How Najimi uses Qloo
The taste graph behind every answer, the agent's Qloo tools, and how to see each call.
Najimi would not work without Qloo. Qloo's taste graph knows how things people love connect across domains: someone who loves Insecure and Americanah tends to love Jorja Smith, a certain kind of restaurant, a certain kind of room. Najimi asks Qloo, and builds everything from the answers.
Your taste, from three loves
Loves + weights
Qloo ids from /search, weights 1–10
/v2/insights × 10 domains
signal.interests.entities with {id, weight}; feature.explainability
/v2/insights tags
Aesthetic, decor, ambience, style, cuisine and dish words
Your taste page
Words, picks, 'because' chips, connections
- Weighted signals. Each love goes in as
{ id, weight }, so a love at 9 pulls harder than one at 3. - Explainability. Qloo says which love drove each result. That is the because chip on every pick, and the "what your taste connects" line ("Insecure led to Jorja Smith").
- Words. Tag insights over places and media give the vocabulary your taste is made of.
The agent's Qloo tools
The chat's model gets Qloo's whole surface as tools, not a narrow "recommend" button, and composes them per ask.
| Tool | Qloo endpoint | What it's for |
|---|---|---|
qlooSearch | /search | Turn a name ("Fela", "Kyoto") into a Qloo id to use as a signal, filter or comparison. |
qlooTags | /v2/tags | Find tags by meaning: moods ("rainy day"), genres, cuisines, decor, dietary options, place kinds. |
qlooInsights | /v2/insights | Anything the graph answers: picks of any type, a taste's words, a heatmap of a city, the demographic lean. Every documented signal and filter. |
qlooCompare | /v2/analysis/compare | Two tastes side by side: yours and a set of things, or Fleabag vs Insecure. |
qlooAudiences | /v2/audiences | Audiences (life stage, hobbies, communities) to lean a taste with. Never political ones. |
qlooInsights takes the person's loves as the signal by default, and the model adds to it freely:
| Ask | How the model composes it |
|---|---|
| "70s films I'd love" | ask: movie, filter.releaseYear {1970–1979} |
| "Hidden gems" | filter.popularity { max: 0.3 } |
| "Where would I hang out in Paris?" | ask: heatmap, filter.city: Paris → a map with named areas |
| "Is Kendrick a fit for me?" | filter.onlyIds: [his id] → a score |
| "Dinner near me on Sunday" | filter.tags: [restaurant], filter.openOn: Sunday, your city |
| "A trip around my taste" | ask: destination, diversifyBy: city |
| "Music for this room" | qlooTags "candlelit" → signal.tags |
Each answer becomes a card that renders whatever came back: picks of any type, a tag cloud, a map, an age and gender chart, a comparison.
Seeing every call
Every Qloo request is logged with its path, parameters, status, time, cache hit and result count.
- On every Qloo card, the Qloo chip opens the exact request behind it.
- On a taste page, Built on Qloo lists every call the reveal made, in order.
Answers are cached by a stable key (path, parameters and body), and identical requests in flight share one call, so a page reload or a retry costs Qloo nothing.