Algolia counts staging and locale indices against the same 10,000-record ceiling

The record ceiling is shared, and that is the part that catches teams. Staging, production and locale-split indices such as `products_en` and `products_de` all draw on the same quota, according to the post ^[8]. Work its own example. A 2,000-product Shopify store with three variants each is 6,000 records before any metadata fields ^[7]. That leaves 4,000, or 40 percent of the allowance ^[1]. Add a German index for the same catalogue and the account sits at 12,000, which is 2,000 over the line ^[2].

The second multiplier is the indexing function. Splitting rich text into per-block records for highlight accuracy raises the record count five to ten times, the post says ^[9]. At five times, the free plan holds 2,000 source documents. At ten times it holds 1,000 ^[3]. A single Sanity document with image URLs and category arrays can expand into three to five records depending on how the split is written ^[10].

The other meter counts API calls that query the index ^[1]. Ten thousand a month is roughly 333 a day, and the post reckons a moderately trafficked search widget on a content site clears that inside the first week after launch ^[6]. What the post does not say is what Algolia does at request 10,001 on a free account ^[19]. That behaviour is worth establishing before the free plan sits behind anything user-facing.

The dollar figures are softer than the ceilings. Algolia quote-gates exact numbers, so the Grow floor here comes from public documentation and indie benchmarks, and the Premium and Elevate minimums come from market reports and quotes the author has seen ^[11][14]. At the low end that Grow floor is $600 a year and at the high end $720, paid whether traffic arrives or not ^[4]. The step from Grow's floor to the bottom of the Premium estimate is roughly eight times, and to the top of it forty times ^[5]. Grow also buys no uptime commitment, since the 99.99 percent SLA starts at Premium ^[12].

The upgrade test in the post hangs on one feature. "If you're not using NeuralSearch, you're paying Algolia prices for what Typesense or Meilisearch could do at a fraction of the cost," the author wrote ^[15]. NeuralSearch, Personalization and Predict sit in Elevate, the top tier ^[13]. Meilisearch and Typesense both have genuine open-source self-hosted options. That matters when the search process runs in a Railway or Render container and the bill is compute only ^[18].

For a Next.js and Sanity build, the post would budget Grow from day one on any site expected to serve more than a few hundred daily active users, and keeps the free tier for prototyping and internal tools ^[17].