Agent Manager with Google Antigravity/Production Multi-Agent Systems
Lesson 2 / 5 · 11 min
Walkthrough

Observability and Tracing

Knowing what every agent did, why, and how much it cost.

When a roster run goes wrong, you need a trace: every agent decision, every tool call, every prompt and response, with timing and cost. Antigravity emits OpenTelemetry traces by default; ship them to your existing observability stack and you can debug agent runs the same way you debug services.

Steps · 0 / 3 done

  1. Configure trace export

    Point Antigravity at your OTLP collector.

    antigravity config set otlp.endpoint https://otel.your-stack.com:4317
antigravity config set otlp.headers 'Authorization=Bearer $OTEL_TOKEN'
    Verifyantigravity config show prints the endpoint; first run after the change appears in your tracing UI.

  2. Trace correlation

    Tag each run with the originating issue, PR, or human user. Makes filtering trivial.

    antigravity run start --tag issue=42 --tag user=alice@example.com pipeline.json
    VerifyTrace appears with the tags as searchable attributes.

  3. Per-step cost in the trace

    Antigravity emits a spend metric per step. Plot it next to latency.

    # In your dashboard, query:
antigravity_step_cost_usd{role="builder"}
    VerifyBuilder cost shows as a time series; you can spot regressions.
Check your understanding
Q1. Why ship Antigravity traces to your existing observability stack instead of using a separate UI?
· Tick off the 3 step(s) above.
· Score 100% on the quiz.