Trace-lit — Agent Observability

# install

pip install "tracelit-sdk[kafka]"

# instrument any agent — one line

import trace_lit

trace_lit.configure(kafka_brokers=["app.trace-lit.com:9093"], api_key="sk-demo-abc123")

@trace_lit.trace(agent_name="my-agent", framework="langchain")

def run(query: str) -> str:

...

✓ trace sent latency=1.4s cost=$0.004 tokens=1203 status=success

WHAT YOU GET

See every agent hop, tool call, and LLM invocation as an interactive graph. Click any node to inspect inputs, outputs, and timing.

Track token spend per agent, per workflow, per team. Know exactly which pipeline is burning your budget before your CFO asks.

Automatically classify failures as hallucination, tool error, timeout, or context overflow — with instant alerts.

Run quality scoring on every production trace — not just in test. Catch regressions before users do.

Works with LangChain, LlamaIndex, CrewAI, AutoGen, or any custom agent. No lock-in, ever.

Single docker compose up. Deploy on your own infrastructure for full data control and compliance.

Visibility into your
AI agents