We evaluated a lot of TTS providers before making Fish Audio the default in InsForge’s model gateway. It came down to quality, speed, and reliability — and Fish Audio checked all three. Now we’re running it across 15,000+ developers and over 1B tokens/month, and it’s been rock solid. The thing that stands out most is the model quality — you can tell there’s real research depth behind it. And the team ships fast and stays close to partners, which matters a lot when we’re scaling.

Been testing Clipto with a local media folder and a few older videos on an external drive.

I had a four-minute Chinese video recommending three books for people working in growth and operations. Normally I’d have to scrub through it to find the titles again.

After connecting the library to Codex through MCP, I asked the question in natural language, and Clipto surfaced the source video with timestamped clips for the exact moments where each book was mentioned.

That’s probably the part I like most so far. It turns a folder of media into context an agent can actually use, without uploading files one by one.