API Cost Tracker#

Analyze and optimize AI API costs across multiple providers with detailed breakdowns, trend detection, and actionable savings recommendations.

Quick Start#

bash

# Analyze OpenRouter usage (from activity page export)
python3 scripts/api_cost_tracker.py openrouter --file activity.json

# Analyze OpenAI usage (from billing export)
python3 scripts/api_cost_tracker.py openai --file usage.json

# Analyze from environment (auto-detect provider from API keys)
python3 scripts/api_cost_tracker.py auto --days 30

# Cost breakdown by model
python3 scripts/api_cost_tracker.py openrouter --file activity.json --by model

# Cost breakdown by day with trend analysis
python3 scripts/api_cost_tracker.py openrouter --file activity.json --by day --trends

# Find most expensive requests
python3 scripts/api_cost_tracker.py openrouter --file activity.json --top 20

# Compare current vs optimized (model substitution analysis)
python3 scripts/api_cost_tracker.py openrouter --file activity.json --optimize

# Set budget alert threshold
python3 scripts/api_cost_tracker.py openrouter --file activity.json --budget 50.00

# Output as markdown report
python3 scripts/api_cost_tracker.py openrouter --file activity.json --output markdown

# Output as JSON
python3 scripts/api_cost_tracker.py openrouter --file activity.json --output json

Supported Providers#

ProviderInput FormatAuto-detectOpenAIBilling CSV/JSON export, API responsesOPENAI_API_KEYAnthropicUsage API, console exportANTHROPIC_API_KEYOpenRouterActivity JSON, API responsesOPENROUTER_API_KEYGoogle AIBilling exportGOOGLE_AI_API_KEYGenericCSV with columns: timestamp, model, tokens_in, tokens_out, costN/A

Analysis Features#

1. Cost Breakdown — by model, day, week, feature/tag, request type
2. Trend Detection — spending velocity, anomaly detection, projected monthly cost
3. Optimization Report — model substitution suggestions, caching opportunities, prompt compression candidates
4. Budget Alerts — daily/weekly/monthly thresholds with projected overrun warnings
5. Top Spenders — most expensive individual requests or sessions
6. Model Comparison — cost-per-quality analysis using common benchmarks

Output Formats#

• Terminal (default) — colored tables and charts
• Markdown — report suitable for documentation
• JSON — structured data for programmatic use
• CSV — spreadsheet-compatible export

How It Works#

The script:

1. Reads usage data from the specified source (file, API, or environment)
2. Normalizes all entries to a common format (timestamp, model, input_tokens, output_tokens, cost)
3. Applies current provider pricing to calculate/verify costs
4. Groups and aggregates by the requested dimension
5. Runs optimization analysis comparing current models to cheaper alternatives
6. Generates the report in the requested format

Pricing Database#

Built-in pricing for 50+ models (updated March 2026). Override with --pricing custom_prices.json.

Requirements#

• Python 3.8+
• No external dependencies (stdlib only)