jub0t/Concat

Concat is worth a look if you edit short-form video on your own machine and dislike uploading footage or paying a subscription. Treat it as a beta desktop editor: the download is straightforward, but rough edges are expected and the documentation says so.

The problem

Someone who wants a quick social clip today usually opens CapCut: import footage, add auto-captions, apply effects, export. That flow depends on logging in and sending media to a service, and the free tier adds a watermark or prompts for payment. Concat replaces that round trip with a local application: install it, import media, add captions or transitions, and export. The caption, voice and background-removal models download once from Settings, after which the app claims to work offline.

What's different

The distinguishing mechanism is that editing and compositing run on a native Rust engine with a GPU compositor, and the AI features are local models rather than API calls. In practice, that means captions from Whisper and TTS/voice cloning run on your own hardware, which matters when the material is private or when you work without reliable internet. The stated benefit is also the main trade-off: local inference pushes CPU and GPU load onto your machine, and model quality depends on which model size you download.

The second differentiator is scriptability. The project exposes a JSON-RPC API and MCP, so a batch of edits can be driven programmatically by another tool rather than assembled by hand. That is unusual for an open-source editor at this stage and makes it interesting for automated pipelines, not just manual cutting. The README also lists 100 effects, filters and transitions, keyframes with a curve editor, multi-track cutting, background removal and 4K export. These are author claims; independent testing is not available here.

Who should try it

A good fit is a creator who already knows CapCut and wants the same rough workflow offline: auto-captions, text and titles, keyframed movement, a 4K export, no watermark. Also a reasonable fit is a developer automating short-video production who wants a real editor behind a JSON-RPC or MCP interface instead of stitching FFmpeg commands together.

A poor fit is anyone who needs a finished, stable tool for client deadlines or uncommon codecs. The project's own download page calls it beta and admits rough edges, so it is a candid try-it-and-report-bugs candidate rather than a drop-in replacement you schedule work around.

Requirements and limitations

Platform support is documented for macOS, Windows, Linux and Android, but Windows ARM builds are paused (issue #326), so ARM Windows users are out for now. Downloads come from the project site or GitHub Releases; the Windows builds are unsigned, and the README explains the SmartScreen bypass step. Building from source is heavier: Rust 1.93+, FFmpeg 7+ development libraries, cmake and a C++ compiler, with FFmpeg version matters differing across Linux distributions.

Ongoing upkeep has two parts. First, model downloads for captions, voice and cutout take disk space and are fetched separately from the editor, so the first useful session includes that setup. Second, local processing leans on the GPU; how well 4K and background removal perform on a given machine is not established by the README. Licensing is AGPL-3.0, which is relevant if you plan to redistribute a modified build or embed it in a service.

HotRepo's take

HotRepo's dated signals show #8 on the monthly Rust board and #4 on the weekly Rust board on 2026-10-26, but no daily-board history, so whether interest persists cannot be judged from this sample. The project itself is 61 days old at that snapshot. A comparison raised on Hacker News was OpenCut, and one commenter's negative verdict on that project's feature completeness does not establish anything about Concat's. Before installing, check the release notes and issues for the codecs and platform you rely on, since the beta status is the key risk. The editor screen is shown on the project homepage.