Hi Product Hunt, I'm Ahmad. I kept writing the same glue code. Upload a video somewhere, call a model, parse whatever came back, then add defensive code for the times it came back in a shape I hadn't asked for. That last part is what I got tired of. Fovea does it properly. You send a video and a JSON Schema, or just a sketch of the shape you want, like {"decisions": [{"at": "string", "text": "string"}]}. What comes back has already been validated against that schema. If the model's answer doesn't conform, Fovea retries internally instead of handing you something half-right. Conformance is the guarantee. Judgement isn't, and I'd rather say so up front. Any field can carry a timestamp. The shortest way in is one line: claude mcp add --transport http fovea https://api.fovea.run/v1/mcp --header "Authorization: Bearer $FOVEA_KEY" Codex works the same way. There's one REST endpoint if you'd rather curl it, and a web app if you'd rather click. 30 free minutes on every account, no card. Prepaid top-ups are coming soon. What it doesn't do yet, so you don't find out the hard way: videos are capped at 10 minutes, there are no webhooks so everything comes back on the same request, and payments aren't live. Videos are deleted once the job finishes, and nothing sent to Fovea is used for training. One thing I'd like to know: what are you actually trying to pull out of a video, and where does the schema start getting awkward? That's the part I want to fix next.

Fovea

Video in, your JSON out — validated against your schema

Video in, your JSON out — validated against your schema