OpenChatCut: A Local Video Editor for AI Agents With a Real Timeline

OpenChatCut is a local video editor for AI agents, built so an agent can cut, caption, and grade a project directly in a native multi-track timeline instead of talking to a plain document. Projects and media stay on the machine running it, and nothing leaves the device unless a cloud provider is deliberately wired in. That puts it in the same rough category as FableCut, the browser-based, JSON-timeline editor covered earlier on this site, but the two solve the agent-editing problem in close to opposite ways. FableCut skips the timeline and hands an agent a single project file to patch. OpenChatCut keeps a full timeline, and wraps agent edits in a formal review step before anything reaches the live project.

What this local video editor for AI agents actually does

OpenChatCut is an Electron desktop app (also runnable from source in a browser during development) that pairs a professional non-linear timeline with a conversational agent panel. The timeline itself is not a toy: multitrack editing, transitions, keyframes, ripple edits, motion graphics, and WebGL-based effects with custom shaders, rendered through Remotion. Export options cover MP4 (hardware-accelerated on both macOS and Windows, with a software fallback), separate audio tracks, SRT captions, FCPXML for round-tripping into Final Cut Pro, and the full project as JSON.

The editing surface is transcript-driven as much as it is timeline-driven. Word-level transcription lets a person or an agent make cuts by deleting words in a text view, compress pauses automatically, identify speakers, and generate captions that stay linked to the underlying audio. That transcript layer is what an agent actually reasons over when asked to tighten a scene or remove filler, rather than guessing at frame numbers.

Agent access comes through Model Context Protocol. The repo ships an MCP endpoint any external agent can connect to, plus documented one-line setup for Claude Code and Codex specifically. On top of that sits "Agent Skills," OpenChatCut's own internal skill system: running npx skills add 0xsline/OpenChatCut registers a local MCP router and loads 26 specialized skills on demand, with custom skills droppable into ~/.openchatcut/skills/<slug>/SKILL.md. Built-in and external agents share the same underlying editing tools, so a skill written for the in-app assistant works the same way over MCP.

The draft, review, approve workflow

The detail that most separates this from a tool where an agent edits a shared live file is how OpenChatCut isolates agent changes before they count. An agent opens a session with begin_edit_session, working in manual approval mode by default (an auto mode exists for cases where review isn't wanted). Every timeline tool call inside that session acts on a draft, not the live project. When the agent calls review_edit_session, a person previews the proposed changes inside the actual editor and either approves or rejects them. An approved session commits as a single undo step; a rejected one leaves the live timeline untouched.

Manual edits from a person and agent edits both route through the same underlying command system, so everything is inspectable and reversible the same way regardless of who made the change. It's a more cautious model than watching an agent write directly to a shared file and catching mistakes as they render, and it's the practical answer to the "keep a normal editor's controls" half of what makes this tool distinct.

Setup, platform support, and the license difference

Desktop installers are available for macOS (Apple Silicon and Intel), Windows x64, and Linux x64. Running from source needs Node.js 24.x; npm install, copying .env.example to .env.local, and npm run dev bring up a local dev server. The core timeline, local projects, built-in media, and manual editing genuinely don't need any cloud service to work. Optional API keys, for image, video, speech, or music generation, or for transcription, live server-side in .env.local and are never exposed to the browser. Built-in agent models cover Anthropic, OpenAI (including the Codex CLI), Gemini, and several others (Kimi, Qwen, GLM, DeepSeek, MiniMax, Mistral among them), so the choice of model isn't locked to one provider.

One thing worth flagging plainly: OpenChatCut ships under AGPL-3.0, a copyleft license, not the permissive MIT license FableCut uses. AGPL requires that a modified version made available over a network also have its source published under the same terms. That's a real consideration for anyone thinking about building a hosted product on top of this codebase specifically, even though it changes nothing for someone just running it locally or scripting against it as a personal tool.

Where this fits and where it doesn't

This is the right pick when the work involves footage that shouldn't leave a machine, when the edit needs a real timeline with transitions and effects rather than a flat cut list, or when a person wants to see and approve an agent's proposed edit before it becomes permanent. The transcript-driven editing and linked captions make it a reasonable fit for talking-head or interview footage specifically, since so much of that editing is really text editing.

It's also a heavier install than a single-file browser tool: a desktop app, Node 24 if building from source, and potentially several API keys depending on which generation features get used. The AGPL license is a real constraint for anyone building a commercial derivative rather than using it as-is. And this is a young, fast-moving repo from an independent developer, not a maintained commercial product, so the usual caveats about depending on a single-maintainer open-source project apply.

The traction, verified

As of today, the OpenChatCut repo shows 1,061 stars, 140 forks, and 26 open issues on GitHub, licensed AGPL-3.0, with commits landing as recently as yesterday. The repository was created July 15, 2026, so this traction has built up in under a month, and the star count has kept climbing well past where it stood at scan time. That's a fast trajectory for a project this new, though it's still early enough that the usual rules apply: check the open issues list and recent commit activity before depending on it for anything time-sensitive.

Bottom line

OpenChatCut is worth adding to an AI stack when the job needs a real, local, multi-track editor that an agent can drive without a person losing the ability to see and approve each change before it lands. It trades FableCut's minimalism for a heavier, more complete editor with a genuine review workflow built in. Anyone weighing the two should look at the actual editing task first: a flat cut-grade-caption pipeline fits FableCut's document model fine, while footage that needs real timeline work, transcript-based editing, or a formal approval step before changes commit is where OpenChatCut earns the extra setup.