BridgeClip earns trust by telling you exactly what leaves your machine

BridgeClip takes a long video — a podcast file, a YouTube link, a Twitch VOD — and produces short vertical or horizontal clips with burned-in captions, entirely on your own machine. No BridgeMind server sees your footage. Transcription and moment-selection calls go directly to OpenRouter under your own API key; rendering happens locally via FFmpeg. The architecture is refreshingly honest about what leaves your computer and what doesn't.

The moment-selection logic uses an LLM to rank segments for virality, and the results land in a Library view with scores, timecodes, and tags. Nine caption styles — Viral, Hormozi, Bold, Clean, Minimal, Fire, Glow, Neon, Karaoke — each have a live preview before you commit to a render. Word-by-word timing comes from MAI Transcribe 2 (Quality mode) or Whisper Turbo (Economy), both routed through OpenRouter. If the source has no speech at all, BridgeClip samples video frames and sends those to OpenRouter for visual-only planning, though those clips get no speech captions.

Speed control is a small but considered feature: you can export every clip in a job at 1×, 1.1×, 1.25×, 1.5×, 1.75×, or 2×, with pitch preservation and caption sync maintained. Clip length references always refer to original footage — a 60-second clip at 1.5× exports in roughly 40 seconds before any dead-air cuts. That kind of concrete, user-facing arithmetic in a README signals actual attention to the editing workflow rather than a demo.

The Twitch VOD support has explicit constraints worth knowing: completed public VODs only, no live channels, no subscriber-only content, source capped at six hours and 20 GB, downloads halted if free disk space would drop below 1 GB. These aren't buried footnotes — they're in the README alongside the feature description.

For developers, the stack is Electron with a React renderer, a Python 3.12 clipping engine, and yt-dlp for downloads. Dependencies are hash-locked. Linux development is supported but a packaged Linux binary isn't available yet; macOS and Windows installers are signed and notarized. Auto-update checks every four hours and verifies downloads against a published SHA-512 before installing.

The social publishing path routes through a third-party service called Zernio, which handles platform authentication and media upload. The README is explicit that Zernio's own data policies govern what happens to clips once they're uploaded — a reasonable disclosure that most tools in this space quietly omit.

What BridgeClip is not: a hosted SaaS with a free tier, a no-code tool for non-technical users, or a finished product with Linux support. What it is: a well-structured, MIT-licensed desktop application that treats API costs and data flows as first-class concerns rather than fine print.