Cutmark keeps video projects on your Mac and lets your own AI agent help edit

What Cutmark is

Cutmark is a video production app for the Mac, built by TriHerm Labs and created by Fernando Abishai. As github.com reported on its project page, the idea is a single workspace where a person and an AI agent handle the same job: recording talking-head video, then moving through scripts, storyboards, edits, review notes, thumbnails and drafts of the posts that go with a video.

The key design choice is that the source files stay on your computer. You do not upload footage to a Cutmark account, and the project says it does not need a hosted project database for this workflow. The agent you already pay for or run does the assisting.

Our newsletter flagged it as early, with 3 points and no comments on Hacker News, and that still holds. The project page is the main evidence, and we have not tested the app ourselves.

Bring your own agent

Cutmark does not ship its own AI model. During first-run setup you pick one of four command-line agents: Claude Code, Codex, pi or opencode. The page says nothing is installed until you choose a provider and press Install.

Inside the app, a chat panel talks to whichever one you chose. Under the hood, the project says each provider gets its own adapter, a small translation layer, while the rest of the app works against one shared interface. In practice, that should make it easier to switch agents without the app changing around you.

Each of those tools keeps its own login and model settings. Any billing and privacy terms come from the provider you pick, not from Cutmark.

How the files and tools fit together

The default library lives at ~/Movies/Cutmark, in a normal folder structure. Projects contain sessions, and sessions hold scripts, recordings, edits, storyboards, thumbnails, post files and metadata. You can move the library from within the app.

Editing relies on ffmpeg, a widely used free tool for processing video, and on Whisper-based tools, which turn speech into text. The app also exposes the same workspace through MCP, the Model Context Protocol, a standard way for AI apps to connect to outside tools. That means other compatible agent apps on the same Mac can work with your projects, using either stdio or Streamable HTTP connections.

For ChatGPT, the page points to OpenAI's Secure MCP Tunnel on a private Mac and refers readers to a separate document in the repository.

Privacy claims and their limits

The project makes several specific privacy statements. The release sets up no usage analytics endpoint, and the HTTP server for MCP listens only on the loopback address, meaning only your own computer can reach it. The page warns against exposing that server to the public internet and says a public setup would need proper authentication and separation between users.

There is a catch that the page states plainly. Your video, scripts and chats stay on the Mac unless the provider you chose, or an action you take such as publishing or generating something, sends data to an outside service. If your agent runs in the cloud, the text you give it goes there.

Camera plus screen recordings may also save a small side file with pointer movement, clicks, scroll amounts and manual markers, for editing features later. The page says keystrokes are not recorded in this format.

Requirements and rough edges

You need an Apple silicon Mac running macOS 15 or newer, Apple's Command Line Tools, ffmpeg and ffprobe, and at least one configured agent. The first-run setup can guide you through missing pieces.

The notable rough edge is signing. Version 0.3.0, like 0.2.0, uses an ad-hoc signing channel and has not been notarized, which is Apple's automated security check for downloaded apps. On first launch, macOS may make you approve it through System Settings under Privacy & Security. The repository has a separate production path with a stricter Apple signing process, but it waits on Apple credentials being set up.

The iPhone companion source is in the repository but is not part of the current macOS release. Cutmark is under the MIT License, and you can build it yourself from the source using the included test and build scripts.