Recombee Alternative: A Recommendation API With a Written Reason for Each Pick
Recombee learns from what your users do. TasteRay reasons from what a person says right now and writes down why. Plenty of products should pick the first, and this page says which.
TasteRay is an alternative to Recombee when you want ranked picks from what a person says in their own words, with a written reason for each, and little or no interaction data. Recombee fits large catalogs with interaction data that need millisecond updates, rules and SDKs. The two can run side by side.
Recombee names catalog and interaction data as its two most important inputs and takes new interactions into account within milliseconds. TasteRay keeps no interaction model on the self-serve API: each request carries its own context, and the answer comes back as text a person can read. Most of the table below follows from that one difference.
Who each one is for
- Recombee
- Teams with large catalogs in video streaming, sports and live events, news, music and podcasts, e-commerce, travel or recruitment, who can send a stream of views, purchases and ratings. Its homepage shows DAZN and The Telegraph among others.
- TasteRay
- Teams whose user can say what they want, and who want the answer to carry its reasoning: a new subscriber, a gift buyer, a small catalog without much history, a typed or spoken request. You send up to ten candidates per request and show the written reason beside each.
Recombee and TasteRay side by side
Recombee counts interactions, recommendation requests, active users and catalog items. TasteRay counts requests. Read the unit before you compare a price.
| Question | Recombee | TasteRay | Source |
|---|---|---|---|
| What it is | "An AI personalization engine" for real-time recommendations and personalized search | A recommendation API that ranks your list or suggests its own picks, with a written reason per result | Recombee, TasteRay docs |
| Data to start | Catalog data. Its site labels interaction history optional, and also calls interactions one of the two most important data types | A sentence or a list of preferences in the request, plus up to 50 rated items if you have them | Data and results, TasteRay docs |
| New signals | New interactions and content used "within milliseconds", through incremental training | Nothing carries over between requests on self-serve. You send fresh context each time, and Enterprise can train on your content | Data and results, TasteRay docs |
| Why a result was picked | Not listed in their docs (as of Nov 2026) | why_match and key_factors on every result | Recommendations API, TasteRay docs |
| Rules | ReQL filters and boosters | Filter items before you send them. No boosting parameter in the docs | Recommendations API, TasteRay docs |
| Integration | REST API, SDKs in ten languages, low-code and no-code widgets | REST API called from your server, plus the ForMe widget for a single product page | Recombee, TasteRay docs |
| Free tier | Under 100,000 interactions and under 100,000 requests a month, under 20,000 users and items, without support | 1,000 requests a month, 5 a minute, community support | Recombee pricing, TasteRay docs |
| First paid tiers | Standard from $99 a month: 100,000 interactions and requests, 20,000 users and items, 99.9% uptime backed by an SLA | Basic $49 a month for 10,000 requests, Pro $199 for 100,000. Self-serve tiers carry no SLA | Recombee pricing, TasteRay prices |
| Above that | Pro from $1,699 a month, Premium from $4,499, Platinum custom | Enterprise, custom, with an SLA and limits set to your traffic | Recombee pricing, Enterprise |
What a Recombee control becomes on TasteRay
If your team already thinks in ReQL, this is the part that matters when you plan a move. Each Recombee parameter below comes from its recommendations API reference, and each answer is what you would do instead.
- reqlExpressions as a filter
- Run the same logic in your own query, then send only the survivors in
items. Withitemsset, the API only returns items from the list you send. - booster
- The TasteRay docs list no equivalent. To push a title, place it yourself outside the ranked block, or keep the campaign logic in your own code.
- minRelevance
- Read
metadata.taste_match_scoreon each result and set your own cut-off. Test it on your data first, since the score is the model's own estimate. - rotationRate
- The API remembers nothing, so rotation is yours: drop what a user has already seen from the next candidate list, or send those titles in
context.history. - returnProperties
- The response returns your own
idand the exactnameyou sent, and nothing else of yours. Match results to your catalog byitem.idand read every property from your database. Fewer results than you asked for can come back.
Running both
A team on Recombee does not have to replace anything to try this. Recombee keeps filling the home rows, where an answer has to be ready immediately and interaction data earns its keep. TasteRay answers a different moment: someone typing "something gentle, I'm wrecked", where your interface shows a reason under each result.
To start, pass a viewer's sentence and the top ten candidates from a broad Recombee or catalog query to POST /v1/recommend. Plan for a few seconds per response, returned in meta.processing_time_ms, and cache what repeats.
When to choose Recombee
- You have interaction data and a large catalog, and you want both reflected within milliseconds.
- You need a recommendation on every row of every screen, immediately.
- Your team wants a rule language, boosters and SDKs in the languages you already use.
- You want a vendor whose homepage shows streaming and sports references such as DAZN.
- You need an SLA at a published price. Recombee lists one from $99 a month, and ours comes with Enterprise.
- You want account-manager and Slack support at a published tier, as Recombee lists from its Pro tier.
When to choose TasteRay
- Your interface shows the reason, so a person can read it, disagree with it and decide.
- Your users say what they want in words, by typing or by voice, and you want that sentence to drive the ranking.
- You have too little history for a behavioural model, or you are entering a new category.
- A few seconds per live request is acceptable, or you can precompute and cache.
Recombee is a separate company; see its own site. TasteRay is not affiliated with it.
Sources and further reading
- Recombee homepage (Recombee, opened 5 November 2026)
- Data and results (Recombee, opened 5 November 2026)
- Recombee pricing (Recombee, opened 5 November 2026)
- Recommendations API reference (Recombee, opened 5 November 2026)
FAQs about Recombee Alternative
What is the main difference between TasteRay and Recombee?
Recombee personalizes from catalog and interaction data, updating within milliseconds. TasteRay reasons from the context in each request, such as a sentence from the user, and returns a written reason with every result. Choose by the signal you hold: behaviour at scale points to Recombee, stated intent points to TasteRay.
Does Recombee explain its recommendations?
Explanations are not listed in Recombee's docs as of November 2026. We checked its homepage, its data and results page and its recommendations API reference, which document filters, boosters and returned properties. Ask Recombee directly if a written reason per result matters to you, since a page can leave out what a product does.
Is TasteRay cheaper than Recombee?
It depends on the unit. TasteRay costs $49 a month for 10,000 requests and $199 for 100,000. Recombee starts at $99 a month for 100,000 interactions and 100,000 requests, and its free tier is far larger than ours. Generating text costs more per request, so a high-volume row usually favours Recombee.
Can TasteRay and Recombee run together?
Yes, because TasteRay only needs a list of candidates and some context in each request. Keep Recombee on home rows and send its top ten candidates to TasteRay when a user types a request or opens an explanation panel. Your server joins the two, and neither vendor needs to know about the other.