Troubleshooting overview

Start by identifying which layer failed. Account access, cloud provisioning, VM readiness, workspace connectivity, model providers, voice, integrations, and billing have different recovery paths.

Collect evidence first

Record:

• the exact visible error, without secrets;
• agent name and immutable ID;
• provider, region, zone, runtime profile, and machine type;
• current and previous status;
• the action and approximate time;
• whether the issue affects one browser, one agent, or the whole account;
• relevant connector or billing period.

Do not include passwords, API keys, OAuth secrets, full tokens, or private conversation content in a support message.

Choose the failing layer

Symptom	Guide
Launch stays provisioning or reports an error	Launch and status
UI URL cannot be opened or voice fails	Connection and voice
Connector authentication, allowlist, or invoice looks wrong	Integrations and billing
Unsure what a badge means	Agent statuses

Safe first actions

Refresh the dashboard once, wait for any active transition, and read the complete error. For a workspace problem, confirm the Cloud card says ready before restarting services by stopping the entire VM.

Avoid repeated launch, start, stop, checkout, or OAuth requests. They can create overlapping operations and hide the original evidence.

When to contact support

Contact support when a transition exceeds its documented expectation and does not produce an error, provider cleanup appears incomplete after termination, reconciled charges cannot be explained, or a security boundary behaves differently from its configured policy.

Include the collected nonsecret evidence and what you already tried.