Everything in EditAI

One native Mac app for multi-camera video: line the angles up, let Claude cut between them, then fix anything by hand. Here is the complete list, grouped by what you are trying to do.

Import and sync

Drop several cameras: Drag in one MP4 or MOV per camera (or choose them all at once). Add more cameras to a project at any time. Projects are plain folders in ~/EditAI Projects/.
Synced by sound: Each camera is matched to the others by its audio using GCC-PHAT cross-correlation, so different microphones, gains and room tone still line up. Offsets are found to a fraction of a millisecond, and cameras that started minutes apart are fine.
You can check it: Every camera shows its offset and how strong the match was. Nudge by a frame or 100 ms, or re-run the sync. Cameras you set by hand are left alone.
Waveforms and previews: A waveform and a filmstrip of stills are built for every camera in the background while you start working.
Transcript, on your Mac: whisper.cpp runs on the Metal GPU (Fast, Balanced or Best). The model is downloaded once from Hugging Face, only when you ask, and the transcript is what lets Claude follow the conversation.

Claude directs the cut

Local MCP server: editai-mcp ships inside the app. Claude Desktop or Claude Code launches it on your Mac. Nothing is hosted and there is no account or API key.
Reads the show: The transcript comes in numbered sentences with pause markers, a few thousand tokens for a long episode when asked by range.
Sees every camera: get_frames returns one picture, a row per moment and a column per camera, so Claude knows who is on which angle. It can leave a note on each camera.
Cuts with switch points: set_edit takes "from here show camera B" points and ranges to remove. The result appears live in your timeline as one undoable step.
First pass and tightening: auto_cut and remove_silences do the mechanical parts without spending tokens on them.

Timeline and manual editing

One lane per camera: Filmstrip, waveform and the stretches where that camera is on air. An edit lane above shows every shot and every section you removed.
Paint a camera: Drag across a camera lane to use it for that stretch. The shots split and merge for you.
Cut live: Press 1 to 9 to cut to a camera at the playhead, paused or while playing.
Trim, split, delete: Drag a shot's edge (it snaps to sentence starts and other cuts; hold ⌥ for none), press S to split, ⌫ to delete and close the gap.
Choose the sound: Use any camera's microphone for the whole edit.
Undo everything: Unlimited undo and redo, and the project saves itself after every change.

Auto-cut and silences

Auto-cut: Rotates through your cameras on sentence starts, with your shortest and longest shot, and never shows a camera that has no footage at that moment.
Remove silences: Ripple-deletes pauses longer than you choose, leaving 0.24 s before speech and 0.14 s after so nothing is clipped.

Export

One video from every camera: 4K, 1080p, 720p or the camera's own size, H.264 or HEVC, AAC audio, at the fastest camera's frame rate.
Same plan as the preview: The preview and the export are built from the same composition, so they match.
OS-licensed encoders only: Encoding uses AVFoundation and VideoToolbox. There is no ffmpeg and no openh264 in the app.

Private by design

Everything on-device: No telemetry. The only network use is the optional model download from Hugging Face.
Open source: MIT licensed, written in Rust with egui. No web view, no Electron.

Download EditAI