Kestra tutorial: Docker setup for an AI news digest

This sponsored tutorial builds a self-hosted Hacker News digest with Kestra. The workflow fetches story IDs through the Hacker News API, retrieves the first story, asks an LLM for a two-sentence summary and posts the result to Slack through an incoming webhook. The presenter uses Anthropic Claude 3 Haiku and explains how to change the YAML provider to Ollama for a local LLM. The demonstration completes with Claude; it does not establish how an Ollama model performs.

The setup uses Docker Compose to start Kestra and PostgreSQL 18, with the Kestra interface available at localhost:8080. The presenter adds credentials to an .env file, creates a Slack app and configures its webhook. YAML defines the tasks and supports a daily 8:00 a.m. schedule or an on-demand webhook trigger, though the successful demonstration uses manual execution. Running the orchestrator locally still leaves this example dependent on external Hacker News and Slack services.

Much of the walkthrough covers failed executions. The presenter checks logs and the task graph, corrects secret encoding and environment-variable mistakes, then resumes the failed task. In this configuration, the Slack webhook uses base64 encoding while the Anthropic API key stays unencoded. The tutorial distinguishes restarting a failed task from replaying the flow as a new execution. It closes with the speaker's overview of enterprise access controls and deployment options.