Adding an agent
Describe the agent you need, start from a ready-made agent, connect an agent running elsewhere, or make a Work exchange into an agent. Nothing is added before you confirm.
Four ways to start
- Describe it: the builder drafts the steps, the instructions and the gates; the draft shows alongside, live.
- Start from an example: a ready-made agent from the Library — translator, summariser, classifier… — is added as it is.
- Connect a remote agent: by the URL of its Agent Card, and only after approval (see Remote agents).
- From Work: « Make it an agent » takes an exchange's request and the collections it cited, and says what it does not carry over.
The preview before adding
The preview shows only what the draft really declares: its steps, its tools with their privilege level, the knowledge it reads, whom it may delegate to, its human gates. An agent never approves its own action.
You choose the team that owns it — or none, and it belongs to the whole organisation: its team's rules apply, its approvers decide its gates, its cost is attributed to it.
Studio screenshot · Add an agent — to come
Who does what
- An approver or an owner has a draft written.
- An owner adds it to the Mesh. The agent is saved as a version, and every run then pins its exact definition by its SHA-256 fingerprint.
Not yet
SoonA described agent is not bound to any of the organisation's tools yet — Slack, e-mail, search…: tools are bound afterwards in the editor.