Token meter & budgets
Han AI runs on foundation models that bill by tokens — the small fragments of text a model reads and writes. Your workspace has a monthly token budget tied to your tier. The token meter records every model call against that budget and tells you when you are approaching it. The point of the meter is a predictable monthly cost.
Budgets by tier
| Tier | Monthly token budget |
|---|---|
| Starter | 20,000,000 tokens |
| Growth | 40,000,000 tokens |
| Enterprise | Unlimited |
How it works
- Every model call goes through
meter.preCall(budget check) andmeter.recordCall(ledger write). - The ledger lives on your VPS and is per-tenant — no other workspace shares it.
- The month resets on the first of the calendar month.
Alerts
At 80% of your monthly budget you receive a Telegram alert from your Han AI bot. The default overage policy is alert — you are notified, not blocked. If you choose to upgrade your tier, the new budget takes effect immediately.
What counts as a token
Every input and output token across:
- Reasoning calls (the model thinking before replying).
- Conversation in Telegram and on the dashboard.
- Document and image processing.
- Background jobs — dreaming loop, daily brief, scheduled tasks.
Local tools that do not call a foundation model (page fetches, OCR, code execution, vector memory writes) do not consume tokens.
Where to see current usage
Live usage is recorded in the on-VPS ledger and is exposed in the dashboard.
TODO: confirm whether the public dashboard /api/usage endpoint is live. The canonical platform note records that the Worker-to-Node bridge for /api/usage returns 500 until the bridge is built; usage is still recorded in the ledger and visible in the operator view in the meantime.