ClickStack Demo Day (2026-08-28) - LLM observability out of the box

Teams running LLM apps and coding agents already send their telemetry to ClickStack, but the product had no LLM awareness at all: no chat rendering, no token or cost concepts, no model analytics. There's now an LLM dashboard alongside the existing preset dashboards for ClickHouse and Kubernetes. Anything exporting through the common conventions, OpenLLMetry or OpenInference, so the OpenAI and Anthropic SDKs, the Vercel AI SDK, LangChain, Claude Code and opencode, is read at query time out of the traces and logs you already have. There are no dedicated tables and no ingest-time processing. The dashboard covers token usage, cost, model calls, tool calls, cache hits and response times, and breaks usage down by user when a user attribute is present. A latency view prioritizes the AI-related spans in a trace, so you can see which model call was slow without picking through everything else. A sessions view groups a conversation's spans into a single lookup, and clicking through gives you the span detail, including the prompt where it's recorded. Logs are searchable with the same session filter. This is deliberately narrower than a dedicated LLM observability tool such as Langfuse, which does considerably more, evaluations included. What it gives you is the common monitoring, error rate, cost, tokens and latency, without leaving your existing telemetry. It's experimental and not recommended for production use yet. Part of the ClickStack Demo Day series. See the full post at https://clickhouse.com/docs/use-cases/observability/clickstack/demo-days/2026/2026-08-28