Agent run monitoring

On each Relevance AI run, the agent reads the execution record, its inputs, status, and any errors returned. It checks the result against your quality rules, updates the task record, and notifies the owner when a run finishes or fails. Runs that complete inside expected parameters are logged and closed. A run that errors repeatedly, returns output far outside the usual shape, or exceeds a cost or time limit you set is held and escalated to the team that owns the workflow, with the run data attached so a person can decide whether to retry or adjust it.

Task queue handling

As work items enter a Relevance AI queue, the agent reads each task, its payload, and its priority. It applies your routing rules, assigns the task to the correct agent or workflow, and updates the record so progress stays visible. Standard items move through on their own and are noted. Tasks with missing inputs, an unknown type, or a priority that conflicts with your rules are set aside and routed to a human operator, who supplies what is needed or redirects the item before it runs, keeping malformed work out of automated pipelines and protecting downstream results.

Output review routing

When a Relevance AI workflow produces a result meant for a customer or a record change, the agent reads the output and grades it against your acceptance rules. Outputs that pass are written back or released and logged. It updates the relevant record and notifies the owner as each item clears. Low confidence results, answers touching sensitive topics your rules protect, or outputs that would alter high value data are held for a person to approve, so no unchecked generation reaches a customer or overwrites a system of record without human sign off.