Your footage. Your control.
Browser privacy
Videos, captions and reference notes selected in the public browser editor are processed on your device. They stay local unless you explicitly share sampled images and text with your connected assistant for a request. Original video and audio files are never uploaded by this connection. Exported MP4 files are created in your browser. The hosting service receives ordinary web requests, including your IP address and browser request information, when you load the site. There are no advertising trackers or analytics scripts.
Connecting your assistant
Each request requires separate approval of its displayed sampled images, transcript and direction. Approved samples pass through a temporary Snagedit relay to your chosen assistant. Pending requests expire after ten minutes and are cleared on cancellation or completion; original video and audio stay on your device. Temporary data can be lost if the service restarts. OAuth connection metadata is stored separately for authorization. Access expires within one hour and can be revoked in Connection settings. Your assistant provider’s usage and retention terms apply to anything it reads. These tools cannot apply edits, export, publish or browse your files. Direct prompting requires a compatible host and an active prompt bridge; MCP alone does not supply model inference.
Save your work
Use Save edit backup to download your settings. Keep the original video separately. Accepted project edits can recover in this browser. Keep original footage and download a project backup; browser storage can be cleared or unavailable. Backups include your caption, style notes and reference link, so share them thoughtfully.
Current capabilities
The browser editor offers trimming, framing, timed captions, transcript selections with reviewed word removals, Taste preferences and reference notes. It accepts source files up to 500 MB and cuts up to 90 seconds. Export requires compatible browser encoders; a current Chrome or Edge browser is recommended. Videos without compatible codecs may be rejected. Review the downloaded video before sharing it.
Reference links are notes only. The assistant connection is under private evaluation. Compatible MCP assistants can inspect approved sampled images and text and return candidate-based pacing variations for preview. It does not provide unrestricted effects, link ingestion or automatic publishing. Optional private transcription downloads approximately 762 MB of pinned Whisper model files from Hugging Face and runs on your device. The model host receives ordinary download requests, never your audio. Browser word boundaries are estimates; preview and adjust every join. Only measured quiet gaps are proposed automatically; review each join. Automatic filler removal is withheld. Release the model to free memory or remove its browser download in Captions. Model-assisted reference analysis and native audio polish remain desktop features under private evaluation. Consumer ChatGPT or other chat-plan login is not enabled. The desktop installer is not publicly distributed yet.
Open-source notices
Browser media processing uses unmodified Mediabunny 1.61.6 under the Mozilla Public License 2.0. License · Source. Corresponding package source. The distributed library is available at this URL. Browser codecs are supplied by your browser and operating system.
Private speech uses Transformers.js 4.3.1 (Apache-2.0), ONNX Runtime (MIT), and Whisper large-v3-turbo (MIT). Versions and model revision · Transformers license · ONNX license · Whisper license.
Contact
Updated October 18, 2026.