Costs and usage

The cost dashboard shows what the workspace uses: tokens, estimated costs and answer times, by day, model, agent and channel.

Page: Monitoring → Costs and usage in the main menu, for workspace administrators. Prices are set in Settings → Model prices.

What is counted

Every assistant answer already records the tokens the provider billed (input and output) and the time before the answer. The dashboard adds these answers up over 7, 30 or 90 days: estimated cost (tokens × prices below), tokens, answers, average time to answer and the time 95% of answers stay under. Deleted conversations still count, since their tokens were spent. Answers without a token count from the provider are reported.

Model prices

In Settings → Model prices, enter for each chat model the price per million input and output tokens, as shown by your provider (OpenRouter shows it on each model page), and pick the currency. Prices only estimate: nothing is blocked or capped. Cached input tokens are counted at the normal price, so the estimate is slightly high.

API

GET /api/v1/usage-dashboard?days=30&tz=Europe/Paris, and GET/PUT /api/v1/tenants/kv/cost-config with {"currency": "EUR", "prices": {"<model id>": {"input_per_million": 0.4, "output_per_million": 1.6}}}.