Agents

An Agent performs a Line stage. A prompt Agent runs a selected harness, model, and prompt. A script Agent runs commands. Both types belong to a factory and require a Machine.

In the dashboard

1. Open your factory and go to Agents (/factories/my-work/agents or factory → Agents when the default factory is selected).
2. Choose New agent and pick Prompt or Script.
3. Select a verified Machine, harness, model, and prompt (or script commands). The form loads available models from your tenant; you do not need to paste model IDs by hand unless you prefer the CLI.
4. Save the Agent. Names are unique within a factory and cannot change after creation.

Update an Agent from its detail page when you need a new prompt, model, or Machine assignment.

Build with a local agent

Install the Islo skill in your local coding agent and ask it to prepare an Agent for your Line. Give it the repository, the work the Agent should do, and any Machine or model requirements. The agent checks islo schema agent, chooses a compatible model from /inference/models when needed, creates the Agent configuration, and validates it before creation.

Ask the agent to show you the Machine assignment, harness, model, prompt or commands, and validation result. Review those choices before it creates or updates the Agent. The same skill handles updates; the update manifest omits name because the CLI identifies the Agent by its name.

Use on a Line

Ask the local agent to assign the Agent to a Line stage, or pick the Agent in the dashboard Line builder. A Machine can serve multiple Agents and Lines. A Line’s selected Manager reads the Agent’s current configuration on each Manager turn.

For direct Agent sessions, see the API reference. The CLI agent integration describes a separate direct sandbox command, islo use --agent.