aiusage

CLI tool that reports AI API usage and costs for OpenAI, Anthropic, and OpenRouter. Two modes:

• API key usage — token counts and dollar costs from provider admin APIs, broken out by provider, API key, date, and model
• Subscription quota — live quota percentage and reset times for Claude Code and Codex subscriptions

Setup

API key usage (requires admin API keys)

export OPENAI_ADMIN_KEY="..."
export ANTHROPIC_ADMIN_KEY="..."
export OPENROUTER_ADMIN_KEY="..."  # management key, not regular API key

Subscription quota (requires CLI login)

Log into Claude Code (claude) and/or Codex (codex) on your machine. aiusage reads their stored OAuth credentials automatically.

Usage

# Show both subscription quota and API key usage (last 7 days)
aiusage

# API key usage only
aiusage api
aiusage api --provider openai --start 2026-03-01 --end 2026-03-05
aiusage api --group-by model
aiusage api --json

# Subscription quota only
aiusage subs
aiusage subs --provider anthropic
aiusage subs --force   # bypass 5-minute cache
aiusage subs --json

Pricing

Dollar costs come from provider cost APIs when available (cost_src=truth_allocated). When not available, costs are estimated from model rates (cost_src=estimated_pricing).

Pricing is resolved in order: built-in fallback → LiteLLM (fetched live, cached 24h) → --pricing-file (manual override). The report header shows the active pricing source and its date.

# Show current pricing table and source
aiusage prices

# Compare built-in rates against LiteLLM (writes pricing-override.json if mismatched)
aiusage prices --check

# Manual override
aiusage api --pricing-file ./pricing-override.json

Build

go build ./cmd/aiusage

Notes

• OpenRouter returns actual USD costs directly; no pricing estimation needed. Data covers last 30 days only.
• API endpoints are configurable via --openai-usage-path, --anthropic-usage-path, etc.
• If pricing is missing for a model, cost is zero and a warning is shown.
• Subscription quota is cached for 5 minutes to avoid rate limits. Use --force to bypass.
• LiteLLM pricing is cached for 24 hours. If LiteLLM is unreachable, built-in fallback rates are used.