Concat: a Rust, offline CapCut replacement that writes H.264 MP4 only
Open-Source CapCut replacement.
At a glance
- What is it?
- Concat is an AGPL-3.0 desktop video editor built in Rust with Slint, pitched as a CapCut replacement that runs captions, speech and cutout models locally. It is a 0.2.x beta, and the export path is narrower than the comparison table suggests.
- Who is it for?
- Adopt Concat if you need offline editing with local captions and text-to-speech, and H.264 MP4 is the only deliverable you owe anyone. Skip it if you need ProRes, WebM or a curve editor, or if you depend on CapCut's cloud AI features.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository last received commits 2 days ago.
- What is it written in?
- Mainly Rust, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 16, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What Concat is aimed at, and who it is not for
The README frames Concat as "everything you use CapCut for" without watermarks, paywalls or subscriptions, and the repository topics list it as a capcut-alternative, a non-linear-editor and a free-video-editor. The intended user is someone cutting short-form video on their own hardware who does not want an account, an upload step or a monthly fee. That is a real gap: CapCut's terms, as summarised in the project's own comparison table, grant ByteDance a perpetual licence to uploads, and its best features are cloud-side.
The audience is narrower than the tagline. Concat is a 0.2.x beta, and the comparison table in the README marks its own effects library, AI tools, keyframes, export formats and stability as partial. Anyone whose work depends on CapCut's tracking, reframe or avatar tools will not find equivalents here. The project is also not a library: despite the name, this is not a string-concatenation utility, and the repository has no published package for programmatic use. There is no homepage; distribution is through GitHub Releases.
The Rust and Slint engine, and where the models live
The core is a native Rust engine with a Slint interface, which is why the README can claim Concat and CapCut are both native while OpenCut runs WebAssembly FFmpeg in a browser. Editing operations the README lists are split, trim, merge, transitions and speed control, across multi-track timelines, with several timelines per project.
The offline features are the interesting architectural choice. Auto-captions run through whisper.cpp locally, with model sizes from 78 MB to 488 MB depending on the whisper size chosen. Text-to-speech adds 132 MB or 349 MB of voices, person cutout 15 MB, object cutout 179 MB, and the cutout brush 40 MB. The README states these download from the settings panel on first use and then never need the network again. That is a distinct data flow from a cloud editor: the model weights are the only network transfer, and the media never moves. The repository carries a top-level models/ directory, and ffmpeg appears in the topic list as the underlying media layer.
Hardware sets the ceiling, since nothing is offloaded. The minimum is any 64-bit CPU from 2013 or later, 4 GB of RAM and 500 MB of storage; the recommended column is 6 cores, 16 GB for 4K timelines and the larger caption models, and a GPU with Metal, DirectX 12 or Vulkan. Without a usable GPU the README says the window and monitor fall back to the CPU, which will be felt on a 4K timeline.
Installing Concat and cutting a first clip
Concat is in beta, and the README points to the Releases page for the latest build rather than to a package manager. There is no Homebrew formula, no winget manifest and no apt repository documented, so download the archive for your platform. The Windows and Linux builds are plain archives.
To keep settings, recents and downloaded models inside the folder you unpacked, create a directory named portable beside the concat executable. The README states nothing is then written to the user profile, which is what makes a USB stick viable.
mkdir -p portable
./concatOn macOS the binaries are unsigned, so the README gives this command before the first launch:
xattr -dr com.apple.quarantine /Applications/Concat.appOnce the window opens, the workflow the README describes is a multi-track timeline: import media, place clips, then split, trim, merge, add transitions and adjust speed. Auto-captions are not bundled. They download from the settings panel on first use, so expect a one-time transfer of 78 MB to 488 MB depending on the whisper model size, after which the README says no network is needed. Export writes H.264 MP4. If a run misbehaves, every run writes a log, the last ten are kept, and Settings > About has a button to open the log and another to copy your system information; the README asks for both in an issue.
Where Concat falls short of CapCut
Export is the hard limit. The comparison table states plainly that Concat writes H.264 MP4 only, while OpenCut writes MP4 and WebM. If your delivery pipeline expects ProRes, DNxHR or a mezzanine format, Concat is the wrong tool today, and there is no documented workaround in the README.
Keyframes are the second gap, and it is a subtle one. Concat keys position, scale, rotation and opacity with bezier easing, but the README says there is no curve editor and no keyed effect parameters yet. That means animation of effect settings, the thing that makes CapCut's transitions feel authored, is not available. The effects and templates library is described as a few dozen against CapCut's thousands, and the AI tools are limited to local captions, speech and person cutout.
