AVE — a local Mac video editor that turns Whisper transcripts into searchable, reviewable timeline edits

[lemberalla]

I am building AVE, a local-first video editor for Apple Silicon Macs.

Most AI video tools begin with generation. AVE begins by understanding the footage already on your Mac.

AVE uses MLX Whisper for local transcription on Apple Silicon. Transcript segments retain their source time ranges and, when available, word-level timing. That transcription drives several editing workflows:

• Search existing footage by spoken words, phrases, or dialogue.
• Jump directly to the matching source moment.
• Extract matching ranges into the editing timeline.
• Generate captions from transcript timing.
• Remove silent pauses or filler words while keeping the resulting timeline change reviewable.

Visual analysis sits alongside the transcript rather than replacing it. A search result can therefore be supported by spoken content, visual content, or both, with the original source range preserved.

AVE also exposes scoped editing tools to locally connected coding assistants such as Codex CLI and Claude Code. The assistant works with structured project information and allowed review frames; source footage does not pass through the MCP connection. Proposed timeline changes remain reviewable before export.

Current boundaries:

• macOS 14 or later on Apple Silicon, M1 or later.
• AVE is a video editor, not a hosted transcription service or standalone ASR SDK.
• Media analysis is local. Assistant-driven workflows still use the connected assistant selected by the user.

Product: https://www.aivideoeditor.app/

More about the footage-first approach: https://www.aivideoeditor.app/blog/ai-video-editing-starts-with-footage/

I would be interested in comparing notes with anyone using Whisper timestamps as an editing primitive rather than only producing a transcript file.