Tracing
The foundation for evals.
Capture full agent trajectories — every tool call, LLM hop, and decision branch — so your evals have the signal they need to score accurately.
- One-line instrumentation for every major agent framework
- Works with any LLM provider out of the box
- Captures spans, and metadata automatically
agent.py
python| 1 | from pandaprobe.integrations.google_adk import GoogleADKAdapter |
| 2 | |
| 3 | # Call once at startup — before creating any agents |
| 4 | adapter = GoogleADKAdapter( |
| 5 | session_id="session-abc", |
| 6 | user_id="user-123", |
| 7 | tags=["production"], |
| 8 | ) |
| 9 | adapter.instrument() |
| 10 | |
| 11 | # All ADK runners are now fully traced |
| 12 | # — tool calls, LLM hops, token usage, TTFT |
