OpenVids
bazodev (open source) · OpenVids: Open-Source Agent-First Video Editor for macOS and Windows
OpenVids is a free, Apache-2.0 desktop video editor for macOS and Windows in which you describe the video in chat and a Director agent and its specialists cut your footage, build the timeline, add captions, motion graphics and sound, then render and check the result. It suits creators and editors who want agentic editing of their own footage on a real timeline, with media kept on their disk and their own model provider keys.
Best for
Solo creators, YouTubers and editors on Apple Silicon Macs or Windows PCs who want AI agents to do rough cuts, captions, motion graphics and sound passes on their own footage, keep media on their disk, and are comfortable bringing their own model provider key and installing early-stage software
Not ideal for
Teams that need a stable, supported or signed release, Linux, Intel Mac or browser users, people who want a hosted tool with no API keys or extra installs, and anyone who mainly wants to generate video from scratch rather than edit footage
Who it's for
Creators, YouTubers and video editors on Apple Silicon Macs or Windows who want agentic editing of their own footage with local media and their own model provider
Capabilities
- Chat-driven editing where a Director agent plans each request and delegates to Editor, Vision, Motion Designer, Research and Audio specialists that can be switched on or off
- Agent edits are written to the project files, so the timeline and preview update live and every change can be reviewed or taken over by hand
- Local long-form analysis: transcript with word timestamps (whisper.cpp or Parakeet), speaker detection, pauses, shots, black or frozen picture, retakes, false starts and fillers, reused across turns to plan and build rough cuts
- Story mode: plan the video as a graph of chapters with A-roll, B-roll, music, voiceover, sound effects and motion, then Build or rebuild only the affected parts of the timeline
- Motion Designer builds titles, lower thirds, logo stings and caption styles from a sentence as live templates you can scrub, tweak and reuse
- Render QA after each agent turn checks the rendered file for black or frozen picture, audio holes and silence, gaps, cuts inside a word and caption or layout collisions, with Vision reviewing frames for wrong B-roll; agents fix issues for up to 5 passes (2 by default)
- Research specialist searches built-in open sources (Wikimedia Commons, Openverse, Internet Archive, NASA and several museum collections) plus Pexels, Pixabay, Flickr, Freesound and Smithsonian with your own free keys, recording origin, author and license for every import
- Voiceover from your own text-to-speech key (Gemini, OpenAI, OpenRouter, ElevenLabs or your own server), with music ducking and captions generated from the voice
- One prompt equals one checkpoint: a whole agent turn, including QA fixes and imported assets, can be reverted at once
- Per-project agent cost view broken down by agent, model or chat, and design systems (colors, fonts, logo, motion) that can be saved and applied to other projects
Limitations
- Early-stage, pre-1.0 software: the repository was created on September 28, 2026, and the security policy says only the latest commit on main receives security fixes
- Desktop only, for macOS 11+ on Apple Silicon and Windows 10+ on x64; there is no Intel Mac build and no web, mobile or Linux version
- macOS builds are ad-hoc signed and not notarized, and Windows builds are unsigned, so the first launch needs a Gatekeeper or SmartScreen override
- Rendering and thumbnails need Google Chrome and FFmpeg installed on the machine; OpenVids does not ship them (Windows offers an in-app FFmpeg download)
- Agents need a model provider you connect, which bills you directly and receives your prompts, transcripts and sampled video frames
- While a project is open, its local Studio server on loopback is unauthenticated, so any local process can reach it (web pages are refused); the security policy lists this as a known limitation
- No team collaboration, shared cloud projects or vendor support are documented; help is through GitHub and Discord
Use cases
- Turning a long interview or talking-head recording into a rough cut that keeps the strongest answers, removes retakes and fillers, and stays under a target length
- Laying out a short documentary or product video as chapters in Story mode and building a first cut from your own clips
- Adding a caption pass, lower thirds and a title sting to a finished cut with one prompt each
- Finding openly licensed B-roll for a section and keeping a credit list of every imported asset
- Adding narrated voiceover from your own text-to-speech key with the music ducked under it
Our take
OpenVids is one of the few agentic video tools where the agents work inside a real, local timeline editor rather than producing a finished file in a cloud app, and its design choices back that up: plain-file projects, per-prompt checkpoints, local transcript analysis for long recordings, license tracking for imported footage, and a Render QA loop that checks its own output. The README is also unusually clear about what leaves your machine. The catch is maturity. It is less than two weeks old, pre-1.0, unsigned, Apple Silicon and Windows only, and depends on separately installed Chrome and FFmpeg plus a model provider you pay for. It is worth trying on a non-critical interview or explainer edit today; teams should wait for signed, stable releases before relying on it.
