instrument-llm-analytics

Track LLM analytics costs and latency by instrumenting LLM calls to emit $ai_generation events via PostHog. This captures model, provider, input/output tokens, latency, and cost data automatically to enable end-to-end observability across your stack.

Instrument OpenTelemetry traces for LLM calls by applying PostHog instrumentation with minimal code changes. This approach auto-generates $ai_generation events without restructuring your architecture, linking traces and spans to identified users for debugging and performance optimization.

Associate LLM analytics events with identified users by instrumenting LLM calls through PostHog. This generates $ai_generation events linked to users and traces, enabling per-user analytics, session tracking, and debugging across multiple providers and frameworks.

Yes, PostHog LLM analytics instrumentation broadens coverage by supporting multiple providers and frameworks as listed in its configuration. It applies OpenTelemetry instrumentation across your codebase to capture LLM generations regardless of the specific provider used.

Capturing LLM generations requires configuring environment variables and installing the required SDKs for PostHog and OpenTelemetry. The setup enforces environment variable configuration to enable minimal code changes for instrumenting LLM calls.

LLM traces may lack token data if environment variable configuration is incomplete or the required SDKs are not installed. Properly instrumenting LLM calls via PostHog ensures $ai_generation events capture input/output tokens, latency, and costs correctly.