exploring-llm-costs

Analyze LLM costs in PostHog by aggregating total spend, per-model and per-provider breakdowns, token usage, and cache metrics from $ai_generation and $ai_embedding events. It reads cost properties like $ai_total_cost_usd and $ai_model to quantify spend and identify waste.

Yes, you can break down LLM spending by user, model, provider, and trace in PostHog. The analysis aggregates cost data from ingested $ai_generation events to surface actionable insights and identify your most expensive models or individual traces.

Tracking LLM cost analytics requires PostHog ingestion of $ai_generation and $ai_embedding events with properties like $ai_total_cost_usd, $ai_model, $ai_provider, $ai_input_tokens, and $ai_output_tokens. These fields enable detailed spend aggregation and cache-efficiency analysis.

Create dashboards and alerts for LLM cost spikes by materializing the aggregated spend and token usage insights into PostHog monitoring views. You can generate shareable cost breakdowns and set up alerts based on per-model or per-user spending thresholds.

Yes, this approach works for optimizing token usage and cache efficiency by analyzing $ai_input_tokens, $ai_output_tokens, and related cache metrics. It quantifies spend across AI features to help you identify waste and drive cost-efficient usage patterns.