Run observability¶

RunAgents treats the run as the operational unit for debugging and governance.

Instead of piecing events together from multiple systems, operators can follow a single run record from request to outcome.

What operators can inspect¶

A governed run can include:

• execution status and timestamps
• LLM and tool-call events
• approval-required pauses
• approval grants or rejections
• blocked actions and payload hashes
• model budget failures and other runtime errors

Where to inspect it¶

• Dashboard for active runs, pending approvals, and recent operator work
• Run lifecycle for the state machine and event model
• Runs API for programmatic inspection and export

Why this matters¶

Observability is more useful when it stays attached to the action path.

That means the same operational surface can answer:

• who triggered the run
• which agent and tool were involved
• whether policy allowed or paused the action
• who approved it, if approval was required
• what outcome completed or failed the run