Runs
Create, inspect, steer, pause, resume, and recover durable Agent executions.
A Run is one durable execution accepted for an Agent and session. Acceptance pins the Agent deployment, Resource bindings, Credential resolutions, aggregate Policy, pricing, limits, principal authority, and customer context. Controls change future execution at explicit safe boundaries; they do not rewrite committed history.
Choose a path
| Goal | Guide |
|---|---|
| Invoke an Agent from the Console | Start a Run |
| Inspect state, facts, and journal pages | Operate Runs |
| Resolve waits, steer, pause, resume, rebind, or rewind | Resolve waits and control Runs |
| Consume reconnectable output | Stream a Run |
| Understand the deployed behavior | Agents |
| Route external input | Channels |
Choose an interface
| Interface | Use it for |
|---|---|
| Console | Interactive inspection, debugging, waits, and controls |
| SDK | Defining durable work inside an accepted Run |
| CLI | Starting, streaming, inspecting, and controlling Runs from a terminal or pipeline |
| Platform API | Application-owned invocation and workflow integration |
Durable identity
The short Run id is not the full address. A stable API address includes namespace, Agent, session, and Run id. Copying the address from the Console preserves enough context to retrieve or control the same execution without guessing its owner.
Next steps
Agent code creates durable operations through Ctx; callers and operators create and control Runs through the Console or Platform API. Start with a stable session and caller-stable event id, inspect the accepted snapshot and journal, and use controls only after reviewing their recovery consequences.