dion236Q: Gmail and token usage
Hi, a few questions before I buy:
1. Gmail: Your help page still says Gmail is send-only and the inbox isn't read. Has Google's security review (CASA) been submitted, and what's your realistic ETA for Gmail inbox reading and Gmail in "Needs you"?
2. Token usage with BYOK: Roughly how many input/output tokens does a typical Jarvis chat, a Draft reply and a routine run use? How much of that is fixed overhead (system prompt, tool definitions) versus retrieved brain passages?
3. Prompt caching: Do you use Anthropic/OpenAI prompt caching on BYOK runs, so the fixed overhead isn't billed at full price every time?
4. Background costs: Do "Checks first" routine checks, Needs you ranking and brain syncing make model calls on my own key, even when no run happens?
If you launch a newmajor version or plan structure, do we keep receiving it?
ahmadhajj
Oct 22, 2026A: Hi Dion, thanks for the detailed questions.
1. Gmail: The help page is right. Gmail is send-only today. We submitted the CASA review, it will be around 3-4 weeks until the whole process is completed. Outlook works fully today, including inbox reading and Outlook in Needs you.
2. Token usage: These are per-run medians from real usage over the last two weeks, counting every model turn in the run:
- Jarvis chat: about 50K input and 1.1K output tokens. About three quarters of the input is read from cache.
Most of the input is the fixed part: instructions, tool definitions and earlier turns, re-read on each step. Brain passages are a smaller part that depends on the question.
3. Prompt caching: Yes. BYOK runs use Anthropic prompt caching and OpenAI's automatic caching, so the fixed part is billed at the cached rate after the first turn.
4. Background costs: No. Needs you ranking, routine checks, the brain's syncing and extraction, and search indexing all run on our keys, never yours. Your key is only used for runs you start: chats, Draft reply, and routine runs that actually go ahead. A routine that is skipped because nothing changed doesn't touch your key.
5. Future versions: Yes, your license includes future updates to your plan.
Thanks Ahmad, very clear and helpful
A few follow-ups: OpenRouter: is there a firm date?
White-label pages (tier 4): what exactly is branded: only the logo, or also colours and domain? Is every "Agently" mention removed (footer, badge)? Can each client workspace have its own logo, and do PDF/PPT exports keep the branding?
Full white-label: are you planning white-label login (own domain)