Usage & Cost

Lattis attributes every request and tracks token usage broken down by app, model, and project. For paid cloud models it also estimates the dollar cost from the provider’s reported token counts.

What’s tracked

• Per-app — which client/tool made the request.
• Per-model — local or cloud model id used.
• Per-project — work attributed to a project context.

Token counts come straight from the inference engine (local) or the provider’s usage fields (cloud), so the breakdown reflects actual consumption.

Cost estimation

Cloud providers report token counts, not dollars. Lattis multiplies those counts by a per-model price table (split by billing category — standard input, output, cached-input reads, and cache writes) to produce an estimate. Local models are free, so no cost is shown for them.

Because pricing is an estimate derived from a built-in table, treat the numbers as a close guide rather than an invoice.

Why it matters

A single endpoint serving many models and providers makes it easy to lose track of where tokens — and money — go. Per-project attribution answers “what is this project costing?” and per-model breakdown shows whether a task belongs on a local model or a frontier cloud one.

See Cloud Providers to connect a paid provider, and the Control API for the daemon snapshot that surfaces these figures.