Who should use it
Creators and editors on Apple Silicon Macs or Windows who cut interviews, talking-head videos or short documentaries, want agents to handle rough cuts, captions, motion and sound passes on their own footage, and are happy to connect their own model provider and use early-stage software.
Who should skip it
Production teams that need signed, supported and stable software, Linux or Intel Mac users, people who want a hosted editor with no extra installs or API keys, and anyone looking mainly for text-to-video generation.
Strengths
- Free and Apache-2.0 licensed, with no account and no cloud editing backend
- Agents work on a real timeline you can review, revert per prompt or take over by hand
- Long recordings are transcribed and analysed locally, and the analysis is reused across turns
- Render QA renders and checks each agent turn and fixes what it can before you look
- Research keeps the source, author and license of every imported asset
Weaknesses
- Very new and pre-1.0, with frequent releases and a Windows agent-chat regression fixed only in v0.5.4
- Unsigned and non-notarized builds, and Chrome plus FFmpeg must be installed separately
- No Linux, Intel Mac, web or mobile version
- Model usage is billed by the provider you connect, and footage frames and transcripts are sent to it
- Published from one developer's personal GitHub account, with community support through GitHub and Discord only
OpenVids pricing
Open source
Free
- Apache-2.0 license
- macOS (Apple Silicon) and Windows (x64) installers on GitHub
- No OpenVids account
- Community help on GitHub and Discord
Note: OpenVids has no paid plan. The model provider you connect for the agents, and any text-to-speech provider used for voiceover, bill you directly at their own rates. Stock sources such as Pexels, Pixabay, Flickr and Freesound need your own free API keys.
Technical specs
Modalities
Video, Audio, Text
Available models
Where OpenVids excels
Interview and talking-head rough cuts
Local transcription with word timestamps, speakers, pauses and retake detection lets the agents assemble a cut around the strongest answers, and every turn can be reverted or refined by hand on the timeline.
Short documentary or explainer with archival B-roll
Story mode keeps the film as chapters, Research imports openly licensed footage with its credits recorded, and Render QA catches gaps, black frames and caption collisions before export.
Branded social or product videos
Motion Designer builds reusable titles, lower thirds and caption styles from a sentence, and design systems carry the same colors, fonts and logo across projects.
OpenVids vs. competitors
OpenVids vs. invideo
invideo is a hosted, browser-based editor where AI agents log footage, build cuts and generate missing shots from more than 200 models, with agent work paid in credits. OpenVids is a free, open-source desktop app that runs locally with your own model provider and focuses on editing your footage rather than generating new shots. invideo is the more complete, supported service; OpenVids keeps media on your disk and costs nothing beyond your model usage.
OpenVids vs. Descript
Descript edits video and podcasts through the transcript and adds AI cleanup tools and its own co-editor in a commercial, account-based product. OpenVids uses a multitrack timeline with several specialist agents, runs locally and is open source. Descript suits podcasters and teams that want a mature, document-style editor; OpenVids suits users who want open, local agentic editing.
OpenVids vs. Eddie AI
Eddie AI is a paid assistant editor that logs footage and builds rough cuts, then hands them to Premiere Pro, DaVinci Resolve, Final Cut Pro or Avid for finishing. OpenVids is a free editor in its own right, where agents and you finish the cut in the same timeline, with captions, motion graphics, sound and render QA included. Eddie AI fits professional post workflows built around an existing NLE; OpenVids fits creators who want one local, open-source app.
OpenVids vs. OpenMontage
OpenMontage turns an AI coding assistant into a video production pipeline, with many formats, generated media and zero-key stock footage, but no timeline editor. OpenVids is a desktop editor with a timeline, preview and inspector where agents edit footage you can then adjust by hand. OpenMontage fits developers producing explainers and montages from a coding agent; OpenVids fits people who want to edit their own recordings in an app.
Frequently asked questions
What is OpenVids?
OpenVids is an open-source desktop video editor for macOS and Windows that you work in together with AI agents. You describe the video in chat, and a Director agent with Editor, Vision, Motion Designer, Research and Audio specialists cuts footage, builds the timeline, adds captions, motion graphics and sound, then renders the result and checks it. You can review every change and keep editing by hand.
Is OpenVids free and open source?
Yes. OpenVids is free and released under the Apache-2.0 license, with code and installers on GitHub. The agents use a model provider you connect with your own API key or sign-in, and that provider bills you for usage; voiceover likewise uses your own text-to-speech key. OpenVids shows what a project's agents have cost when the model has a known price.
Which platforms does OpenVids run on?
macOS 11 or later on Apple Silicon and Windows 10 or later on x64. There is no Intel Mac build and no Linux, web or mobile version. macOS builds are not notarized and Windows builds are unsigned, so the first launch needs an Open Anyway or Run anyway step, and rendering requires Google Chrome and FFmpeg on the machine.
