Monitor every AI call across every provider
Track latency, token usage, costs, and prompt quality for OpenAI, Anthropic, and other LLM providers. Every AI call is a span in your distributed trace.
AI observability built into your APM
No separate tool needed. LLM calls are just another span type in your traces.
Latency tracking
See time to first token, total generation time, and streaming latency for every LLM call.
Cost monitoring
Track token usage and estimated costs per model, per endpoint, per user. Set cost alerts before bills surprise you.
Provider dashboard
Compare performance across OpenAI, Anthropic, Cohere, and others. See which models are fastest and cheapest for your use case.
Trace context
Every LLM call is a span in the parent trace. See the full request -- API handler, database query, LLM call -- in one waterfall.
Automatic instrumentation
SDK auto-instruments OpenAI and Anthropic client libraries. No manual span creation or wrapper functions needed.
Agent observability
Trace multi-step agent workflows. See tool calls, reasoning chains, and decision points in the trace waterfall.
How it works
Install the SDK
Add Tracekit to your app. LLM client libraries are auto-instrumented -- no wrapper code needed.
AI calls become spans
Every LLM call is captured as a span with model, tokens, latency, and cost. Nested in the parent request trace.
Monitor and optimize
Use the LLM dashboard to track costs, compare providers, and spot latency regressions before users notice.
Built-in vs bolt-on
Stop flying blind on AI costs and performance
See every LLM call in the context of the request that triggered it.