An AI assistant for your monitoring

Bring your own key

The assistant runs on the AI provider you choose: Anthropic, OpenAI, Google Gemini or OpenRouter. You add your own API key in the app, choose a model, and pay the provider directly for what you use. Uptime Dog doesn't add a margin or a separate AI charge.

Keys are encrypted at rest and never sent back to the browser; the app only shows the last four characters so you can tell which key is set.

What you can ask

• "Which monitors have been flaky this week?"
• "What happened with the checkout incident last night?"
• "Who's on call this weekend?"
• "Is there any maintenance planned?"
• "Show me errors from the payments service around 14:00."
• "Monitor https://example.com/health every minute and check it returns ok."

Read-only tools, scoped to your team

The assistant answers by calling tools that read your team's data, and only your team's: monitors and their recent checks, response times by location, incidents and their timelines, uptime and mean time to recover, who is on call, maintenance windows and logs. It can't change anything. When you ask it to set up a monitor, it drafts one for you to review, and nothing is created until you click to create it.

Incident investigation

With auto-investigate on, the assistant looks into each new automatic incident as it opens. It reads the monitor's recent checks and history, looks for other monitors that are down, checks for recent maintenance and searches logs from the 15 minutes before the failure. Then it adds a short internal note to the incident saying what is most likely wrong and what to check first.

The note is internal, so it never appears on a status page. It's a starting point for whoever picks up the incident, written as the incident opens, and the assistant is told not to speculate beyond the evidence. You can turn auto-investigate off and use the assistant only when you ask it something.

What it costs

The assistant is available on every plan, including Free. The only cost is what your AI provider charges for the tokens used, at your own rates on your own account. An investigation uses a handful of tool calls; a question in chat usually uses fewer.

Getting the most from it

The assistant is only as good as what it can read. Name monitors clearly, send logs with a service name and a level, and post incident updates as you work. The same habits make your status page and reports clearer to people too.