Does OpenVids upload my footage?
Editing, analysis and rendering run on your machine and your media stays on your disk. The agents send prompts, transcripts and sampled video frames to the model provider you connect, Research downloads from the sources your Asset Search policy allows, and OpenVids itself receives only anonymous usage statistics, which you can turn off, and bug reports you choose to send.
Which AI models does OpenVids use?
It does not bundle a model. You connect a provider in Settings by adding an API key, which is stored locally, and choose which model each agent uses. The agent runtime is built on the oh-my-pi agent library, so providers already set up there are picked up too, and some providers can be signed in to from the app. The manual editor works without any model.
OpenVids vs Descript: what is the difference?
Descript is a commercial, freemium editor that centres on editing through the transcript, with its own AI co-editor and cloud accounts. OpenVids is a free, open-source desktop app where several agents edit a conventional multitrack timeline, media stays local and you bring your own model provider. Descript is the more mature, supported product; OpenVids suits users who want open-source, local agentic editing and accept early-stage software.
OpenVids vs OpenMontage: what is the difference?
Both are free, open-source, local video tools that use AI agents. OpenMontage is a production system you run inside an AI coding assistant such as Claude Code or Codex, with pipelines for generated and stock-based videos and no editor of its own. OpenVids is a standalone desktop editor with a timeline, preview and inspector, built around editing your own footage with chat-driven agents.
Integrations & fit
Alternatives to consider
invideo
Creators, agencies, and in-house marketing and film teams who edit in the browser and want AI agents to handle logging, first cuts, multicam, sound design, effects, and missing shots on a timeline they can take over at any frame
Freemium
Descript
Creators, podcasters, and content or marketing teams who want fast, text-based video and audio editing with AI cleanup
Freemium
Eddie AI
Professional video editors, producers and post-production teams who cut interview-driven, documentary, podcast or corporate video and want AI to handle logging, syncing and the first assembly before they finish in Premiere, Resolve, Final Cut or Avid
Paid
Calesthio (open source)
Developers, technical creators, educators, and small teams who already use an AI coding assistant, are comfortable installing Python and Node.js tooling, and want an inspectable, local video pipeline with approval gates and their own choice of media providers
FreeAbout OpenVids
OpenVids is a local desktop editor rather than a hosted service: there is no OpenVids account or cloud editing backend, and a project is a folder of plain files you can open and edit by hand. You chat with a Director agent that plans the work and hands it to specialists you can switch on or off: Editor (timeline edits), Vision (looks at selected frames), Motion Designer (titles, lower thirds, logo stings and caption styles built as live templates), Audio (levels, fades, music and voiceover) and Research (finds outside footage and records its license). The official website presents a simpler picture of four agents (Main, Motion Designer, Audio and Research); the README and release notes document the Director with five specialists described here. Edits land in the project files, so the timeline and preview update while the agents work, and the full manual editor (timeline, preview, inspector and code view) works without any agent. For long recordings it analyses media locally (transcript with word timestamps via whisper.cpp or Parakeet, speakers, pauses, shots, retakes and false starts) and reuses that analysis to build rough cuts. Story mode lays the film out as a graph of chapters that Build turns into a cut while keeping edits you have locked. After each agent turn, Render QA renders the project and checks for black or frozen picture, audio holes, gaps, cuts inside a word and caption or layout collisions, and the agents correct what they can, for up to five passes (two by default). Every prompt is one checkpoint you can revert as a whole. The app is free, but agents call the model provider you connect (prompts, transcripts and sampled frames go to that provider, which bills you), and voiceover uses your own text-to-speech key. The tradeoffs are maturity and setup: the repository was created on September 28, 2026, the project describes itself as early-stage and pre-1.0, macOS builds are Apple Silicon only and not notarized, Windows builds are unsigned, and rendering needs Chrome and FFmpeg installed separately. OpenVids began as a snapshot of HeyGen's open-source HyperFrames and is not affiliated with HeyGen.
Updates from OpenVids
OpenVids 0.5.4 restored the agent chat on Windows, which failed to start in 0.5.2 and 0.5.3 because the installer left out files the agent engine needs. macOS builds were not affected.
Version 0.5.0 added a view of what each project's agents have cost by agent, model or chat, project forking, a render queue, a pinned plan for running turns and an opt-in beta update channel.
Version 0.3.0 brought OpenVids to Windows 10 and 11 (x64) with its own installer and an in-app FFmpeg download, alongside plan approval in Agent Chat and @-mentions of project media.
OpenVids 0.1.0 was released for Apple Silicon Macs as an open-source desktop video editor with a Director agent and specialists, local long-form analysis, Story mode, research with licenses, Render QA and a full manual editor.
Are you the founder? Claim this listing →