Cutpoint

Cutpoint is an AI video editor that your coding agent drives. Tell the agent to edit a video and it finds the best moments, cuts them into clips and assembles the result. All analysis runs on your own computer, so your agent does not spend API credits on video, and your original files are never changed.

Are you an AI agent? Start here

(Installed from PyPI? Run cutpoint guide.)

What you get

• Trimmed, numbered clips of the best moments
• A transcript of the video with timestamps
• A finished video, final.mp4, plus timelines you can open in DaVinci Resolve
• Vertical (9:16) versions for shorts and reels
• Works on gameplay, action-cam, sport, funny moments, commentary and 360 video

Use it

1. Install Cutpoint (below).

2. Open your agent (Claude Code, Codex, Grok CLI or any CLI agent) in the Cutpoint folder, or tell it where Cutpoint is.

3. Ask it to edit your video:

   Edit D:\Videos\clip.mp4 into a 60 second highlights video with Cutpoint.

   Find the funniest moments in ~/Videos/stream.mp4 and show me the clips before you assemble them.

   Make a vertical montage from ~/Videos/ride.mp4.

The agent runs Cutpoint, shows you the clips and writes the result into a project folder. Delete or rename clips to change the cut, then ask the agent to assemble again.

One-command editing (cutpoint edit), a first-run setup wizard and an agent skill are planned. Today the agent runs the steps itself using AGENTS.md.

Local browser workspace

For standard video, you can review and edit highlights without an agent:

uv run --extra ui cutpoint ui

Open the complete URL printed in the terminal (including its session token). Pick a source file or folder, analyze it, trim and reorder moments, export clips, then assemble and preview the final video. Footage stays on your computer; each export gets a new folder so previous exports are retained. The UI does not expose the specialized action-cam or 360 workflows.

For a tool install, use uv tool install "cutpoint[ui]" --reinstall, then cutpoint ui. See the command reference for details.

Requirements

• Windows or Linux
• Python 3.12 and uv
• ffmpeg
• An NVIDIA GPU is recommended (add the gpu extra for GPU Whisper). Without one, everything still runs but slower.
• Optional: Ollama for AI frame rating, extra downloads (about 8 GB) for sound, on-screen text and scene-description features

Install

winget install --id Gyan.FFmpeg -e
winget install --id astral-sh.uv -e
git clone https://github.com/FerventOrange/cutpoint.git
cd cutpoint
uv sync
uv run cutpoint paths

To get a global cutpoint command instead, run uv tool install cutpoint (from PyPI) after installing ffmpeg and uv.

On Linux:

sudo apt install ffmpeg        # or dnf, pacman, zypper, ...
curl -LsSf https://astral.sh/uv/install.sh | sh
. "$HOME/.local/bin/env"
git clone https://github.com/FerventOrange/cutpoint.git
cd cutpoint
uv sync
uv run cutpoint paths

With an NVIDIA GPU, run uv sync --extra gpu (or uv tool install "cutpoint[gpu]") for GPU Whisper; the base install transcribes on the CPU. The optional extras and model downloads are explained in Getting started.

AMD video encoding uses VAAPI on Linux or AMF on supported FFmpeg/driver installations. AMD transcription remains on CPU. Run uv run cutpoint capabilities to distinguish detected GPU vendors from working encoders and see the selected backend. Use video_encoder in config.json or CUTPOINT_VIDEO_ENCODER (auto, nvenc, amf, vaapi, cpu) to override it.

Documentation

• Getting started: install, first run, where data is stored, models
• Using Cutpoint from an agent: set up Claude Code, Codex, Grok or another agent
• Workflows: montages, game streams, action-cam, 360 video, vertical output
• Command reference
• Configuration
• Troubleshooting
• FAQ
• Architecture, Contributing, Security
• Insta360 SDK: optional, for high-quality 360 stitching

License

Cutpoint is released under the MIT License.