FAQ
Short answers to the questions we get most.
Which platforms are supported?
macOS on Apple Silicon and on Intel, as two separate downloads, and Windows 10 or 11 64-bit. The prerequisites page lists versions, chips, and which features need what.
Which Mac download do I need?
Apple menu → About This Mac. “Apple M1 / M2 / M3 / M4” means the Apple Silicon build; “Intel Core” means the Intel build. The wrong one does not open.
What leaves my machine?
Sign-in talks to app.agent28.com. When you ask the chat for something, your request and the transcript or timeline context it needs go to the model provider: Agent28's managed models, your own key's provider, or the ChatGPT / Claude app you are using. Local transcription engines never upload audio; the OpenAI engine does. Rendering, scene indexing, and person matting run entirely locally. Footage is never used for training without your opt-in.
Where are my projects stored?
In Agent28's own data folder on this computer, inside local databases rather than as a project file. They are not synced. To share a result, export the video; to move footage, keep the source files.
Can I edit without the AI?
Yes. Under the chat is a multi-track editor with keyframes, masks, 17 blend modes, speed control, effects, transitions, and per-element properties. Every AI change is one of those same edits, so you can adjust it by hand afterwards.
Why does the editor say GPU acceleration isn't available?
The renderer fell back to software rendering. The editor works, but preview and export are slower. Updating graphics drivers on Windows, or closing other GPU-heavy apps, usually restores it.
Do captions or exports use AI tokens?
No. Local transcription, caption rendering, person matte, scene indexing, and exporting never consume tokens. Only the managed AI in the chat panel does, and only on paid plans.
Where do I report bugs or request features?
Use the feedback option in the app menu, or the contact options on this site. Early access feedback goes straight to the team.