AI budget & cost caps

The AI Budget page shows how much AI your store is consuming this month: total tokens, estimated cost, and request volume, with a breakdown by capability and model. Find it under Settings → Data → AI Budget, or go to /app/<storeId>/settings/ai-budget. The header reads AI budgetwith the kicker “AI cost cap.”

This view is read-only

AI Budget reports spend — it does not let you edit a cap here. To change what your store can spend, manage your plan from Billing(the page links straight to it with a “Manage plan” action).

The three headline metrics

Across the top, three cards summarize the current calendar month:

Tokens this month— the total input + output tokens consumed across every AI feature for this store.
Estimated cost— the dollar value of those tokens, shown to the cent (e.g. $12.40). It is an estimate derived from per-model pricing, not an invoice.
AI requests— how many calls were made to the models this month.

Spend by capability and model

Below the cards, a table breaks the same numbers down per capability and model, sorted by cost (most expensive first). Columns are:

Capability— the kind of work the model did, such as chat, embeddings, or recovery.
Model— the concrete model id that handled the requests (e.g. an OpenAI, Anthropic, or Gemini model). Shows unknownif the model wasn’t recorded.
Requests— call count for that capability/model pair.
Tokens— tokens consumed by that pair.
Cost— estimated dollar cost for that pair.

This breakdown tells you where your AI spend actually goes. If one capability dominates, that’s the lever to pull — for example, a high chat cost might justify pinning a cheaper model on the AI Setup page.

Empty until shoppers interact

Before any AI usage is tracked, the table shows “No AI usage tracked yet.” Once shoppers start chatting with the widget — and catalog embeddings and recovery emails run — the rows populate. New stores should expect this page to be blank on day one.

How costs are calculated

Zubby records token usage per AI call and multiplies by the provider’s per-token price for that model to produce the estimated cost. Numbers are aggregated daily and summed for the month, so the figures here are an internal estimate to help you budget — your actual bill is governed by your plan. See Billing for what you are charged.

Keeping costs down

Prefer a smaller model for chat if answer quality is acceptable for your catalog — the mini/haiku tiers cost far less per message.
Turn off AI conversion flows you don’t use (guided selling, smart discounts, browse abandonment) on the AI Setuppage so they don’t generate extra calls.
Keep your catalog and knowledge base tidy so retrieval is efficient and the agent doesn’t burn tokens on noisy context.

Choose the model that drives most of this spend under AI provider & model. Manage your plan and see invoices on Billing, and compare plan tiers on the pricing page.

AI budget & cost caps

This view is read-only

AI Budget reports spend — it does not let you edit a cap here. To change what your store can spend, manage your plan from Billing(the page links straight to it with a “Manage plan” action).

The three headline metrics

Across the top, three cards summarize the current calendar month:

Tokens this month— the total input + output tokens consumed across every AI feature for this store.

Estimated cost— the dollar value of those tokens, shown to the cent (e.g. $12.40). It is an estimate derived from per-model pricing, not an invoice.

AI requests— how many calls were made to the models this month.

Spend by capability and model

Below the cards, a table breaks the same numbers down per capability and model, sorted by cost (most expensive first). Columns are:

Capability— the kind of work the model did, such as chat, embeddings, or recovery.

Model— the concrete model id that handled the requests (e.g. an OpenAI, Anthropic, or Gemini model). Shows unknownif the model wasn’t recorded.

Requests— call count for that capability/model pair.

Tokens— tokens consumed by that pair.

Cost— estimated dollar cost for that pair.

Empty until shoppers interact

How costs are calculated

Keeping costs down

Prefer a smaller model for chat if answer quality is acceptable for your catalog — the mini/haiku tiers cost far less per message.

Turn off AI conversion flows you don’t use (guided selling, smart discounts, browse abandonment) on the AI Setuppage so they don’t generate extra calls.

Keep your catalog and knowledge base tidy so retrieval is efficient and the agent doesn’t burn tokens on noisy context.

AI budget & cost caps

The three headline metrics

Spend by capability and model

How costs are calculated

Keeping costs down

Related

AI budget & cost caps

The three headline metrics

Spend by capability and model

How costs are calculated

Keeping costs down

Related