OpenChatCut

AI video tools mostly force a choice between two bad options: a traditional editor where you do everything by hand, clip by clip, or a one-shot AI generator that hands you a finished video you cannot actually edit, only regenerate from scratch with a different prompt. OpenChatCut tries to refuse that choice. It puts a conversational AI agent inside a real multitrack video editor, so the agent works the same tracks, clips, transitions and captions a human editor would, instead of producing an opaque rendered file that stops being a project the moment it exists.

Under the hood it is a desktop app (Electron, React 19, TypeScript, Vite) built on Remotion and FFmpeg for rendering, with word-level transcript-driven editing, linked captions, and AI generation for images, video, speech, music and sound effects pulled in through the Vercel AI SDK across multiple providers. Motion graphics templates and WebGL effects sit alongside the timeline, and an MCP server exposes the whole project to external coding agents like Codex or Claude Code, so an agent can drive edits the same way the built-in conversational one does. Exports cover MP4, audio, captions and FCPXML for handoff to other tools.

It fits creators and small teams who want AI to speed up editing - transcript cleanup, rough cuts, caption generation - without losing the ability to go in and manually fix a specific frame, transition or line of dialogue afterward. Because projects and media stay local by default and desktop installers exist for macOS, Windows and Linux, it also suits people who want a self-hosted alternative to cloud-only AI video editors, and the MCP support makes it a natural fit if you already automate other parts of your workflow through coding agents.

AGPL-3.0 is the important caveat here: it is a strong copyleft license, so if you build a hosted service on top of OpenChatCut and offer it to users over a network, you are generally required to release your modifications as source too - worth checking with a lawyer before building a commercial product on it. The AI generation features are not self-contained either; image, speech, music and video generation all call out to external provider APIs (OpenAI, Anthropic, Gemini, Fal.ai and others), so you need your own keys and accept those providers' costs and terms. No independent or even self-reported performance benchmarks are published; its advantages are architectural claims, not measured ones.

The verdict: OpenChatCut is a compelling pick if the core problem you have with AI video tools is that they throw away editability, and you want an agent-assisted editor where every AI-driven change lands as a real, inspectable timeline edit rather than a black-box render. It is under active development with a steady release and installer cadence, so treat it as usable today rather than experimental. Just budget for external AI API costs since nothing is bundled, and read the AGPL-3.0 terms carefully before wrapping it in a commercial, hosted offering.