Analytics

The Analytics screen shows the spend recorded in the proxy's store, as counters and tables. It refreshes every 30 seconds. The screen sends no date range, so the figures cover every record in the store.

Counters

Tile	What it shows
Total Requests	Number of recorded requests
Total Spend	Sum of the recorded cost, in USD
Prompt Tokens	Sum of input tokens
Completion Tokens	Sum of output tokens

Tables

• Spend by Model and Spend by Provider: requests, cost and average latency per group, from GET /api/v1/analytics/spend
• Cost Efficiency: requests, total cost, average cost per request and average tokens per request for each model, from GET /api/v1/analytics/cost-efficiency

Each table has a button that saves its rows as a CSV file. The screen has no charts.

Budget

The cost of a request is computed from the per-model prices in core/pricing.py.

• budget.daily_limit is the daily cap. A request whose estimated cost would take the day's total over it is refused with HTTP 402
• budget.soft_limit is a warning threshold for the budget webhook
• The day's total is saved in the store and read back at startup

The shipped config.yaml also sets budget.fallback_to_local_on_limit. No code reads that key: there is no fallback to a local model when the budget is exhausted.

API

# Spend grouped by model, from a date
curl "http://localhost:8090/api/v1/analytics/spend?group_by=model&from=2026-01-01" \
  -H "Authorization: Bearer your-key"

# Top models by cost
curl http://localhost:8090/api/v1/analytics/spend/topmodels \
  -H "Authorization: Bearer your-key"

group_by accepts model, provider, key_prefix or date; any other value is treated as model. from and to bound the period by date.