Most idea tools invent plausible titles. This one fetches real competitor videos, works out why they performed, and every idea it returns links back to the videos behind it. If there is no evidence, there is no idea.
Google and GitHub sign-in. Spend limits on by default.
Ideas reference evidence by index into a list the app controls, so the model cannot cite a video that does not exist. An idea with no source is rejected before it is stored.
Competitor descriptions and pages reach the model inside delimited untrusted blocks. When content tries to give the agent instructions, the run says so instead of quietly obeying.
Every provider call is metered against a per-run ceiling. A run that reaches it stops and tells you, rather than continuing on your card.
One spoken request runs the whole chain. You watch each step land in the workspace while the agent narrates progress.
Fetches real competitor videos for your niche and scores each against its own channel baseline, so a big number from a big channel is not mistaken for an outlier.
Finds the repeatable reason videos worked — a format, a promise, a hook shape — not a summary of what they were about.
Builds distinct angles on those patterns for your audience. Every idea must name the evidence it came from.
A separate agent rejects vague, unsupported or duplicate ideas. Survivors are scored on six dimensions and ranked by arithmetic, not vibes.
Writes title options, a hook, a beat-by-beat outline and thumbnail concepts for the strongest ideas.
Ask once. Get outliers, patterns, scored ideas and a brief you can film against — each one traceable to the video it came from.
Start researching