Trace complete agent workflows—every decision, tool call, and model switch—across OpenAI, Anthropic, and Google in one unified view.
Built for multi-agent systems. Debug agent behavior, not API logs.
When agents call other agents, switch between models, and use tools—you can't see what's actually happening.
Kalibr traces the complete workflow, showing agent decisions, tool usage, and model switches in one place.
See every agent call in a hierarchical tree. Which agent called which. What they decided. What tools they used. Complete parent-child relationships.
One workflow switches between GPT-4, Claude, and Gemini? We capture all three in a single unified trace across all providers.
Track when agents use external tools—vector DBs, APIs, calculators. See input data, output results, and execution time.
Know what complete workflows cost by agent and operation—not just individual API calls. Track spend by business outcome.
Agent A in Service 1 calls Agent B in Service 2? Context propagates automatically. Full tracing across your infrastructure.
One import. Zero refactoring. Works with LangChain, CrewAI, AutoGen, custom agents—anything.
pip install kalibrimport kalibr
from openai import OpenAI
from anthropic import AnthropicEvery agent call, tool use, and model switch is automatically captured with full context and relationships preserved.
✗ Track individual API calls
✗ Show one provider at a time
✗ Miss agent relationships
✗ No tool call visibility
✗ Can't see workflow structure
✓ Trace complete agent workflows
✓ All providers in one trace
✓ See which agent called which
✓ Track tool usage with full context
✓ Visualize workflow hierarchy
The difference: We trace agent behavior and decisions, not just API requests. See how your multi-agent system actually works.
Start locally with the open-source SDK. Upgrade to Kalibr Enterprise for hosted analytics and production benchmarks.