# AI Usage

> Reported tokens, costs, subscription limits and the limits of accounting.

Open **Settings → AI → Usage**, or run **AI Usage** from the launcher. Choose a day, week or month to see token totals, a timeline and a breakdown by model or ACP agent.

## What is counted

Saved chat replies contribute usage from the chat database. A separate local numeric log records requests that do not belong to a saved conversation: Quick Actions, voice requests, cloud dictation, extension AI and chat titles.

The log stores date, origin, route, model and reported usage. It does not store those requests' prompts or answers.

## Read cost carefully

Redcast uses the cost a route reports. It does not multiply tokens by a bundled price table. Some providers report exact billed cost, others provide their own estimate, and some provide no price at all.

A missing cost is **unpriced**, not zero. If only some replies have cost data, the summary says how many. ACP agents that report a cumulative session cost are counted by the increase per reply so earlier spend is not charged again to each period.

Deleting chat history can remove the usage attached to those chats. The local Usage page is a convenience for understanding activity; your provider's account remains the place to check billing.

## OpenRouter balance

**Check Balance** requests fresh information when pressed. Depending on the key you supply, it can show key spend and account credit information. Redcast does not continuously poll a balance while the pane is open.

## ChatGPT plan limits

The Codex (ChatGPT) connection shows the last limit snapshot received with replies and its age. It does not make a separate polling request. [ChatGPT sign-in](/docs/ai/chatgpt) explains this route.

## Clear the numeric log

Use **Usage log → Clear…** in the Usage section. This affects the separate numeric log, not the conversations whose replies also contribute to the dashboard. The log retains roughly 400–430 days and is pruned during writes, without a background timer.

A [scheduled agent's spend cap](/docs/agents/background#daily-spend-cap) only works with costs the route reports. A model with no reported price cannot provide a reliable monetary stop condition.