Limits & quotas
A reference of the operational limits that apply to your workspace.
Monthly token budget
| Tier | Budget | Behaviour at cap |
|---|---|---|
| Starter | 20,000,000 tokens | Telegram alert at 80%. Default overage policy: alert (notify, do not block). |
| Growth | 40,000,000 tokens | Telegram alert at 80%. Default overage policy: alert (notify, do not block). |
| Enterprise | Unlimited | No monthly cap. |
See token meter.
Per-tool limits
| Tool | Limit |
|---|---|
execute_code | 15-second timeout per execution. |
| Tool output payload | Capped at ~50,000 characters per call. |
page_fetch | Returned page text capped at ~20,000 characters. |
| Telegram message | Telegram’s 4,096-character limit applies. Longer outputs are split or delivered as a PDF. |
| Voice note | Transcribed by local whisper.cpp where installed, with OpenAI Whisper API fallback. |
| Vector memory | Bounded by your VPS disk allocation. |
Concurrency
Han AI handles one conversation turn at a time per workspace. Scheduled jobs run on cron and do not block your live conversation.
Storage
Operational state on your VPS is bounded only by your VPS disk allocation. Backups follow the DPA retention defaults — 30 days rolling.