How the AIworks.
Trackgleam is not one prompt. It is a short pipeline of focused steps, each with one job and its own checks, built so the output reads like a careful publicist wrote it and says nothing you did not say first.
Any good model,one standard.
Trackgleam is model-agnostic. The language models, the image model, and the voice are swappable parts behind one interface. What does not change is the fact sheet, the checks, and the schema every piece has to pass.
OpenAI or Anthropic
Both sit behind one interface. Routing is set in our server configuration, and either can take over if the other is unavailable.
gpt-image-2
OpenAI's image model for cover directions. The model is a configuration setting, not a rewrite.
ElevenLabs, optional
Narrates the radio intro when it is configured. Otherwise you get the script as text.
Schema checked
Every response is parsed and checked before it reaches you. Malformed output is retried, never shown.
Watch it writeyour kit.
Put a real single through the demo and read what comes back. Then check it against what you told it. How we handle care.