What is obsidian?

Obsidian is an open source tool for watching over software systems in production, built as a free alternative to paid services like Datadog or New Relic. It collects three kinds of information companies use to understand how their applications are behaving: traces, which follow a single request as it moves through different parts of a system, metrics, which are numeric measurements over time like response speed or error counts, and logs, the text messages applications write out as they run. What sets this project apart is a group of five separate AI agents that work together to make sense of all that data. One watches for unusual patterns in metrics, one studies traces to find slow parts of a request, one looks for clusters of similar errors in logs, one groups related alerts together so a team is not flooded with duplicate notifications, and a final agent combines everyone else's findings into a plain written explanation of what likely caused an incident. The system is built from several standard pieces working together, including a message queue, a background task processor, a fast column based database for storing telemetry data, and a web dashboard built in React that updates live. It follows OpenTelemetry, a widely used industry standard for sending this kind of data, so it should work with services already set up to send data that way, and it can send alerts to Slack, PagerDuty, email, or a custom webhook. Setting it up requires Docker and a reasonably capable machine, with the README suggesting at least 8 gigabytes of memory and four processor cores. It is aimed at engineering and operations teams who want to self-host their application monitoring instead of paying for a hosted service, and it is released under the MIT license.