Metrics - Luno OS

Metrics summarize what your agents and workflows did over a time range. They are derived from data Luno OS already records: trace spans, workflow runs and usage. Nothing extra to instrument.

Where to find metrics

View	Scope
Projects → your project → Metrics	Every agent and workflow in the project
Agent → Metrics tab	That agent’s sessions, model calls and cost
Workflow → Metrics tab	That workflow’s runs, plus the model calls and cost of agents it called

Pick 24 hours (hourly buckets), 7 days, 30 days or This month (daily buckets; the UTC calendar month so far, like billing). Buckets are aligned to UTC. The view refreshes every minute. A Space’s Observability tab in Governance opens on This month so its totals match the month’s spend shown there; Studio opens on 24 hours.

What is measured

Metric	Source
Runs and error rate	Agent prompts (`agent_session` spans) and workflow runs. A failed prompt or a `failed` workflow run counts as an error; canceled runs do not
Latency (p50 / p95)	Duration of finished prompts or workflow runs in each bucket
LLM tokens	`llm` spans (one per model step): input, output and cached input tokens, plus the most-called models
Cost	Billed usage events (credits), plus the provider cost reported for your own API keys

Each run is counted in the bucket where it started. Very busy ranges are capped at the most recent 20,000 samples per kind; the view says so when that happens, and counts are then lower bounds.

From the CLI and SDK

luno metrics get                       # linked project, last 24 hours
luno metrics get --range 7d --agent support
luno metrics get --range 30d --workflow daily-digest --project my-project

The command prints the same JSON the dashboard reads (GET /api/projects/:projectId/metrics/timeseries?range=24h|7d|30d|month&agent=|workflow=), also available as platform.projectMetrics.timeseries(projectId, { range, agent, workflow }) in @meetluno/sdk/platform. Any project member can read it.

⌘I