Game Recommendation API for Video Game and Board Game Stores
How someone plays matters more than the genre tag: thirty minutes after work, alone, nothing punishing. TasteRay ranks your games against that. Players check facts, so your data states them.
TasteRay is a game recommendation API for video game and board game stores. You send what a player says and up to ten games from your catalog, on any platform or none, and one request returns them ranked with a written reason. Keep playtime, platform and age rating in your own data, because the model can get them wrong.
A store full of games is hard to browse by genre: "roguelike" covers a calm deckbuilder and a brutal action game. Players describe what they want by how long they have, who is with them and what they refuse to do again. They also know the games, so a wrong claim on your page is noticed fast.
Thirty minutes after work, video games and board games in one list
We sent five kinds of play in one request: action RPGs, a party game, a campaign board game, a kart racer and a farming sim. The player plays alone for 30 minutes and dislikes punishing difficulty. Elden Ring, Gloomhaven, Codenames and Mario Kart all stayed out, which is the part worth noticing.
{
"vertical": "games",
"context": {
"profile": "Plays for 30 minutes after work, alone, likes cosy and calm games. Dislikes punishing difficulty.",
"preferences": [
"relaxing",
"short sessions"
]
},
"items": [
{"id": "g1", "name": "Stardew Valley", "description": "Cozy farming sim, play in short sessions, solo.", "metadata": {"type": "video game", "players": "1"}},
{"id": "g5", "name": "Wingspan", "description": "Calm engine-building board game about birds.", "metadata": {"type": "board game", "players": "1-5"}},
// ...and 8 more: Elden Ring, Codenames, Gloomhaven and others
],
"options": {"count": 3}
}meta.processing_time_ms 1,824. The model's sentences are verbatim.Stardew Valley
98 out of 100, taste_match_score
You're not looking for a challenge; you're looking for a sanctuary. Stardew Valley is the digital equivalent of a warm bath and a quiet evening with a good book—perfect for decompressing after a long day alone.
- It’s a 'short session' champion. You can plant seeds, chat with one villager, and leave feeling accomplished and relaxed. It’s a complete experience in 30 minutes.
Cult of the Lamb
82 out of 100, taste_match_score
This game wraps its darker theme in a sugary, cute aesthetic that feels incredibly safe and 'cozy'. It lets you build a community (village) which adds a layer of warm connection to your solo play, without the frantic pressure of pure action roguelikes.
- The management loop is highly suitable for short 30-minute sessions. You can spend the time just tending to your village, upgrading buildings, and feeling a sense of domestic accomplishment.
Fields shown: item.name, metadata.taste_match_score, explanation.why_match, explanation.key_factors.
What we withheld, and why it matters here
The second result was Wingspan, scored 88. Picking a calm board game for a solo player is a fair idea. We left its text out because it placed the game inside a 30-minute window, while the publisher lists 40 to 70 minutes. We had not sent a playing time, so the model supplied one.
We also removed a key factor on the top result. It said Stardew Valley has no timer and no enemies, and a player who has watched the in-game clock run down or met the monsters in the mines would laugh at that. On Cult of the Lamb we kept one sentence about the village loop and dropped a broader claim about avoiding combat, because the game does have combat, and a player who dislikes difficulty needs to know.
None of this makes the ranking useless. It tells you how to build on it: let the model argue about fit, and take the checkable facts from your own data. potential_concerns in the response is also worth reading for the caveats it raises.
What changes in a game store integration
- Send the numbers the model should not guess
- Put playing time, player count and platform in
descriptionandmetadata, then show them on the card from your own database. Our request gave player counts and no playing times, and the wrong claim was about playing time. - Platform comes before taste
- A perfect match that does not run on their console is a wasted slot. Filter by what the player owns before the call, and the API only returns items from the list you send.
- Their library is the best history
- Send owned games with a rating you can defend, such as many hours played as liked and a quick refund as disliked, up to 50 per request. Take owned games out of the candidates at the same time.
- Editions and add-ons
- A base game, a deluxe edition and three expansions look like five items and can fill the list. Collapse them to one game, and offer the add-ons only to someone who owns the base.
- Age ratings
- Filter on your own rating field, before the call, whenever you know or must assume the shopper's age. A reason is no substitute for a rating.
- Mixed shelves
- One
vertical: "games"request took video games and board games together. If you sell only one kind, say so invertical, since it is a free string.
Where it fits in a store
Use a request for a finder ("what should I play tonight?") or a prepared row by situation, such as "solo, under an hour", cached from your own list. On a single game's page, ForMe answers "Is this for me?" with one script tag and a button. It accepts any non-empty medium: its install page lists computer-game, and board-game worked in our test. See Stores for the install.
If you need "players also bought" across a very large catalog on every page, a behavioural model on your own events fits better. Compare Recombee and build vs buy.
Sources and further reading
- Wingspan (Stonemaier Games, opened 5 November 2026)
- TasteRay API documentation
FAQs about Game Recommendation API for Video Game and Board Game Stores
Can TasteRay recommend video games and board games together?
Yes. An item is anything you can name and describe, and in a capture on 5 November 2026 one request held video games and board games side by side and returned both. Filter by what the player can actually play first, such as platform or player count, before you send the list.
Are the game facts in TasteRay reasons reliable?
Not always, so check them. In our capture the model gave a board game a playing time shorter than its publisher lists, and said a farming game had no timer or enemies. We left those out. Send playing time and platform yourself, show them from your data, and treat the reason as opinion.
How do we use a player's library with the API?
Send up to 50 rated games per request in context.history, using a rating you can defend, such as long play as liked and a quick refund as disliked. Remove owned games from the candidates at the same time. The API stores nothing between requests, so your server sends this each time.