> ## Documentation Index > Fetch the complete documentation index at: https://checkly-422f444a-giordanodaniel-rocky-ai-innvocations.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. # Rocky AI > Learn about Rocky AI — Checkly's AI agent — and configuring and using AI features Rocky AI is Checkly's AI agent. Rocky AI can help you automatically determine user impact and root cause on any failing check using error messages, code, metrics, OTeL traces, trace routes, packet captures: essentially all artefact generated by a check or test. ## Rocky AI Capabilities ### Root Cause and User Impact Analysis * **User Impact**: Rocky AI analyzes the failure and provides you with a list of impacted user groups and features. * **Root Cause**: Rocky AI analyzes the failure and provides you with a root cause analysis. * **Step Summary**: Rocky AI summarizes the steps in the check script, making it easier to understand what the check is doing and where it failed. This typically works for any Playwright based check. [**Learn More**](/resolve/ai-root-cause-analysis/overview) ## Consumption and Invocations By default, Rocky AI automatically analyzes every new *error group*. You can disable automatic analysis in the [AI settings](https://app.checklyhq.com/accounts/rocky-ai) to only spend invocations on manual analyses. Every AI analysis, triggered automatically or manually, counts as one invocation against your plan's quota. If the same analysis result is pushed to multiple alert channels, it is still counted as a single invocation. When your invocation quota is reached, no overages are charged. Invocations refresh at the start of your next billing cycle. You can monitor your current usage and configure automatic analysis in the [AI settings](https://app.checklyhq.com/accounts/rocky-ai). ### Error groups An **error group** is a deduplicated bucket of check failures that share the same underlying error. Instead of treating every failure as a separate event, Checkly cleans the raw error message and clusters failures with matching cleaned messages into a single group per check. A group records when the error was first and last seen, so it reads as "this specific error has occurred on this check, starting on date X, most recently on date Y." A new group is created the first time a check fails with an error that doesn't match any existing group — and that's the event that triggers an automatic analysis. Because minor variations are normalized away, repeated failures of the same error reuse the existing group and don't consume additional invocations. ## Data Privacy and Security * All AI features are designed to operate on the data you provide or the data generated by your checks. This data is already stored in Checkly, and we do not send any additional data to third-party AI providers. * Any provided [environment secrets](https://www.checklyhq.com/docs/api-checks/variables/) are scrubbed before being sent to the AI provider. * We do not train any custom models on your data. * Any data used by the AI features is processed in real-time and not stored for future training. ## Built-in and Custom Models By default, Checkly uses OpenAI's GPT-5.1 model for AI features. However, you can also bring your own model (BYOM) and provider to Checkly. To do this, you need to set up a custom AI provider in the [AI settings screen](https://app.checklyhq.com/settings/account/ai). Note that custom model providers need to have the following capabilities: * Able to handle tool calling. * Able to stream responses. * Available for the Node.js-based AI SDK, [see the list of providers here](https://ai-sdk.dev/providers/ai-sdk-providers) The Bring Your Own Model (BYOM) feature is available for Enterprise customers only.