Loading...
Loading...
Single-agent traces are easy. The moment you have a supervisor delegating to four specialists with possible loops, escalations, and final-answer agents, traces become the only way to debug. This course wires Phoenix tracing into a real LangGraph multi-agent insurance copilot, span by span.
Message a mentor about fit, prerequisites, or where to start. Replies come on WhatsApp, usually within a day.
Engineers are learning here from
Trace a real LangGraph multi-agent system end-to-end. Wrap every supervisor handoff, specialist agent, and tool call in OpenTelemetry spans that land in Arize Phoenix. A guided curriculum built around a production-grade insurance copilot, span by span.
Stop debugging supervisor handoffs by guessing. Trace every agent, every tool, every routing decision.
What you'll ship
What you'll learn
Curriculum
The multi-agent observability problem
Why single-agent traces do not scale, what supervisor handoff bugs look like, and where Phoenix fits.
The @trace_agent decorator
Wrap every LangGraph node in an OpenTelemetry span with the metadata Phoenix queries on.
Trace the supervisor and specialists
Apply the decorator to every node in the LangGraph workflow and attach the right attributes for queryable traces.
Correlation, dashboards, replay
Use Phoenix to follow a conversation, build dashboards for handoff health, and replay failing traces as tests.
Who it's for
You shipped a multi-agent system and now you cannot tell why the supervisor sometimes routes wrong. You need a trace tree that surfaces handoff decisions.
You want OpenTelemetry-native tracing so the choice of backend stays reversible. Phoenix is the open option you need to evaluate.
A team handed you a multi-agent system. You need to instrument it before you can put it on call rotation.
FAQ
The Phoenix course covers single-agent and tool tracing fundamentals. This one focuses specifically on multi-agent systems: supervisor handoffs, specialist routing, cross-agent correlation, and replay. The two compose: take Phoenix first, then this for multi-agent depth.
No. The source repo ships with synthetic data: 1,000 customers, 1,500 policies, billing records, claims. The patterns transfer to any multi-agent system, regardless of domain.
Phoenix is OpenTelemetry-native, so the instrumentation you write here works against any OTel backend later. Langfuse adds production tracing plus eval and prompt management. Both have their fit; this course teaches the OTel patterns so you can run them on either.
No. Phoenix runs locally and is free for self-hosted use. The hosted Arize plan is optional for production, but every span call in this course works against either.
Pricing
One subscription unlocks every paid course and workshop replay. Pick yearly or monthly.
Unlock with Pro
You save 47% with regional pricing
Billed annually. Cancel anytime.
Still deciding? Ask Param a question
Trace the agents like you trace microservices. Same patterns, same payoff.
Multi-agent tracing with OpenTelemetry and Phoenix
From $16/mo with Pro