Skip to content
agents-in-a-box

Observability

How to see what your agents are doing — what they cost, what’s running, why a process exists, and how to ship it all to Grafana Cloud.

ToolAnswersSurface
Usage analytics (burndown)“What am I spending — by day, project, model?”i in the TUI · ainb burndown
abtop”What agent processes are running right now?”t in the TUI · ainb abtop
witr”Why is this process running — what spawned it?”w in the TUI · ainb witr
OpenTelemetry → Grafana Cloud”Ship metrics/logs/traces off-box to dashboards”ainb otel setup

Local vs remote

The first three are local, in-the-moment views — open them in the TUI and read the answer now, no setup. OpenTelemetry is the off-box, historical path: a local Grafana Alloy collector forwards Claude Code / Codex telemetry to Grafana Cloud for cost dashboards, latency percentiles, and trace retention.

┌─ Local (now) ────────────┐     ┌─ Remote (history) ───────────────┐
│ burndown  usage + cost   │     │ OTEL → Alloy → Grafana Cloud      │
│ abtop     live processes │     │ cost, latency p95, traces, prompts│
│ witr      causality      │     │                                   │
└──────────────────────────┘     └───────────────────────────────────┘

Start with the OpenTelemetry guide for the Grafana Cloud setup and example dashboards.