Hi everyone, I'm Krzysztof, and I built Tayga. When a request fails in a traced system, you usually have to find the trace, open the waterfall and work out which span actually broke. Tayga does that work while the traces stream in. For every failing or slow request it builds an error story: • the root-cause span, described in a sentence like "checkout could not reach payment" • the request path across services, and the critical path • a diff against the endpoint's normal behaviour over the last hour: operations that are new, missing or slower • the logs that matter When the same problem repeats, it folds into one story group with a trend, so you don't get hundreds of separate traces. Tayga also mines your logs into templates as they arrive and alerts on new, spiking or silent ones, sending them to a webhook or Slack. A few deliberate choices: • No LLM in the root cause. It uses fixed rules over the span tree, so the same input always gives the same answer. • Standard OTLP in. Any OpenTelemetry Collector connects with one exporter, and Tayga can run alongside Jaeger, Grafana or whatever you already use. • Self-hosted, AGPLv3, written in Rust, with ClickHouse for storage and Redpanda for streaming. Watch the 3½-minute tour: https://softberries.github.io/ta... The name comes from my hunting dog, Tajga, who can trace anything, anywhere. I'd like feedback, especially from people running OpenTelemetry in production: what would you need to see before you trusted a root cause from a tool like this?

Tayga

Turn OpenTelemetry traces into root-cause error stories

Turn OpenTelemetry traces into root-cause error stories