Skip to main content

Tool governance

Tool-level governance settings control how a Tool runs as a whole — its resource limits, concurrency, throughput, priority, and what happens when a run fails. These settings apply to the entire Tool, complementing the per-step error handling and retry controls.

Execution limits​

These limits keep a single run bounded and predictable.

  • Max parallel jobs in a tool run — caps how many jobs (branches/steps) run at the same time within one run.
  • Max node executions in a run — caps the total number of node executions in a single run. This guards against infinite loops or unexpectedly large inputs triggering very long foreach loops.
  • Max timeout (tool-level) — sets an overall time budget for the entire tool run. (Timeouts can also be set per step; this setting bounds the whole tool.)

Concurrency and throughput​

These settings govern how the Tool behaves across many simultaneous or rapid runs.

  • Concurrent runs of this tool — caps how many runs of the Tool can be in flight at once.
  • Rate limit / throttling — limits the number of runs allowed per unit of time.

Run priority​

Set the scheduling priority for the Tool's runs:

  • High — prioritize this tool over others for latency-sensitive use cases.
  • Mid — the default priority.
  • Low — for work that can be processed asynchronously or in batches.

Fallback workflow on failure​

  • Run a workflow when a job fails — designate a fallback workflow that executes when a run fails. Use this to handle errors gracefully, notify a system, or run a recovery path instead of simply failing.

Tips for Effective Use​

  • Bound long loops: Set Max node executions when a tool uses foreach over large or unbounded lists.
  • Protect latency-sensitive tools: Mark interactive, user-facing tools as High priority and batch jobs as Low.
  • Throttle external dependencies: Use rate limiting and the concurrent-runs cap to stay within the limits of any external APIs your tool calls.
  • Plan for failure: Pair a fallback workflow with per-step error handling for layered resilience.

If you have any questions or need further assistance, please contact the helpdesk ([email protected]).