Xumulus Logo
Agent-ready search & discovery on Next.js + Elastic/OpenSearch

Algolia alternative: what changes when agents do the searching

Most teams shopping for an Algolia alternative are looking at one of three things: a bill that grew faster than the catalog, a merchandising limit they keep hitting, or an attribute their product data has and the index has nowhere to put. Here is the honest version of each.

Elastic
Relevance + speed
Next.js
Modern UX + performance
RAG
Agentic discovery ready
AI Discovery Signals
Retrieval-first architecture makes agentic discovery possible.
Scroll

First, Algolia’s model stated fairly

Algolia meters two things at once: search requests and records. On its published pricing (checked August 2026), the Grow plan includes 10K search requests and 100K records per month, then bills $0.50 per additional 1K requests and $0.40 per additional 1K records. Grow Plus keeps records at $0.40 but takes requests to $1.75 per 1K. Above that, Elevate is an annual contract with custom volumes.

For a storefront where a person types a query, scans one page of results and buys, that is a reasonable deal. The engineering is good, the relevance is good, and the hosted model means nobody on your team is on call for a cluster. If that describes your catalog and your traffic, this page is not for you and you should stay where you are.

The model binds in specific, predictable places. All three below come from how the pricing is constructed rather than from anything being wrong with the product.

Discovery bottlenecks

Where the meters start to bite

These are not proof-of-concept problems. They arrive in year two, with a real catalog, a merchandising team, and now agents.

01
Every sort order can be a full copy of your catalog
A standard replica is a copy of the primary index, so its records bill as records. Algolia’s own documentation gives the arithmetic: a 1,000-record index with two standard replicas is 3,000 records. Price ascending, price descending, newest, best-selling — a normal commerce sort menu, priced as four catalogs. Virtual replicas fix this and do not add records, but you have to know to use them.
02
Agentic retrieval multiplies a per-request bill
A shopper issues one query. An agent researching the same purchase issues many, because that is how retrieval-augmented reasoning works. When search requests are the meter, the workload everyone is planning for in 2026 is the workload that multiplies the invoice.
03
Merchandising rules have a plan cliff
Grow allows 10 rules per index. Grow Plus allows 10,000. That is a reasonable way to build a price ladder, but it means a merchandising decision becomes a procurement decision at an arbitrary threshold.
04
Some attributes have nowhere to live
Vehicle fitment, compatibility matrices, kit and bundle relationships, regional compliance. No commerce platform models these natively, so they end up in a side database and a nightly job whose failures surface as wrong results rather than as alerts.
05
Ranking logic sits outside your repository
Rules, synonyms and boosts live in a hosted dashboard. They are not in version control, not in code review, and not in your deploy — so the answer to “why did results change on Tuesday” is an audit log rather than a diff.
06
Both meters grow with success
Records rise as the catalog grows. Requests rise as traffic grows. The two things you are working hardest to increase are the two inputs to the bill.
Key takeaway
If none of these apply, a hosted plugin is genuinely the right call. Most catalogs past a few hundred thousand SKUs hit at least three.
Contrast

Two architectures, side by side

Both ship working search. They differ in what you are billed for, what you can model, and where a failure lands.

01
Comparison
Algolia (hosted, metered)
Billed per search request and per record
Index-first (Next.js + Elastic/OpenSearch)
Billed as infrastructure you run
02
Comparison
Algolia (hosted, metered)
Standard replicas duplicate records; virtual replicas avoid it
Index-first (Next.js + Elastic/OpenSearch)
Sort orders are query-time, not extra copies
03
Comparison
Algolia (hosted, metered)
10 rules per index on Grow, 10,000 on Grow Plus
Index-first (Next.js + Elastic/OpenSearch)
Ranking logic is code — no rule ceiling
04
Comparison
Algolia (hosted, metered)
Custom attributes need a side process to reach the index
Index-first (Next.js + Elastic/OpenSearch)
Fitment and compatibility are fields you query directly
05
Comparison
Algolia (hosted, metered)
Relevance configured in a hosted dashboard
Index-first (Next.js + Elastic/OpenSearch)
Relevance in your repo, in review, in your deploy
06
Comparison
Algolia (hosted, metered)
Agent traffic increases the metered bill
Index-first (Next.js + Elastic/OpenSearch)
Agent traffic increases CPU you already pay for
Bottom line
The trade is real in both directions: you take on running a cluster, and you stop paying per question asked.

The attribute problem, concretely

Fitment is the clearest case, because the shopper’s first question is not about the product at all. They do not want to browse tires. They want the tires that fit a 2019 F-150 with a specific trim and wheel size — and a result that ignores that is not a worse result, it is a wrong one.

Here is that working: best off-road tires for the Ford F-150 on OffroadRig HQ. The picks, the comparison table and the buyer’s guide are all conditioned on which vehicle the shopper owns, which means the fitment relationship has to be queryable — not a filter applied after the fact to a generic product feed.

That is the practical difference. When the index can hold the relationship, a page like that is a query. When it cannot, the relationship lives in a second system and every page built on it inherits that system’s freshness and failure modes.

The stack

So should you switch?

An honest decision tree, including the case for staying put.

01
Stay

Keep Algolia if this is you

There is no prize for migrating away from something that fits.

  • Catalog comfortably inside your plan’s record count
  • Standard product attributes, no fitment or compatibility
  • Human traffic, predictable query volume
  • No team to run and monitor a search cluster
02
Switch

Consider index-first if this is you

The economics and the data model start pointing the same direction.

  • Replica or record growth is driving the bill
  • Attributes the platform cannot model natively
  • Agent and API retrieval rising as a share of traffic
  • Merchandising logic you want in version control
03
How

What a migration actually involves

Usually less dramatic than it sounds, because it runs alongside.

  • Model the index against your real catalog, not a sample
  • Run both in parallel and compare relevance on live queries
  • Move traffic gradually, keeping the old path as fallback
  • Port merchandising rules into code as you go

Want the numbers for your catalog?

Send us your record count, your sort orders and your current bill and we will tell you what index-first would cost — including when the answer is that you should stay on Algolia.

Search is the foundation of AI commerce