Track spend

agents cost rolls up $ and duration across the local, cross-agent session index. agents output joins that burn to what shipped.

agents cost

Every session is priced when it's scanned — tokens × a per-model rate from an offline, versioned price table. No API calls, no telemetry, and an unpriced model contributes $0, never a crash.

agents cost                   # daily histogram + top sessions + per-agent breakdown
agents cost --since 30d       # last 30 days only
agents cost --by project      # break down by project instead of agent
agents cost --by day --json   # machine-readable daily rollup

Cost  ·  pricing 2026-07-01  ·  since 30d

  $412.30 across 214 sessions  ·  61h 12m total

Daily
  ▂▃▁▄▆█▅▃▂▄▇▆▃▁▂▃▅▇▆▄
  2026-08-04   $38.20
  2026-08-01   $31.05
  2026-07-28   $27.90

Top sessions by cost
  $18.40  a7f3e2c1  claude   fix the flaky routines test           rush-cli   42m
  $14.02  e91b6c04  codex    port the billing module to v2         payments   1h 08m
  $11.75  9c2f0a88  claude   audit the secrets broker teardown     agents-cli 33m

By agent
  claude       $261.40  148 sessions  38h 02m
  codex        $102.15   41 sessions  14h 55m
  droid        $48.75    25 sessions   8h 15m

agents cost is distinct from agents usage, which reports live rate-limit / quota status per agent — a different question, a different command.

Per-session figures

Every session carries its own cost and duration, so you don't need the rollup for a one-off question:

agents sessions --sort cost --limit 10 --json | jq '.[] | {shortId, agent, costUsd, durationMs, topic}'
agents sessions --sort duration --limit 10

agents output

agents cost answers what you spent. agents output joins that burn to what actually shipped — output tokens, PRs, commits — the "was it worth it" axis. It leads with output tokens (real generation), not the cache-inflated total.

agents output                    # burn split + shipped output, actual (cache-discounted) cost
agents output --pricing no-cache # lead with the burn as if caching were off
agents output --json | jq '.burn | {costUsd, costUsdNoCache, inputTokens, cacheReadTokens, cacheWriteTokens}'

Where the harness reports a per-message cache split (Claude, Codex, Gemini, Droid), the rollup separates uncached input, cache-read, and cache-write tokens, and shows the saving caching bought you: caching: actual $84.10 vs no-cache $211.40 (saved $127.30, 60%).