How it works

An agent with an audience behind it.

A language model on its own can only guess which film would make a hit musical. Overture gives it Qloo's taste data, a score tested on 58 real shows, and strict rules about what it can claim.

  1. 01

    Brief

    The producer lists films and a goal in plain words.

  2. 02

    Agent

    Gemini plans the steps and calls tools. It can only state what a tool returned.

  3. 03

    Qloo

    Every audience question goes through Qloo's official MCP server, from the server side.

  4. 04

    Evidence

    Results are scored against 58 past Broadway shows and shown with their source.

  5. 05

    Memo

    A one-page greenlight memo, with the evidence behind each claim.

The Qloo tools it uses

qloo_describeMatch each title to one exact filmAmbiguous titles like Ghost come back with choices. The agent picks the likely one and says so.
qloo_rankScore audience fitRanks the films for fans of six hit movie musicals. Scores are stable across calls, so they sit on the back-test scale.
qloo_entity_tagsThemes and stage checkWhat the film means to its fans, and a flag when Qloo already links it to Broadway.
qloo_audience_demographicsWho the audience isAge and gender lean, shown as over or under the average.
qloo_compare_audiencesCommon groundWhat the film's fans share with musical-theatre fans.
qloo_where_popularWhere to tourHeatmap cells across the US, rolled up to touring markets.
qloo_recommendTalent and partnersPerformers, musicians, brands, podcasts and TV the combined audience loves.

One real request

The scoring step from the example run. The options are the producer's eight films; the signals are the six musical-fan films: The Greatest Showman, Les Misérables, Mamma Mia!, La La Land, Chicago, Into the Woods. IDs are shortened here.

Request

qloo_rank {
  "option_type": "movie",
  "options": ["48788AF8…", "DEF93FF9…", …8 films],
  "signals": ["A572B5F1…", "B8A7FAAA…", …6 films]
}

Result, then placed on the back-test scale

Notting Hill  affinity 0.862  -> beats 97%
Clueless      affinity 0.858  -> beats 93%
...
Groundhog Day affinity 0.749  -> beats 24%

Data and trust

  • Key stays on the server. The browser never sees the Qloo or Gemini key.
  • No personal data. Only film titles and public names go to Qloo. Results describe groups, never one person.
  • Claims need evidence. The agent is told to state only what a tool returned, and every Qloo call is listed in the run's cue sheet.
  • Light on the API. Calls run one at a time and answers are cached, so a saved run replays without new requests.
  • Limits are stated. The score is one input, not a forecast. See the back-test.