Features
From single-agent tracing to multi-agent topology — everything you need to debug and monitor your AI agents in production.
Agent Trace Timeline
Every step your agent takes — thinking, tool calls, decisions, outputs — in a chronological timeline. Click any step to inspect the full input, output, and intermediate state.
- → Expand thinking chain for any step
- → Filter by tool type or status
- → Share trace links with your team
Per-Step Token & Cost Breakdown
Know exactly how many tokens each step consumed — not just the conversation total. Identify costly thinking steps and optimize your prompts.
- → Cost per model, per step, per tool
- → Budget alerts and cost thresholds
- → Export cost reports
Failure Replay
Replay any failed trace step-by-step. See exactly what the agent saw, what it decided, and where it went wrong. No more guessing from raw logs.
- → One-click replay from error dashboard
- → Step-by-step playback with pauses
- → Compare successful vs failed traces
Multi-Agent Topology Map
Visualize how your agents collaborate. See the communication graph, handoff points, and message flow between agents in real-time.
- → Agent relationship graph
- → Cross-agent trace linking
- → Handoff failure detection
Real-Time Alerts
Get notified when your agents behave unexpectedly — high latency, excessive token usage, error loops, or budget exceeded.
- → Slack/email/webhook notifications
- → Custom alert thresholds
- → Token budget auto-stop
Framework-Agnostic SDK
Works with any agent framework — LangChain, AutoGen, CrewAI, LangGraph, or custom implementations. One decorator to trace everything.
- → Python & TypeScript SDK
- → Webhook fallback for non-SDK clients
- → OpenTelemetry compatible