Stability is honestly labelled: the table calls Concat a 0.2.x beta and marks it partial. Mobile is worse than partial. The README lists Android and iOS builds as compiling but untested, and the Linux x86_64 and ARM builds as to-be-tested. Treat the desktop builds as the only supported surface. There is also no plugin API yet; the README says it is planned, and the comparison table credits OpenCut with an Editor API, MCP server and plugins that Concat does not have.
How Concat differs from OpenCut and CapCut
OpenCut is the closest open-source alternative, and the difference is architectural rather than cosmetic. OpenCut runs WebAssembly FFmpeg in a browser, which the README's own table marks as weaker on performance and dependent on the browser for 4K export. Concat compiles to a native binary and runs its models on device. The trade runs the other way on extensibility: OpenCut ships an Editor API, an MCP server and plugins, and Concat's plugin API is only planned. A team that wants to script the editor should look at OpenCut; a team that wants a desktop binary with local captions should look here.
CapCut is the closed-source comparison, and the README is direct about the terms: CapCut's terms grant ByteDance a perpetual licence to uploads, its free tier caps at 1080p, and Pro effects, 4K and AI tools cost $9.99 to $19.99 a month. Concat sends nothing anywhere and has no account. What you give up is the library of thousands of effects and templates and the cloud AI features. The licences differ too: Concat is AGPL-3.0-or-later, OpenCut is MIT, CapCut is closed. If you plan to embed an editor in a networked product, the AGPL and the MIT choice are not interchangeable.
Licence, maintenance and what an upgrade costs you
Concat is AGPL-3.0-or-later, and the repository carries LICENSE, LICENSE-EXCEPTIONS.md, THIRD_PARTY_NOTICES.md and TRADEMARK.md as separate files. The presence of an exceptions file and a trademark file means the AGPL grant is not the whole story; read both before shipping anything that embeds Concat or uses its name. This is not legal advice, and the exceptions file is the document to take to whoever gives you that advice.
The project is not archived, and the last push was on 2026-09-16, the same day as the v0.2.2 release. Releases are close together: v0.2.1 on 2026-09-04, a nightly build of 0.2.1 on 2026-09-04, and v0.2.2 on 2026-09-15. That cadence means the upgrade cost is real. A 0.2.x line changes behaviour between point releases, and the README offers no migration notes or project file compatibility statement, so a project file saved in one build may not open cleanly in the next. Budget for keeping the version you validated, and keep the archive you downloaded. The portable folder layout helps here: pin the executable and its models in one directory rather than letting an installer replace them.
Editorial conclusion
Adopt Concat if you need offline editing with local captions and text-to-speech, and H.264 MP4 is the only deliverable you owe anyone. Skip it if you need ProRes, WebM or a curve editor, or if you depend on CapCut's cloud AI features. Before committing, download the v0.2.2 build from the Releases page, run one 4K timeline on your own machine, and read LICENSE-EXCEPTIONS.md and TRADEMARK.md to see what the AGPL grant does not cover.
Frequently asked questions
What is Concat used for?
Concat is a cross-platform desktop video editor positioned as a free, open-source CapCut replacement. The README lists multi-track editing, split, trim, merge, transitions, speed control, auto-captions, text-to-speech, voice filters, titles, templates and person or object cutout, all run locally.
How do I install Concat?
The README points to the Releases page for the latest build, since Concat is in beta and no package manager is documented. Windows and Linux ship as plain archives, and macOS binaries are unsigned, so the README gives the xattr -dr com.apple.quarantine command for /Applications/Concat.app before the first launch.
What are the system requirements for Concat?
The README lists a minimum of any 64-bit processor from 2013 or later, 4 GB of RAM and 500 MB of storage, with no GPU required. The recommended configuration is 6 cores, 16 GB of RAM for 4K timelines and larger caption models, and a GPU with Metal, DirectX 12 or Vulkan.
Does Concat need an internet connection?
Only for the first download of the optional models. The README states that auto-captions, text-to-speech and the cutout models download from the settings panel on first use and then never need the network again, and that nothing is sent anywhere on its own.
What export formats does Concat support?
The README's comparison table states that Concat writes H.264 MP4 only, while OpenCut writes MP4 and WebM. There is no documented option for ProRes, DNxHR or other delivery formats.
Community notes