--- title: AI usage and Fuxam token budget description: >- Understand how Fuxam meters AI features against a monthly token budget — how usage is counted per billing period, when it resets, who used what in AI activity, and why premium models are unavailable in some tools. sidebar: order: 3 --- Fuxam's AI features — Docu-chat, the Workbench copilot, the global-search **Concierge**, and AI chat — draw on a shared institution **AI budget** measured in **Fuxam tokens** (credits). Every AI request an institution makes spends credits from that budget. This page explains how the budget is metered, when it resets, how to read the usage and per-user activity views in Settings, and why some premium models cannot be selected in certain tools. You manage all of this under **Settings → Storage & AI** (usage, budget, and AI activity). No configuration is required to start — the budget and metering apply automatically to every AI feature. :::info **Fuxam tokens** (also shown as *credits*) are the internal unit Fuxam uses to meter AI usage. They are distinct from the raw model tokens a provider counts — one AI request typically spends several credits depending on the model and the length of the request and response. ::: ## Why it matters AI usage has a real per-request cost, and a single institution can run many AI features at once across many users. Without a clear budget and visibility, costs are hard to predict and it is impossible to see who is driving spend. Fuxam solves this with three things: - A **monthly budget** that every AI request is checked against, so usage cannot run away unbounded. - A **usage bar** that shows how much of the current month's budget has been consumed and exactly when it resets. - A **per-user AI activity** log so administrators can see which people, features, and models are spending credits. ## The monthly billing period The AI budget is enforced over a **calendar month**, measured in UTC: - The current period starts at **00:00 UTC on the 1st of the month** (`periodStart`). - It **resets at 00:00 UTC on the 1st of the next month** (`resetsAt`). - Only credits spent **since the start of the current period** count toward the budget. Usage from previous months does not carry over and does not count against the current month. Because the window is measured in UTC, the reset moment can appear on the previous evening or the next morning in your local time zone. The usage view always shows the reset date and a relative countdown (for example, "resets in 12 days") based on the user's language and locale. :::note The budget resets automatically — there is no manual reset. When a new month begins, the usage bar returns to zero and any request that was previously blocked for exceeding the budget is allowed again. ::: ## Read the token usage bar Open **Settings → Storage & AI**. The **Fuxam tokens** section shows a labelled monthly-budget bar with: - The **amount used** versus the total monthly budget. - The **percentage used** of the budget. - A **reset line** — the date the budget resets and a relative countdown (for example, "resets on 1 Aug 2026 · in 12 days"). When usage passes the budget (over 100%), the percentage is shown in a warning/destructive colour to make the over-budget state obvious. While over budget, new AI requests are refused until the next monthly reset. :::tip Check the usage bar early in a heavy AI period (for example, when many teachers are generating Workbench content). The reset countdown tells you exactly how long until the budget refreshes if you are close to the limit. ::: ## AI activity: who used what The **AI activity** section (also under **Settings → Storage & AI**) lists individual AI usage entries so administrators can attribute usage to people. Each line shows: - The **user's name** who made the request. - The **feature** used (for example, Docu-chat, Workbench, chat, or Concierge). - The **model** used and the **credits spent**. - A **timestamp** for when the usage occurred. Use the **time-frame selector** in the section header to switch the range between the **last 7 days**, **last 30 days**, and **last 365 days**. Entries are grouped so that repeated calls by the same user with the same model, feature, and minute are combined into a single line with a total credit count, keeping the list readable during bursts of activity. :::info AI activity is sourced from the institution's audit log. Each time credits are charged, a record is written that captures the user, feature, model, and credits spent — which is what powers the per-user attribution shown here. These `ai_usage` entries also appear in the institution **Audit logs** with a readable description of the feature, model, and credits. ::: ### What a line looks like A typical activity line reads as the user's name followed by the request feature, credits spent, model, and timestamp — for example: > **Jane Doe** · Docu-chat · 42 credits · Claude Sonnet 4.6 · 20 Jul, 14:03 ## Model availability in AI tools To protect the institution's budget, **premium (high-cost) models are unavailable** in the AI tools where quality gains over cheaper models are marginal but cost is high — specifically **Docu-chat** and **Workbench copilot / element generation**. High-cost models are: - **Hidden** in the model selectors for Docu-chat and the Workbench copilot, so they cannot be chosen. - **Blocked on the server** for the Docu-chat and Workbench routes: even a direct request that specifies a premium model is refused (returns an error instead of running). This is intentional. Premium models are excluded from these features because their output-quality gains are marginal there, while their cost makes it easy to exhaust an institution's monthly AI budget quickly. :::note Model availability is filtered by the model's **identifier**, not its display name, so renaming a model in the interface does not change whether it is allowed. ::: ## What happens when the budget runs out When the institution has spent its entire monthly budget, AI requests are refused until the budget resets on the 1st of the next month (UTC). Users triggering an AI feature will be stopped by the budget check rather than partially completing a request. The usage bar shows the over-budget state, and the reset countdown indicates when AI features will be available again. ## Best practices - **Watch the reset countdown, not just the percentage.** If you are near the limit, the countdown tells you whether the budget will refresh before your next big AI task. - **Use AI activity to coach heavy users.** If a small number of people account for most spend, the per-user view makes that visible so you can guide usage. - **Prefer the standard models for Docu-chat and Workbench.** Premium models are deliberately unavailable there; the standard models are more than adequate for those features. - **Plan bulk AI work around the monthly reset** if you expect to approach the budget. ## Common pitfalls | Pitfall | What happens | Avoid by | |---|---|---| | Expecting last month's usage to count | Usage only counts from the 1st of the current month (UTC) | Read the usage bar as a **monthly** figure, not lifetime | | Reset appears "a day off" | The window is measured in UTC, so the reset can fall on a different local date | Trust the displayed reset date and relative countdown | | Looking for a premium model in Docu-chat / Workbench | High-cost models are hidden and blocked in those tools | Use an available standard model | | Assuming AI activity shows raw provider tokens | Activity shows Fuxam **credits** spent, grouped per user/model/feature/minute | Read credits as the internal metering unit | ## FAQ **What is a Fuxam token or credit?** It is the internal unit Fuxam uses to meter AI usage across features. One AI request usually spends several credits depending on the model and the size of the request and response. Credits are counted per institution and checked against your monthly budget. **When does the budget reset?** At 00:00 UTC on the 1st of each month. Only credits spent since the start of the current month count toward the budget. The usage bar shows the exact reset date and a relative countdown. **Why can't I pick a premium model in Docu-chat or Workbench?** Premium (high-cost) models are intentionally excluded from Docu-chat and Workbench copilot because their quality advantage there is marginal while they can exhaust the AI budget quickly. They are hidden in the selector and blocked on the server for those features. **Can I see who used the AI features?** Yes. The AI activity section under Settings → Storage & AI lists usage per user, showing the person, feature, model, credits spent, and timestamp. Use the time-frame selector for the last 7, 30, or 365 days. **What happens when we run out of budget?** New AI requests are refused until the budget resets on the 1st of the next month (UTC). The usage bar shows the over-budget state and the countdown to the reset. ## Related pages **[Settings overview](/docs/de/fuxam-web/settings)** Map of the main Settings areas administrators manage. **[Create and edit a Workbench](/docs/de/fuxam-web/lms/content/workbench)** The Workbench copilot uses AI and follows the model availability rules described here.