Troubleshooting

Most problems fall into a handful of shapes. Find the symptom, follow the checks in order — they're sequenced from most to least likely.

The device shows Offline

"Online" means the device has reported within the last 5 minutes, so Offline is about power or network:

1. Power — is the unit on? Check the socket isn't switched or on a timer.
2. Network — is the Ethernet cable seated at both ends? Link lights? If the device is on Wi‑Fi, has the network name or password changed?
3. Give it two minutes after any power or network change — the device reconnects on its own; no one needs to log into it.
4. The HDMI status screen (plug in any monitor) shows the device's own view of its network state — the fastest way to see whether it has an address and connectivity.
5. Your internet — if the whole facility is offline, the device shows offline too. It recovers automatically when connectivity returns; backups and notifications queue and resume.

A device that stays offline with confirmed power and network is a support case — have the serial number ready (drawer Overview → Identity or the device label).

The agent isn't responding in chat

The device is online but the conversation is stalled:

1. Reload the module page — your session token may simply have expired after a long-lived tab; reloading mints a fresh one.
2. Check for a spend wall. If the agent has hit a spend cap or your facility's AI credit balance is empty, model calls decline — the agent can't think until spend is available. The Budgets & limits tab shows usage against caps; AI Models & Access shows the facility balance.
3. Check the header connection state. If it's cycling paths (Cloud relay → via Performance Hub), the tunnel may be re-establishing — give it a moment.
4. Restart the agent via drawer Configuration → Actions → Reboot device (confirm the prompt). The device restarts and is briefly unavailable; conversations, memory, and files are unaffected.

"Insufficient credits" or requests declined mid-task

The financial wall, specifically:

• Facility balance empty → top up in AI Models & Access. The agent resumes immediately.
• Daily/monthly cap reached → wait for the window to reset (facility local midnight / month start), or raise the cap in Budgets & limits.
• Lifetime cap reached → raise it; there's no automatic reset.
• Scheduled automations that were paused by a cap resume by themselves once spend is available.

"You don't have access to this agent"

You can see the device but get a locked panel instead of chat:

• The agent is in restricted mode and you're not on its allowed list. Click Request access — the owner is notified and can approve you in one click. See Access control.
• If the request seems to vanish: it's already pending — asking again doesn't re-notify. Chase the owner directly.
• If the owner has left the business, the account owner or an administrator can reassign ownership (Access Control → Change next to the owner).
• If you can't see the device at all but colleagues can, it's restricted with admins-only visibility — same fix: ask the owner or an admin.

Pairing fails

In the Add AI Agent Processor dialog:

A backup failed, or backups have stopped

• A single failed restore point shows as unhealthy in the Backups tab list — the next hourly run usually succeeds on its own; treat one-offs as noise.
• Backups stopped entirely: check the device is online, and check backups weren't disabled under Configuration → Cloud backup. Settings changes (including re-enabling) are picked up within about an hour.
• "Back up now" refuses: another backup or restore is already running — one at a time per device. Wait for it to finish.
• Repeated failures with a healthy, online device are a support case.

A restore didn't do what I expected

• Wrong point restored / results worse: use Undo restore on the completed restore — it returns the agent to the automatic pre-restore safety backup. Nothing is lost by a full restore; see Restore & recovery.
• Restore seems stuck: full restores stop the agent while data is replaced — allow for the size of the restore point. The Backups tab shows the request's state; a request that never gets picked up points at the device being offline.
• Can't compare two restore points: they're stored in different regions — comparisons work within a region.

A notice about local network access

When you're at the facility, the module tries the fastest path — Direct (local network) — which can prompt your browser for permission to reach devices on your local network. Allowing it gets you the quickest connection; declining is harmless — the module falls back to the cloud relay automatically and everything still works, just via the internet.

An update seems stuck

• The banner shows phases (Downloading… Installing… Restarting…); on a slow connection the download phase dominates — give it time.
• A scheduled update may start up to 5 minutes after its set time.
• If Lock updates was engaged (Configuration → Actions), the device holds its version by design — unlock to proceed.
• Still stuck after generous patience: Reboot device, which lets the update converge on startup.

When to contact support

Persistent offline with confirmed power/network, repeated backup failures, pairing errors that survive the checks above, or anything involving suspected hardware failure. Have ready: the serial number, the device name, your facility, and roughly when the problem started. If PH support access is enabled, support can inspect the agent directly, which shortens most investigations.

Related pages

• Reference — the exact numbers behind these behaviours
• Device management · Backups · Restore & recovery