Skip to content
House Mix // crowd programming for venues, restaurants, promoters and sponsors

Program the nightfor the crowd that'scoming.

Give House Mix tonight's bill and city. Claude agents read the crowd with Qloo taste data and program the walk-in music, openers, food and drink, sponsor activations and the after-party, with the Qloo evidence behind every pick.

Plan a show+Open a finished plan+

How it works.

  1. 01 // describe · find_tags
    Resolve

    Every name becomes a confirmed Qloo ID. An ambiguous one stops and asks you; nothing is auto-picked.

  2. 02 // compare_audiences · audience_demographics · entity_tags · trends
    Read the crowd

    Shared and distinct taste between the two sides, aggregate age skew, the crowd's vocabulary and momentum.

  3. 03 // recommend · rank · where_popular
    Program

    One agent per slot: walk-in, openers, food and drink, sponsor activations, after-party, plus a stretch pick.

  4. 04 // call ledger
    Cut

    The Critic removes any pick or sentence the cited Qloo call doesn't support. The trace shows why.

  5. 05 // rank (swap) · briefing
    Lock

    Approve, swap or reject line by line. Lock a version, export the run-of-show, brief the crew.

// File 026 · the record

Every pick is on file.

// Powered by Qloo

Qloo's taste AI maps what audiences love across music, film, TV, podcasts, books, brands, dining and places, as anonymised aggregate affinities. House Mix chains all ten of its tools through the qloo mcp harness. Without Qloo there is no plan: no IDs, no crowd read, no candidates, no ranking.

// For sponsors and brand partners
// Ask House Mix

// Tested on live Qloo · 30 Oct 2026

Methods and the negative result →
10/10
Qloo tools verified against the live API (harness 0.1.26)
182/182
Eval-plan picks with a valid live Qloo citation (the Critic removed 23 more)
76/157
Claude-alone picks that are ambiguous in Qloo: a person must choose before they can be used
0/32
Sponsors repeated from one show to another (Claude alone: 6/28). Openers bigger than the headliner: 0/28 (Claude alone: 4/10)
155/182
Plan picks that carry Qloo's own reason; the same call asked twice gave the same answer 5 of 5 times; the same bill in another city keeps only 26% of Qloo's top 10
1–5
Blind A/B vs Claude alone, automated judge (1 tie, 7 test shows; 0–7 before the fixes). Claude alone still wins this one
Human blind ratings are open at /rate; the automated judge is a secondary signal until they are in.
// Index · 00 plans

Open a finished night.

Compare two shows+