Agents

An agent is one LLM worker: a single set of instructions, tools and a model. A peer agent is an external agent on another platform that a process flow can hand a step to. A model is the LLM an agent runs on. An execution is one run, with its inputs, outputs and trace. You create an agent, run it with a goal, and poll the resulting execution for the outcome. Creating or changing an agent needs fabric:orchestrate.write, reading one needs fabric:orchestrate.read, and running one needs fabric:agent.run.

Start with create-agent, which returns the agent_id the others take.

• create-agent: Create an agent.
• get-agent: Read one agent.
• list-agents: List the project's agents.
• update-agent: Change an agent's fields.
• delete-agent: Delete an agent.
• run-agent: Run an agent with a goal.
• agent-versions: Read promoted version history.
• rollback-agent-version: Make an earlier version live.
• register-peer-agent: Register an external agent.
• handshake-peer-agent: Test an external agent endpoint.
• list-peer-agents: List peer agent connections.
• delete-peer-agent: Delete a connection.
• list-models: List models and the default.
• run-model: Call a model directly.
• set-model: Set the organization default.
• get-execution: Read one run.
• list-executions: List runs.
• pause-execution: Pause a run.
• resume-execution: Resume a paused run.
• restart-execution: Re-run an execution.