UiPath Documentation
autopilot
latest
false
Autopilot user guide

Troubleshoot

Conversational diagnostics for UiPath platform failures, triaging RPA, Orchestrator, Maestro, Agents, and Integration Service issues.

Autopilot brings conversational diagnostics to the UiPath platform. Describe a failure in plain language and Autopilot triages the issue, ranks likely causes, tests each hypothesis against live platform data, and tells you exactly what went wrong and how to fix it.

How it works

A diagnostic session follows the same four-stage flow regardless of failure type:

  1. Triage — Classify the failure scope from your description, the job or process identifier, or the error message. If the input is ambiguous, Autopilot asks a clarifying question.
  2. Generate hypotheses — Produce a ranked list of possible causes, drawing on the relevant domain playbook and any evidence gathered during triage.
  3. Test hypotheses — Confirm or eliminate one hypothesis at a time. Autopilot fetches logs, traces, queue items, healing data, configuration, and audit events to gather evidence.
  4. Resolution — Present the root cause, the chain of events that produced the failure, and a concrete fix. Autopilot asks for explicit approval before changing anything.

You can be involved at four moments: clarifying the scope during triage, supplying extra context for hypothesis generation, providing source code or configuration files when needed, and approving recommended fixes.

What you can diagnose

Autopilot covers the following domains. You describe the problem; Autopilot fetches the evidence and returns a root cause.

DomainWhat you can diagnoseExample ask
RPAFailed jobs (SelectorNotFoundException, TimeoutException, and similar), selector mismatches, queue item failures, Healing Agent data analysis"Why did the InvoiceBot job fail this morning?"
OrchestratorStuck or faulted jobs, permission and role issues, deployment and publishing errors, queue configuration problems"Why is my robot not picking up items from the Finance queue?"
MaestroFaulted process instances, step-level failures, incidents, multi-domain failure chains that involve agents or RPA"Why did the approval process instance fault on step 3?"
AgentsLLM step errors, context grounding and resource errors, input schema validation failures, guardrail violations, and Integration Service tool-call failures"Why did my agent stop with a guardrail violation?"
Integration ServiceConnection authentication and token-refresh failures, connector activity errors, missing required inputs, and triggers that don't fire or fail to execute"Why is my Salesforce connection returning a 401?"

Autopilot operates with the same access you have. If you cannot read certain logs or folders as a signed-in user, Autopilot cannot read them either.

Prerequisites

  1. A logged-in session with access to Automation Cloud or Automation Suite.
  2. Access to the relevant folders and logs for the domain you want to diagnose.

Tips for effective sessions

Treat Autopilot like a colleague helping you debug — context and back-and-forth improve the outcome:

  • Be specific. Job IDs, process names, queue names, and exact error messages let Autopilot skip clarifying rounds.
  • Keep the conversation going. If the first recommendation does not match what you observe, say so. Phrases like "that's not it", "I already checked that", or "dig deeper" keep the investigation moving.
  • Redirect when needed. If Autopilot follows the wrong thread, correct it — for example, "the issue is not with permissions; the job ran fine yesterday".
  • Share files when asked. XAML, project.json, configuration files, and similar artifacts sharply improve diagnosis quality.
  • Match access to expectations. Autopilot inherits your permissions. If you cannot view a log, Autopilot cannot either.
  • How it works
  • What you can diagnose
  • Prerequisites
  • Tips for effective sessions

Was this page helpful?

Connect

Need help? Support

Want to learn? UiPath Academy

Have questions? UiPath Forum

Stay updated