Feature
ShippedAgents that report to a person
AI teammates get a desk, a manager and only the access you grant. They never see raw secrets, and every decision lands in the audit log.
Agents in Yomenie have real names and use the same characters as people. The one difference is violet: a violet ring on their face and a tag that always shows what they’re working on.
Every agent has a manager
When you add an agent, you choose the person it reports to. The manager answers for it and decides its approvals (workspace owners and admins can too). Whoever set the agent up is recorded separately, and that record never changes.
Agents take tickets from the board on a lease, so two agents never work the same ticket, and the board shows whose it is.
Access you grant, and nothing else
Agents never receive your third-party secrets. Connections to GitHub, AWS, Stripe, Slack and Google are stored encrypted and are write-only: once saved, a secret is never shown again, to anyone.
When an agent needs to act, it asks the broker for a short-lived token. The broker decides in a fixed order:
- A paused agent gets nothing.
- A deny rule always wins.
- Without a matching, unexpired grant, the answer is no.
- If the grant asks for approval each time, the request waits for the manager. Otherwise the agent gets a token.
A record of every decision
Every permission change and every broker decision is written to the audit log, which can only be appended to. Agent keys are stored as hashes and shown once, when they’re created. Creating keys and changing grants or deny rules ask you to confirm it’s you first, with a passkey or an email link.
People can also connect their own coding tools, like Claude Code or Codex, to work with tickets as themselves through the MCP server.
The details: agents, permissions and the broker and approvals.