See AI site search in action: hybrid search, RAG answers and search by image, on real indexes.
A six-minute narrated tour, two short screen recordings and 39 real screenshots, nothing staged. Live customer catalogues and news indexes searched the way people actually search: a question in plain words, a photo instead of words, a half-remembered product. Underneath it all, fully managed Apache Solr hosting that has been in production since 2011, with a free forever, with AI features from €20 a month and dedicated enterprise clusters when you need them.
Six minutes, narrated, no cuts that hide anything.
Both recordings are hosted Opensolr search pages, the same pages our customers put in front of their visitors. The voice says what is being searched and what comes back, as it happens on screen.
Search by image: a photo goes in, results come out
Photos picked from a folder, nothing typed. A shelf tag photographed in a store, a lily, a kitten, a dog, a bat, a snake, a daffodil, a supplement label, each dropped into search pages that have nothing to do with each other: a Romanian home-improvement retailer, a UK jewellery shop, an international news index.
- The shelf tag is read, and the exact interior door it was stuck to comes back first.
- A lily photo returns Lilium plants and flower seeds; a kitten returns kitten food and a kittens canvas print.
- The same dog photo returns leashes and collars on the DIY store, dog stories on the news index, and dog pendants at the jeweller.
- A bat returns bat earrings and a bat pendant; a snake, a snake necklace and ring; a daffodil, daffodil studs.
- A supplement label photographed in English returns the supplement recall articles on the news index.
It works on every Opensolr index, including the one our web crawler builds from your site, with no extra setup. Visitors see the camera icon next to the search box and that is the whole feature.
Hybrid search and AI answers: ask the way you would ask a person
Three indexes, three clumsy questions. Keyword search alone would have missed every one of them; Opensolr’s hybrid search reads the meaning as well as the words, and the AI layer answers from the documents it found, with the sources shown.
- News: “to which player is that guy in agreement of, to include in the Italian team?” finds the Tresoldi and Mancini story, and AI Hints summarises it.
- Home improvement: “wall art with red flowers in it 150” returns the 150 × 95 cm red-flowers canvas first; change the number to 90, the 90 × 60 cm one comes first.
- The AI reader summarises a product page: dimensions, material, price, availability.
- Jewellery: “something to wear around my neck, with a canine variety” returns terrier and labrador pendants.
The same search and the same answers are available as a REST API, in the free Drupal module and WordPress plugin, and in the LangChain, LlamaIndex, Haystack and Laravel packages.
One photo, three indexes, three different right answers.
The picture goes in, Opensolr turns it into a search, and the results come out of the index you already have. No new fields, no reindexing, no image database to build: a photo of a lily returns a lily pendant on a jewellery site, lily bulbs on a garden retailer, and a lily-flower recall on a news site.
A label or a shelf tag is read rather than recognised, so the product name and article number on it become the query and the exact item comes back first. That is the difference between “something like this” and “this one”.
There are three layers working together on every photo: what it looks like, what is printed on it, and any barcode or QR code on it. A barcode or QR code is decoded straight off the picture and handed back on its own, ready to open or search for.
Click any screenshot to see it full size. Search by image is on every hosted search page and in the REST API, on every plan.
Keyword precision and semantic recall in one query, then an answer with its sources.
Hybrid search runs the keyword and the meaning side by side and fuses them, so exact product codes still match exactly while a question phrased in other words, or another language, still lands on the right document. The RAG API then answers from those documents and cites them, on your data, with no OpenAI key and no vector database to run.
The first three screenshots are from the live sandbox, where you can run the same three-way comparison on your own website in about a minute.
Read more: how hybrid search works · the AI & RAG endpoints · LangChain, LlamaIndex, Haystack and MCP for AI agents · Drupal and WordPress modules.
A video site, crawled like any other site. Duration, channel and category come out as filters.
Everything on these three screens started with one action: the site’s address pasted into the web crawler. While it crawls, Opensolr reads the video markup the web already agrees on, schema.org VideoObject in JSON-LD, the same properties as microdata, Open Graph’s og:video tags and the Twitter player card, in that order of trust, and turns what it finds into typed fields: duration in seconds, upload date, channel, category, tags, views, chapters, the transcript when the site publishes captions. A site that has none of that markup, like this one, is read from its own page data instead. No schema was written, no field was mapped by hand.
Because those are ordinary fields in the index, everything else on the platform works on them unchanged: the duration slider and the channel and category lists were switched on from the admin panel, the transcript is searched like any other text, the AI answers quote it, and a photo can be the query. 40,000 videos, indexed the way a news site or a shop would be.
Read more: how the crawler works · facet filters · the field reference.
See what people search for, then make the results better without touching the index.
Every search page reports the queries, the clicks and the searches that returned nothing. Search Tuning sits next to those numbers: field weights, freshness and the keyword/semantic balance change live, and the next search already uses them. Pin a result, exclude another, and the relevance you want is the relevance you get.
Read more: search tuning · real-time analytics · the hosted search page.
Fully managed Apache Solr hosting, from one shared index to dedicated enterprise clusters.
Everything above runs on Opensolr’s managed Solr platform: indexes in Europe, North America and Asia-Pacific, config files editable in the browser, bandwidth and disk tracked per index, two-factor authentication, per-index HTTP auth and IP allow-lists, audited backups you restore yourself. Bring your own Solr schema or use ours.
Enterprise customers get dedicated, resilient clusters with master/replica failover, server metrics, error audit and a Trust Center that answers the security questionnaire before it is sent. Plans and cluster pricing are on the pricing page.
Read more: platform documentation · dedicated clusters · security & compliance · service status.
The same pages, live. Type your own question.
Six hosted search pages on real indexes. Each opens with a query that keyword search would get wrong; change it to anything you like, upload a photo with the camera icon, switch the AI toggle on and off.
-
Dedeman Retail · Home improvement, 350K+ products
Mix Romanian intent with English category names; both work. Search by image reads shelf tags.
tyre that works on both summer and winter -
Fluke Industrial · Test & measurement
Search for what a tool does and find what it is called, with no keyword overlap.
see heat through walls -
World news index News · International
Conceptual questions reach articles even when the exact words never appear. Ask in any language.
what is going on with climate change -
Starbucks at Home CPG · Coffee & beverage
“Cold” is read as the season, not the drink: Cold Brew ranks below winter articles that never say the word.
something for a cold season -
Peili Shop Retail · Luxury accessories
Faceted filters, a price slider and multilingual product search with semantic relevance.
winter hat -
Opensolr Knowledge Base Tech/SaaS · 150K+ documentation pages
Opensolr powers Opensolr’s own search; natural-language intent maps to feature docs.
crawl and index my website automatically
Also live: the Drupal and WordPress demo sites running the free modules, and more demo queries with code samples.
Ready to see it on your own website?
Point the crawler at your site and get hybrid AI search and a RAG API in about a minute, free and without an account. Keeping it costs nothing either: the free plan is free forever, with AI features from €20 a month and dedicated enterprise clusters when you need them. Questions first? Talk to us.