omnibus-instrument-llm-analytics

You can capture LLM costs and latency by wrapping your LLM clients with the PostHog analytics wrapper, which generates $ai_generation events containing token counts, model usage, and latency metrics.

PostHog LLM analytics supports multiple AI providers and frameworks, allowing you to integrate observability instrumentation into existing LLM stacks to monitor generation traces in your dashboards.

The PostHog LLM analytics wrapper produces $ai_generation events and traces that capture the model, input, token counts, latency, and costs for each LLM generation call.

Setting up LLM analytics tracing without the PostHog SDK is handled by the instrumentation process, which configures the necessary environment variables and handles the initial SDK setup.

You can use LLM observability instrumentation during feature development and PR reviews to ensure all AI-assisted workflow generations are captured with full cost and latency metrics.