OpenObserve

OpenObserve provides OpenTelemetry-native observability for LLM and agent workloads, capturing calls, tool use, and handoffs as spans with payloads and per-span token cost, alongside existing logs and metrics.

OpenObserve is an established observability platform that in September 2026 launched AI and LLM observability as a first-class capability. The design decision that matters is that agent traces are OpenTelemetry spans, not a separate proprietary format, so LLM calls, tool calls, and agent handoffs land in the same store as your application logs, metrics, traces, and Kubernetes telemetry. Each span carries the prompt and response payload plus a token cost computed from your own model pricing rather than a vendor estimate, which makes cost attribution per request, per feature, or per customer straightforward. Optional LLM-as-a-judge scoring adds quality evaluation, though online evaluations are an enterprise feature. Commercially it is unusual in charging on ingestion rather than seats: cloud is 0.50 dollars per GB ingested and 0.01 dollars per GB queried with unlimited users and no per-seat charge, and self-hosted enterprise is free up to 50 GB per day. The core is AGPL-3.0. The vendor holds SOC 2 Type II and ISO 27001 and notes that LLM observability is in preview.