The vocabulary of search and discovery in plain words — what the terms mean and why they matter when you run a catalogue.
Combining keyword search with vector search so a query matches both exact words and meaning.
Numeric representations of text (or images) that let search match by meaning.
Search that returns results by meaning rather than exact keywords.
Returning correct results even when the query contains spelling mistakes.
Reducing words to a root form so «running» matches «run».
Autocomplete hints derived from what users actually search for.
Filters shoppers narrow by — price, brand, size — shown as live counts next to the results they'd produce.
A managed mapping between words a shopper might type and the words your catalogue actually uses.
Common, low-signal words excluded from relevance scoring so they don't drown out the terms that actually distinguish a query.
A stable name your app searches against, pointed at the real collection behind the scenes — so a full reindex can happen with zero downtime.
A named, reusable bundle of search parameters — which fields to query, which filters to apply, how to sort — so a frontend calls the preset instead of repeating the same options on every request.
An API key restricted to exactly the operations, collections, fields and origins it needs — safe to embed in a browser because it cannot do anything else.
A client-supplied header that lets a retried write be repeated safely, instead of applied twice.
Normalized discounted cumulative gain — the standard offline metric for ranking quality.
A curated set of queries with known-good results, used to score ranking changes offline.
Splitting live traffic between ranking variants and measuring which performs better.
The share of searches that return nothing — the single most actionable search metric.
A signed HTTP callback AACSearch sends to your own endpoint when something happens — a quality regression, a finished import, a paused organization — instead of you polling for it.
Three independent ceilings that can each answer 429 on their own: the organization's monthly request quota, a key's rate limit, and a search key's per-session AI allowance.
Editorial control over search results: pins, boosts, exclusions, banners and campaigns.
Paid placements inside search results, tracked separately from organic ranking.
A rule that matches a specific query pattern and takes an action — pin a result, hide it, or redirect the query — without touching the general ranking formula.
Items suggested from behaviour — bought together, viewed together, trending — rather than typed as a query.
Adjusting ranking for the visitor in front of you, from their own session and history, rather than one ranking for everyone.
Grounding an LLM's answers in retrieved documents, with citations.
An open protocol that lets AI agents call tools — AACSearch exposes your index through it.
Turning a conversational query — «red running shoes under $50» — into the structured filters a search index actually needs, automatically.
A structured network of entities and the relationships between them, extracted from your content, that grounds AI answers in more than isolated passages.
Speaking a query instead of typing it — the audio is transcribed, then searched exactly like typed text.
Search by uploading a photo instead of typing words — results are ranked by visual and semantic similarity to the image.
The best way to evaluate a search platform isn't reading one more comparison table.
Only then make the decision.