How can I add autocomplete before Algolia without driving up search costs?

I would start with a small suggestion layer rather than an AI service. Offer category, brand, or other relevant filters as people type, then let them choose a suggestion before sending a search to Algolia. Keep the suggestions local or serve them from a separate lightweight index so every keystroke does not become another paid search.

You can also wait for a short pause in typing before requesting suggestions. Measure both the number of Algolia requests and whether people actually select the suggestions. A vague search is still useful to some users, so I would not block it outright.

Check the request pattern before adding another component. If the interface searches on every keystroke, debouncing those requests may reduce volume without changing the backend at all. You could also cache repeated queries and their results.

Caching is less helpful if most searches are unique, and it will not turn a broad search into a specific one. Suggestions and filters address a different problem from request volume, so track those separately.

Typesense is worth evaluating if cost is the main reason for reconsidering the stack. It supports autocomplete, typo tolerance, and filtering, which may cover a lot of the search interface without a separate AI layer. Compare the hosting and maintenance costs against your current usage rather than assuming the migration will be cheaper.

That said, swapping engines will not make a search for jacket less vague. You would still need useful filters or suggestions if the goal is to help people narrow it down.

There is also the self-hosted route with something like Meilisearch or Elasticsearch. That gives you another way to control search costs, but it trades an easy setup for infrastructure and operational work. I would compare that work with the cost of keeping Algolia before rebuilding the index and search integration.