Installing components and models
Setup fetches processing components from GitHub and PyPI, and model files from Hugging Face. Files are pinned to known versions and verified against their expected hashes. Your projects are not sent.
Your footage stays yours.
A clear guide to what runs locally, what connects to the internet, and what you control.
Local by default.
Your source video and local analysis stay on your computer. Once setup is complete, you can analyse and export without an internet connection.
What happens on your computer
The default workflow handles your source video, transcription, clip selection, visual checks, editing, and export locally. Your recordings do not need to be uploaded for that workflow.
A local Qwen model proposes and reviews moments through MLX on Apple silicon or llama.cpp on Windows and Linux. The built-in picker is available without that model. You choose which clips to keep and how to finish them.
When ClipMill uses the internet
Setup, imports, model acquisition, and optional update notices use a connection.
Finding a model
Looking up a model in Hugging Face from the Models area contacts that service. Downloading the selected model is a separate action you start.
Importing from YouTube
The importer downloads the video you selected, then passes it into the local workflow. This does not sign in to your channel or enable cloud analysis.
Checking for a newer version
When update notices are enabled, ClipMill asks GitHub for the latest release once a day. The first check waits a day after installing. It sends no credential and installs nothing. Turn off Say when a new version is out in Settings → About to disable it.
Cloud analysis is a choice
If you separately enable optional cloud processing, each analysis still needs your consent, a supported provider and model, and a spending limit.
Cloud clip proposals and reviews send transcript context to that provider. Source frames and visual checks remain local. A local task does not automatically switch to a cloud provider if it fails.
Reading the Local Lock indicator
Local Lock is engaged when no network-enabled operation has started in the current background-service session. Starting an operation such as a model download, YouTube import, update check, or cloud analysis disengages it until that service restarts.
“Network operations started this session”
This counts operations the app starts, including work fulfilled from a cache. It does not count packets or bytes, or prove that a remote request completed. Restarting the service resets the session count; job history is kept separately.
What it guarantees
ClipMill checks network permissions in its task system, and local workers reject cloud work. Optional cloud work needs separate consent.
These are controls inside ClipMill. The Local Lock indicator is not an operating-system firewall, and the desktop launcher does not put its worker processes in a network sandbox.
Your files, on your disk
Projects, components, models, and logs are stored in your local application data folder. Use Settings → Storage to see how space is used and remove items the app can safely clear.
Removing the app and deleting its data are separate actions. The installation guide lists the folders for each operating system and explains how to uninstall.
Find your local data folderVisits to this website
This website uses Vercel Web Analytics to measure page visits and referral sources. These statistics help us understand how people find and use the website.
This analytics integration runs on the website; the desktop app’s processing is described above.