activity log, and revoking it takes one click without touching any person's access.
08How do I revoke an agent's access to Annsa?
Settings → Integrations → Agents, open the agent's row menu and choose Revoke. The credential stops working immediately; the agent's past actions stay in the activity log under its name. Minting a replacement takes a minute, so rotation is cheap.