Kestra setup: GitHub AI triage with human approval

This tutorial builds a self-hosted Kestra workflow for GitHub issue triage. The presenter describes Kestra as an open source automation platform whose YAML workflows can be reviewed in Git and deployed through CI/CD. The setup supports macOS, Windows and Linux; the demonstration uses Windows, Docker Desktop and PowerShell. It starts Kestra with docker compose up -d and opens the interface on localhost without a Kestra cloud account.

The workflow retrieves an issue through the GitHub API and extracts its title, text, author and issue number before passing that context to an AI agent. The agent classifies the issue, assigns a priority and proposes a label and response. The presenter explains that fetching the facts in separate tasks reduces the need for the model to search for or invent details.

A Pause task stops execution for human review. In the demonstrated configuration, it waits until the user resumes the workflow. The presenter says the agent has no GitHub token; later tasks apply labels and publish a comment after approval. The tutorial also covers the YAML editor, topology view, execution logs and JSON output.

Kestra runs locally here, but model inference uses the Gemini API with Gemini 2.5 Flash. This is not an offline model setup. Kestra sponsored the video.