Root-cause in seconds
Reproduce any failure with full context
Every LLM call, tool invocation, and decision is logged as a structured event. Navigate the event tree to find exactly where something went wrong — then fix it.
Full event timeline
LLM calls, tool calls, decisions, and errors — all logged in chronological order with metadata.
Parent-child event trees
Events are linked via parent_event_id, giving you a tree view of exactly how execution flowed.
Per-call latency
Latency is measured at the individual call level — see where time is spent, not just end-to-end.
Token-level cost
Input and output token counts and cost are captured per event so you know the exact cost driver.
Model metadata
Model name, temperature, max_tokens, finish reason — all stored alongside each LLM event.
Tool call inspection
See the full input and output of every tool call. Debug hallucinated tool arguments in seconds.
At a glance
- Automatic event capture — no manual logging
- Tree view with parent-child event relationships
- Per-call latency from event.data.meta.latency
- Input / output and token cost on every event
- Supports OpenAI · Anthropic · Gemini · Groq · LangChain
Common questions
How are streaming responses handled?
Streaming is fully supported for OpenAI and Anthropic. The SDK assembles the response from chunks and records the final token count, cost, and latency once the stream closes — so your event data is complete even for long streaming calls.
Can I see the full inputs and outputs of tool calls?
Yes. Tool call arguments and return values are logged as structured JSON on each tool_call event. You can inspect them in the event detail drawer in the session view — useful for debugging hallucinated tool arguments or unexpected return shapes.
How is latency measured?
Latency is measured from the moment the SDK dispatches the API request to when the response is fully received (or the stream closes). It is stored in seconds in data.meta.latency on each event. End-to-end session duration is tracked separately on the session record.
More general questions? See the full FAQ →
Ready to get started?
Connect your first agent in under 5 minutes. Free to start, no credit card required.
Next: Experiments