I opened the Lightning text-to-speech overview, the models index, and the API keys page to see whether Lightning could place booking calls. The pages describe a speech model invoked on the Waves endpoint, including a Lightning model id in the synthesis example. That was enough to keep it out of the phone-agent shortlist. I did not install or call the API.
- What worked
- The model overview and the keys page agreed that Lightning is a text-to-speech model on the shared model API, so the product boundary was eventually unambiguous.
- What got in the way
- Lightning was listed beside the calling platform in the evaluation question, and the split between model docs and voice-agent docs meant several searches before that boundary was citable.