AI limits and costs
When Geeky Bot uses Zywrap or OpenAI, each call costs money. A site-wide daily budget and monthly cap put a hard ceiling on those calls, and per-visitor rate limits stop one visitor from flooding the chat.
Set the AI call limits
Section titled “Set the AI call limits”- Open Geeky Bot → AI & Privacy.
- Under AI connection, set Daily AI call budget and Monthly AI call cap.
- Select Save settings.
| Setting | Default | Range | What it does |
|---|---|---|---|
| Daily AI call budget | 1,000 | 1 to 1,000,000 | Site-wide limit on paid AI calls per day, staff included. |
| Monthly AI call cap | 20,000 | 1 to 30,000,000 | Hard stop for the month. |
The limits count every call to Zywrap or OpenAI: chat answers, Smart Catalog, Rescue and learning from missed searches. Days and months follow your site’s time zone. There is no “unlimited” setting.
The AI calls left today figure at the top of the page, and the AI spend ceiling note below the fields, show how much of each limit is used.
When a limit is reached
Section titled “When a limit is reached”- Chat answers aren’t sent to AI. Shoppers get the local reply, built from your catalog and Store Knowledge pages, so the chat keeps working.
- The AI spend ceiling note says how many calls were declined today.
- Smart Catalog stops early, to keep part of the limit free for shopper chat. It pauses while less than a quarter of the daily limit (at least 20 calls) or a tenth of the monthly limit (at least 100 calls) is left. Product Search then shows Paused: AI limit, and it continues by itself when calls are free again.
Calls that don’t count
Section titled “Calls that don’t count”- Cached chat answers. An AI reply is reused for 30 minutes when the same question arrives with the same product data. A reused reply isn’t sent again and isn’t counted.
- Rescue answers. Each is stored for 30 days, so the same failed search isn’t sent again.
- Smart Catalog searches. The words are written once per product and stored on your server. Searching them makes no AI call.
- Local replies. Policy answers, product-detail answers, recommendations and the other local replies never call AI.
Shopper rate limits
Section titled “Shopper rate limits”Rate limits apply per visitor, in every answer mode. Set the first two under History, retention and rate limits on AI & Privacy.
| Limit | Default | What it does |
|---|---|---|
| Rate limit shopper messages | 120 | Chat messages one visitor can send per window, from 20 to 1,000. |
| Rate limit window minutes | 5 | The length of the window, from 1 to 60 minutes. |
| Product searches | 240 per 5 minutes | Fixed. Product catalog requests from one visitor. |
| Rescue | 20 per hour | Fixed. New failed searches one visitor can send to AI. Over it, the search simply isn’t rescued. |
A visitor over the chat limit sees “Too many assistant messages. Please wait a few minutes and try again.”
A visitor is told apart by IP address and browser. If a CDN such as Cloudflare sits in front of your store, set Trusted proxies in front of this site to 1, or every shopper may share one limit. Leave it at 0 otherwise: a higher number lets visitors fake their address.
Logged-in administrators, shop managers and users who can edit products or orders are never rate-limited, so you can test freely. Their AI calls still count against the daily and monthly limits.
Rough costs
Section titled “Rough costs”- Local grounded mode: nothing. No AI service is called.
- OpenAI with
gpt-4o-mini: a chat reply typically costs well under one US cent, billed by OpenAI. - Zywrap: a chat answer uses about 100 credits, roughly $0.001. New accounts get 10,000 free credits.
As an example, with Zywrap and the default daily budget of 1,000 calls, chat answers can’t cost more than about $1 a day. Smart Catalog writes words in batches of products, so a large catalog uses calls when you first switch it on and again when products change. See Smart Catalog.