Usage Analytics

The Usage view shows where your tokens and money go — all computed from the persistent activity history (also visible in the Models tab).

Breakdowns

Four grouped summaries, each with calls, input/output tokens, and estimated cost:

• By provider — e.g. OpenRouter vs local GPU
• By model — e.g. anthropic/claude-3.7-sonnet vs gpt-4o-mini
• By day — 2026-09-28 buckets
• By month — 2026-09 buckets

A search box filters everything by provider, model, execution mode, or task description. A Sort selector orders groups by Cost, Calls, Tokens, or Name.

Credits and balances

• + Add Credit records a manual top-up (e.g. $20 for OpenRouter) — persisted locally.
• The balance table shows per provider: credits added, estimated spend, and the remaining balance (color-coded).
• ⟳ Live Balance fetches the actual remaining credit from a provider API. Currently supported: OpenRouter (GET /v1/credits, using the API key stored on its enabled fallback tier). More providers can be added as they expose endpoints.

Where the data comes from

Every AI call records: timestamp, provider, model, input/output tokens, estimated cost (built-in per-model pricing table), execution mode, and router profile. Every auto-continue step records its unique ID and action. Everything persists in a local SQLite database (activity.db) — see the Models and AC tabs.