August 2026 - AI / Insights

The following updates were released to the Kentik platform for AI / Insights during August 2026.

Features

Runbooks Tool: AI Advisor can now create runbooks based on conversation context.
New AI models available for participating users: Google Gemini 3.7 was added to the models available through Kentik's AI features, and support for OpenAI's Responses API enables access to newer OpenAI models.
AI context for OTT Service Tracking: AI Advisor now understands the OTT Overview, Connectivity, Capacity, and Subscriber views, so it offers more relevant suggested prompts when you analyze OTT services. Also includes a tool for AI Advisor that lists all detected OTT services, categories, and providers
AI context for CDN Analytics: AI Advisor now understands CDN transport, offload/connectivity, OTT service, and performance data, producing more contextual suggested prompts on CDN Analytics pages.
Longer device command timeouts: AI Advisor device command execution now supports timeouts up to 600 seconds (previously about 30 seconds), so long-running commands such as show route complete successfully.
More accurate investigation summaries: AI Advisor investigation summaries now focus on actual findings rather than speculative recommendations, reducing hallucinated content in both short and detailed summaries.

Cause Analysis dimensions and tab behavior: Forensic Packet Analysis cause-analysis results now include the AWS packet address, interface type, service, and ENI dimensions that were previously omitted. The FPA tab also auto-selects when activated, resets on cancellation, and no longer forces you back if you switch tabs manually.
Cause Analysis time window: Enabling Cause Analysis with a "Local" timezone preference no longer shifts the query window into the future, which previously caused Data Explorer to hang with a "start time in the future" error.
Custom Network Context page access: The Custom Network Context settings page is now visible to all authenticated users.
Interface status filtering: AI Advisor's interface tool now correctly filters by operational status (up, down, unknown, testing, notPresent).
Long-running turn timeouts: Complex multi-step AI Advisor turns no longer fail when report generation exceeds the timeout, and the MCP server poll timeout now matches AI Advisor's 600-second response timeout.
Auto Investigation cleanup: Deleting an alerting policy now also removes its associated Auto Investigation configuration, preventing orphaned investigation settings.
Cancelled requests: If your client disconnects while an AI response is being generated, the request to the model provider is now cancelled instead of continuing, and incorrect error responses are no longer returned.