INTENT LAYER EST. 2025

Zero pipeline. Zero RAG. Zero training.

Intent infrastructure that gets what your visitors want â€” and what their AI agents are asking for. Seven lines of code.

Read the Docs Get Early Access

CarsÂ· Real EstateÂ· StorefrontsÂ· Marketplaces

THE RECORD

30Y / 4 FAILURES / 1 FIX

1995

KEYWORD
COERCION

Binary matching. "Cushioned boots" returned zero results for "comfortable shoes." We forced humans to speak like database columns.

01 / 04

2010

FILTER
INTERROGATION

Checkboxes and dropdowns. Search became a form. Users were interrogated by database structures instead of understood by them.

02 / 04

2023

SEMANTIC
FRACTURE

LLMs understood language but couldn't connect to databases in real time. The industry got stuck on unstable unstructured pipelines.

03 / 04

2025

THE
akropolys
LAYER

Intent resolved to structured query in <20ms. No synonyms. No sidebars. A translation layer between human thought and catalog databases.

04 / 04

LATENCY & PERFORMANCE

From human thought
to database query.

Under 20 milliseconds. Zero configuration.

STEP 01

Capture raw intent

Instead of fragile keywords, users query in natural human language. The SDK captures this raw text stream instantly on the client side.

STEP 02

Resolve and structure

The Akropolys Engine parses the query into a deterministic JSON object containing categories, price bounds, attributes, and negative filters in <20ms.

STEP 03

Direct catalog routing

The structured intent is matched semantically against your database, completely bypassing the need for complex synonyms, search-index tuning, or manual routing.

INTENT INGESTION

USER INPUT:

"family SUV under Â£28k, not diesel"|

RESOLVED INTENT STRUCT:

"intent": "search",
"category": "SUV",
"filters": {
  "price_max": 28000,
  "currency": "GBP"
},
"negations": ["diesel"]

Intent Resolution Latency

Traditional LLM Agent 2,450ms

Vector DB + Router 380ms

Akropolys Engine 18ms

THE SDK

Five lines.
Full semantic
intelligence.

â†’

Auto-ingest on browse.

usePageIngest() fires once per product view. Your catalog enters vector space as users navigate â€” no batch jobs, no uploads.

â†’

Queries resolve to meaning.

Natural language is parsed into structured intent: category, constraints, attributes, price. Matched semantically. No synonyms file needed.

â†’

Every search becomes data.

Resolved or not, every query is logged as structured intent. Your dashboard shows what people want â€” including what you don't stock.

@akropolys/sdk â€” product-page.jsx Full docs â†’

import { usePageIngest, SearchBar } from '@akropolys/sdk'

export default function ProductPage({ product }) {
  // Fires automatically on client-side mount
  usePageIngest({
    id:     product.id,
    title:  product.name,
    url:    window.location.href,
    fields: {
      price:    product.price,
      category: product.category,
    },
  })

  return (
    <SearchBar
      placeholder="family SUV under Â£28k, not diesel"
      onResult={(results) => setItems(results)}
    />
  )
}

Zero pipeline. Zero RAG. Zero training.

From human thoughtto database query.

Capture raw intent

Resolve and structure

Direct catalog routing

Five lines.Full semanticintelligence.

Your catalog. Finally understood.

From human thought
to database query.

Five lines.
Full semantic
intelligence.