AI features consume **tokens** — the unit of work for language models. Doxbrix meters that consumption transparently so you always know how much AI you're using...
AI features consume **tokens** — the unit of work for language models. Doxbrix meters that consumption transparently so you always know how much AI you're using...
AI features consume tokens — the unit of work for language models. Doxbrix meters that consumption transparently so you always know how much AI you're using and whether you're on track against your budget. Paired with the Workspace AI policy, metering keeps AI costs predictable as your team scales.
What counts as usage
Every AI feature draws from the same token pool:
| Feature | Typical cost |
|---|---|
| [Writing Copilot](/ai/copilot) | Low — short, frequent completions |
| [Improve with AI](/ai/improve-with-ai) | Low–medium per page |
| [Scan with AI](/ai/scan-and-content-quality) | Low — runs continuously as you write |
| [Workspace assistant](/ai/workspace-assistant) | Medium per conversation |
| [AI agent](/ai/agent) | High — multi-step, multi-page tasks |
| [Site assistant](/reader/site-assistant) | Scales with reader question volume |
The Spark and Atlas tiers consume tokens at different rates — Atlas (deeper reasoning) costs more per request than Spark (fast).
The Usage page
Go to Usage (/usage) to see your workspace's consumption:
- Token wallet — your current balance of available tokens.
- Monthly grant — the tokens your plan includes each period, which top up the wallet.
- Active token packs — any additional token packs you've added on top of the monthly grant.
As features run, they draw down the wallet; the monthly grant refreshes it each period.
Budgets and alerts
Set spending guardrails in the Workspace AI policy Token Budgets section so usage never surprises you:
- Monthly workspace budget (tokens) — a ceiling for the period (or No limit).
- Alert threshold (%) — get notified when usage crosses this share of the budget (e.g. 80%).
- Alert recipients — the emails that receive budget alerts.
Reducing usage
If you're consuming more than expected:
Use the Copilot tier override and Scans tier override to keep everyday work on Spark; reserve Atlas for complex agent jobs.
Narrow agent task scope and schedules so runs do only necessary work.
Turn off any AI surface you're not using in Feature Controls.
A budget with an 80% alert gives you time to react before a hard stop.
Plans and token grants
Each plan includes a monthly token grant. When the wallet runs low you can add token packs for more headroom, or raise your plan. Your current wallet, grant, and packs are shown on both the Usage and Billing pages.