Intelligence

Cost Attribution

Know what each feature costs

npx promptreports-cli costs

Groups AI costs by model, git commit, or provider. Shows unit economics and cost anomalies.

costs answers three questions: which models are you paying the most for? which git commits cost the most AI tokens to produce? which providers account for your spend? Groups costs across your entire stack (Claude Code + all providers) and shows per-unit breakdowns. Use --by commit to see your cost-per-PR.

Prerequisites

Session files in ~/.claude/projects/
Git repo (for --by commit)
Env vars for providers (for --by feature)

Flags & Options

Flag	Description	Default
--days N	Lookback	7
--by <group>	Group by: model \| commit \| feature	model
--json	JSON output	—

Examples

By model

npx promptreports-cli costs --by model

Opus vs Sonnet vs Haiku spend.

By commit

npx promptreports-cli costs --by commit --days 30

Cost of every commit last month.

By feature

npx promptreports-cli costs --by feature

Which providers dominate your spend?

Output

Table grouped by selected dimension + totals (Claude Code + Providers = Grand Total).

  Cost by Model
  Model                Cost      Turns   Unit Cost   %
  claude-opus-4        $3.45      12     $0.29       45%
  claude-sonnet-4      $2.10      18     $0.12       27%
  claude-haiku-4.5     $0.68      34     $0.020      9%
  gpt-4o-mini          $1.42      23     $0.062      19%

  ──────────────────────────────────────────────────────
  Total Claude Code:   $7.65 (5 sessions, 7d)
  Total Providers:     $12.45 (5 services, 7d)
  Total Burn Rate:     $20.10 / 7d
                       $86.14 / month (projected)

What it reads and writes

Reads

~/.claude/projects/**/*.jsonl
.env.local
git log

Writes

Nothing (read-only)

API calls

Provider APIs (same as providers)
Git log for commit grouping

Free vs Pro usage

Free tier

Identify your most expensive model — usually Opus
Spot cost-per-commit outliers for retro
Share breakdown with CFO or manager

Pro tier

Upgrade

Tag sessions with feature labels → cost per feature
Attribute costs to specific team members (who's burning the most Opus?)
Alert when any single commit exceeds cost threshold
Export to QuickBooks / Stripe Revenue categories

Pro tips

--by commit reveals the real cost of large refactors
If Opus is 70%+ of cost, run `models` to find downgrade candidates
Unit Cost (cost per turn) is the key efficiency metric

Free to try

Install the whole toolkit.

One command unlocks all 23 tools.

Start Free Back to all tools