Pop-Search

What it is

A new kind of search: similarity without vectors

Pop-Search is a structured similarity & anomaly engine. You load your data as ordinary records, and it finds records that are similar to a given one, and records that are statistically out of place — using the fields you already have.

The whole idea in one line: vector-less search. A vector database needs you to train an embedding model, push every record through it, and accept a black box. Pop-Search skips all of that. Your numeric fields already carry meaning — price, rating, latency, temperature — so it works on them directly.

Every similarity result comes with a score you can read and a reason you can explain, which matters when the answer has to stand up in a pricing review, a fraud queue, or a maintenance ticket.

How it works

Load it. Query it. It runs on a shoe-string.

You load your data. Pop-Search indexes your numeric fields so that similarity, range filtering, and anomaly detection all run fast — even on a small 2-core / 4 GB server.

What you get

Everything the "find similar + find outliers" job needs

Why it's different

Not a vector DB. Not a text search engine.

The tools people reach for today were built for a different job.

Hardware

Built for the box you already have

The entire engine is one static binary with local file persistence. It runs comfortably on a small VPS — the kind of box that hosts a dozen little services — with no JVM, no cluster, no external database, and no GPU.

That's the whole footprint — and there's still room for the rest of the stack.

Quick start

Five calls to a working demo

Base URL is the host serving this page (the API serves its own docs, so the playground and the API are the same origin). With auth enabled, send your key in the X-Pop-Api-Key header.

1. Create a collection — declare your fields.

curl -s $BASE/collections -H 'Content-Type: application/json' -d '{
  "name": "products",
  "fields": {
    "price":    { "type": "numeric", "min": 0.99, "max": 10000, "bins": 20, "log": true },
    "rating":   { "type": "numeric", "min": 0, "max": 5, "bins": 16 },
    "category": { "type": "categorical", "levels": ["electronics","furniture","outdoor","kitchen"] }
  }
}'

2. Ingest records (JSON array, or POST /collections/products/ingest/csv for bulk).

curl -s $BASE/collections/products/records -H 'Content-Type: application/json' -d '[
  { "id": "e1", "price": 249.99, "rating": 4.6, "category": "electronics" },
  { "id": "e2", "price": 89.99,  "rating": 4.2, "category": "electronics" }
]'

3. Search — numeric range + categorical filter, ANDed.

curl -s $BASE/collections/products/search -H 'Content-Type: application/json' -d '{
  "must": [ { "field": "price", "gte": 50, "lte": 500 }, { "field": "category", "eq": "electronics" } ],
  "limit": 20, "with_values": true
}'

4. Find similar — nearest records to a given one.

curl -s $BASE/collections/products/similar -H 'Content-Type: application/json' -d '{
  "record": { "price": 249.99, "rating": 4.6, "category": "electronics" },
  "k": 10, "with_values": true
}'

5. Detect anomalies — statistical outliers on a field.

curl -s $BASE/collections/products/anomalies -H 'Content-Type: application/json' -d '{
  "field": "price", "z": 2.5, "limit": 50
}'

Live playground

Try it right here

No account, no setup. The button below loads a 45-product catalog into this server (collection demo) and you query it live. It's the exact API from the docs — the playground just calls it from your browser.

API reference

Every endpoint

JSON in, JSON out. All query endpoints take a JSON body. Errors return a JSON { "error": "..." } with an appropriate HTTP status.

Query conditions support gte / lte (numeric range), eq / neq / in (exact or set), label (field label), anomaly (z-threshold), and exists (optional fields). Combine with must (AND), any (OR), must_not (exclude).

Feedback

Report a bug or request a feature