Cost dashboard + price tables

What it does

The Cost dashboard aggregates inference costs by model (or by day) using actual token counts from completions and configured price rates (price per 1K input tokens, price per 1K output tokens). You configure prices once, and the dashboard automatically rolls up costs from the inference log. Use it to track spending, identify expensive models, and optimize resource allocation.

Prerequisites

• At least one chat completion has been logged (guide 01).
• Token counts are captured in the inference log (all models should report usage.prompt_tokens and usage.completion_tokens).

Steps

1. Open the Cost page

Click Cost in the left rail. The page displays two tabs: By Model and By Day.

2. Configure prices (one-time setup)

At the top of the page, click Edit Prices. A form appears with a row for each model:

• Model id — the model name (e.g. qwen2.5:3b).
• Input price — price per 1,000 input tokens (default $0.00).
• Output price — price per 1,000 output tokens (default $0.00).

Fill in your pricing (get rates from your provider, e.g. Ollama is free, HF inference has a price card). Click Save. Prices are stored and applied to all future rollups.

3. View cost by model

Stay on the By Model tab. The table shows:

• Model id — the model name.
• Total input tokens — sum across all inferences.
• Total output tokens — sum across all inferences.
• Cost — calculated as (input_tokens / 1000) * input_price + (output_tokens / 1000) * output_price.

Click a row to expand and see # of inferences and avg tokens per inference.

4. View cost by day

Click the By Day tab. The table shows:

• Date — day in YYYY-MM-DD format.
• Input tokens — sum for that day.
• Output tokens — sum for that day.
• Cost — daily total.

Use the time-range selector (top-left) to view:

• Last 1 hour
• Last 6 hours
• Last 24 hours
• Last 7 days
• All time

5. (Direct API) Fetch costs by model

curl -s -H "Authorization: Bearer $BIGWAVE_TOKEN" \
  "http://localhost:8080/admin/cost/by-model" | jq '.results[0]'

Returns [{ model_id, input_tokens, output_tokens, cost }, ...].

6. (Direct API) Fetch costs by day

curl -s -H "Authorization: Bearer $BIGWAVE_TOKEN" \
  "http://localhost:8080/admin/cost/by-day?since=2024-01-01" | jq '.results[0]'

Returns [{ date, input_tokens, output_tokens, cost }, ...].

Verify

• Set prices for your models (or use $0 for free models).
• Send a few chats; observe token counts in the inference log.
• Open the Cost page; the By Model tab shows the calculated costs.
• Switch to By Day; the costs are aggregated by date.
• Verify a day's total cost matches your manual calculation: (input_tokens / 1000) * input_price + (output_tokens / 1000) * output_price.

Common issues

Symptom	Cause	Fix
"Prices not configured"	No prices set for any models	Click Edit Prices and fill in input/output rates for each model
Cost is $0.00 for all rows	Prices are set to $0 (free tier) OR token counts are not being captured	Check inference log to confirm token counts are present; if missing, the model may not report usage
Cost by day is empty	No inferences exist in the date range	Send chats first (guide 01) or expand the time range
Costs look wrong (too high/low)	Input and output prices are reversed or an order of magnitude off	Double-check your price configuration; prices are per 1,000 tokens
Price changes don't apply to past rows	Prices are applied at query time; historical costs are recalculated	Expected behavior — changing prices updates all past rows automatically on the next page load

Related

• Send a chat message — generates inferences that are costed.
• Inspect inferences (Observability page) — see token counts used in cost calculations.