Troubleshooting

Pick the problem you're seeing and follow the steps. Most issues clear up in under a minute.

Connection & messaging

Ghost isn't replying

You @mention Ghost or send a DM, but nothing comes back.

1Open My Agent and confirm the status reads "Running."
2Make sure it's a real mention — type @your-agent and pick it from the autocomplete, not plain text.
3Check the channel is connected: My Agent → Channels, the row should say Connected.
4Give it a moment — Ghost adds a 👀 reaction within ~30s when it picks up a task.
5Open the Activity tab on your agent and check for a recent error.
6Still silent? Open the agent's ⋯ menu → Restart to reboot the VM, then try again.

The bot's status dot is grey in Discord and it ignores every message.

1Open My Agent and check the agent card for a failed status or an error banner.
2Open the Activity tab and check for a recent error.
3If you supplied your own bot token, re-check it in Channels → Discord → Settings.
4Click the ⋯ menu → Restart to redeploy. The dot turns green once it's back online.
5Still offline a couple of minutes after a restart? Reach out to support below.

Its status dot flips to grey every so often — usually right after you change something — then comes back on its own.

1This is expected, not a fault. Each time your agent's setup changes, it reloads its messaging gateway to apply the change, and the bot is offline for those few seconds (occasionally a minute or two).
2Anything that changes its config triggers a reload: connecting or disconnecting a channel, switching the model, saving channel settings, or restarting the agent.
3You don't need to do anything — the status dot turns green again on its own once the gateway is back. A message sent during that gap can be missed, so resend it if you didn't get a reply.
4Only worth a look if it stays offline for several minutes without you changing anything — then follow "Bot shows as offline" above.

Instead of an answer, the agent posts "Error code: 400" or a "Non-retryable error" — on Discord, Telegram, or any channel.

1Open the Activity tab on your agent to see the failing request and its error detail.
2Most 400 / 500 errors come from the AI model, not the channel. On the agent card, open the model menu and switch to a different model — a Claude model is the most reliable.
3Send your message again. A fresh model usually clears the error right away.
4Still erroring after switching models? Reach out to support below with the request id from the error (e.g. request_id: 952beaac…).

Account & billing

You redeemed a code but the balance didn't move.

1Redeem codes in the API console — open it from Settings → API access.
2Balance can take a few seconds to refresh; reload Settings to confirm it landed.
3A code only applies once. If it shows as already used but you never saw the credit, contact support.

VM & performance

"Start agent" does nothing, or provisioning ends in an error.

1Verify your email first — an unverified account can't start an agent.
2Check your plan's agent limit on Settings; destroy an unused agent or upgrade for more.
3If the card shows failed, open it and read the error, then click Retry.
4Stuck on the progress bar for minutes? Reload the page — it re-checks the real status.

You updated your agent, and files you'd created on the VM are gone.

1Only `/workspace` — plus your agent's memory, schedules, and channel connections — is preserved across an update. Files saved anywhere else, including your home directory (/root), are reset to the fresh image.
2Move your work into `/workspace` so it survives future updates: mv ~/my-project /workspace/.
3When you SSH in, run cd /workspace before creating files.
4Full details are in the Backups doc under Docs. Files saved outside /workspace before the update can't be recovered.

Still stuck? Email support with your account email and what you tried.