Docs/Integrations/Crewai

CrewAI Integration

ANTS Platform gives you full observability for CrewAI — traces, cost, latency, agent attribution, and error analysis — with zero changes to your crew logic.

Choose your integration

Native OTel (CrewAI Enterprise)

Zero-code. Set two env vars and point CrewAI's built-in OTel exporter at our endpoint. Recommended for CrewAI Enterprise customers.

SDK + EventListener

For self-hosted CrewAI or projects that want full Python-SDK control. Three-line EventListener registration in crew.py.

What gets captured

Both approaches produce the same trace shape in ANTS Platform:

Crew executions — kickoff to completion, with framework:crewai and crew:<name> tags
Agent steps — role, goal, tool inventory, per-agent cost/latency rollups
LLM calls — model, prompt/completion messages, tokens, cost (computed server-side)
Tool usage — humanised tool name, arguments, output
Task execution — task description, expected output, actual output
Errors — attributed to the exact agent / task / LLM call

code

MarketIntelligenceContentPipelineCrew (root) ├── research_market_signals (task) │ └── Senior Market Research Analyst (agent) │ ├── LLM Call (gpt-4o-mini) │ └── Search the internet with serper ├── draft_content_strategy │ └── Content Strategist → LLM + tools └── editorial_review_and_publish_plan └── Editor-in-Chief → LLM

ℹ

Already running CrewAI Enterprise? Native OTel is the fastest path — no code changes, no pip install. Start with the Native OTel guide.