Operations
AI assistant
Explain failed jobs and site errors in plain words, with concrete next steps — powered by Claude.
When something fails, Explain with AI asks the assistant (Claude, by Anthropic) what went wrong and what to do next.
Where it is
- Activity: next to every failed or refused job — a deploy whose build failed, a certificate that could not be issued, a package that would not install.
- A site's Logs pane: after Load, below the log lines — PHP errors, 502s, a crashing app process.
The answer names the most likely cause and the next steps, in your language. The assistant never changes anything: you decide, and high-risk actions still need your approval passkey.
What is sent
Only the failure itself: the job's log or the loaded log lines, with the job type, the site's type and PHP version, the server's operating system and the agent version. Before anything leaves the panel, private keys, passwords, tokens, Authorization headers, long keys and hashes and email addresses are removed. Log content is treated as data: instructions hidden in logs are ignored.
Credits
Each answer uses one AI credit; asking the same question again within an hour is free. Your plan includes a monthly number of credits (shown under each answer); they renew on the first of the month.
Updated October 18, 2026