OpenObserve analyzed Claude Code traces to determine the time-to-first-token for each API call, finding that 83% of turn time is spent waiting on the model API, 18% is spent on tools, and each API call averages 33 seconds with a 14-second time-to-first-token.

OpenObserve is an open-source, high-performance platform for unified observability — bringing logs, metrics, traces, and real user monitoring into a single system so teams stop juggling disconnected tools. With over 20K GitHub stars and 8K+ active deployments at organizations like Walmart, Banorte, Sony Music, and Hitachi Ventara, it's proven at scale. Built on open standards like OpenTelemetry with S3-native storage, it avoids proprietary lock-in while delivering up to 140x better storage efficiency than Elastic and 30x better compute efficiency than Splunk, at roughly 4x lower cost than Datadog. Beyond traditional observability, OpenObserve is expanding into AI-driven operations. Its AI SRE capabilities auto-correlate signals across logs, metrics, and traces to speed root-cause analysis and cut alert fatigue, shifting teams from reactive dashboards to proactive incident response. As LLM adoption grows, OpenObserve also supports LLM observability, tracking prompts, completions, latency, token usage, and quality signals — letting teams monitor and optimize AI applications through the same unified pipeline used for the rest of their stack.