Understand Vox usage and limits
Vox consumes credits from the active workspace. A turn is priced from the configured platform-message fee, uncached input, cached input, output, completed web searches, and successfully analyzed attachments.
Vox credits = platform message fee
+ uncached input tokens / 1,000 × rate
+ cached input tokens / 1,000 × rate
+ output tokens / 1,000 × rate
+ completed web searches × rate
+ analyzed attachments × rate
Wirevox reserves an amount appropriate to the predicted request mode before work begins, settles the provider-reported usage afterward, and releases the unused reservation. The credits shown below a Vox response are the settled credits for that turn. Use the live Billing page for current rates.
Current product limits
| Boundary | Current value |
|---|---|
| User-message length | 12,000 characters |
| Messages in one thread | 200 total |
| Threads returned in history | Up to 100 used threads |
| Files per message | 2 |
| File size | 10 MB each |
| Attachment retention | 7 days |
| Pending action expiry | 10 minutes |
| Largest recent-message context | 20 messages |
| Largest context-character budget | 24,000 characters |
| Website pages read per call | 5 |
| Candidate links from a website scan | Up to 100 |
| Knowledge-base processed size | 50 MB |
Tool calls also have mode-specific budgets. Current action mode allows up to 12 tool calls across six rounds; research allows up to 10 calls across six rounds. These are internal per-turn ceilings, not monthly quotas.
Why short messages still cost credits
The visible answer is only part of a turn. The model may receive recent context, stable instructions, capability schemas, and live tool results. Therefore “Hi” can cost less than a workspace question even when both answers are short. Vox routing intentionally gives casual turns no tools and smaller context.
For a full explanation of balances, rates, reservations, and settlement, see Understand Website Chat, Vox, and tool usage.