HeyGen ships HyperFrames Studio for Mac and Linux

HeyGen's open-source video team put a desktop editor on Product Hunt on October 5, 2026. The listing is HyperFrames Studio (Desktop). The tagline is "The First Video Editor built for Agents." Product Hunt marks it free. Tags on the launch are Open Source, Artificial Intelligence, and GitHub.

The same day, the hyperframes repo cut v0.8.134 and added Studio app pages for Mac and Linux. The README was updated on October 5 to put Studio first and to show a native Mac and Linux app. On October 6 the repo shipped v0.8.136, a preview fix for compositions with large inline background images. This is not a waitlist. The docs site already has a Studio app section, and the product site links a Mac build.

I care about this one because the video is a folder of files, not a timeline trapped in a vendor format.

What shipped

HyperFrames is the open-source framework from the HeyGen team. Compositions are HTML. Timing lives on data attributes such as data-start, data-duration, and data-track-index. Animation can come from GSAP, CSS, Lottie, Three.js, Anime.js, WAAPI, or a custom frame adapter. The renderer seeks the page and encodes frames. Same input, same frames. The license is Apache 2.0. The project says there are no per-render fees and no commercial-use thresholds on the open-source engine.

The CLI, core, catalog, and agent skills were already public. Studio existed as a visual editor package. What landed on October 5 is the desktop app that wraps that loop for someone who does not want to live in a terminal. Docs call it the easiest way to make a video with HyperFrames. You describe the video. An agent on your computer builds it. You change it by typing, pointing at the picture, drawing, or commenting. Framey, a small green guide in the app, points at the next step.

Platforms in the docs are macOS and Linux. The Linux build named in the docs is HyperFrames-x86_64.AppImage, for 64-bit Intel and AMD machines. I did not see a Windows build in the Studio app docs. The open-source CLI is the broader surface if you already render from a repo.

Sign-in is a HeyGen account. On first launch the app asks you to continue with HeyGen. The browser opens hyperframes.dev. You have two minutes before the sign-in window expires. You can skip the optional survey. Settings holds account sign-in and sign-out.

The agent is not theirs

This is the part that changes the bill. Studio does not ship its own model. The agent runs on your computer. Docs list two: Claude Code and Codex.

Claude Code needs a Claude account or an Anthropic Console account. The app can set that agent up. Codex needs a ChatGPT account. Docs say the app does not set Codex up for you. You confirm it from Settings, then the row shows Ready.

Runs bill to the plan of the account the agent is signed in with. If you want an API key instead, Settings has a "Bill runs to" control. If both agents are ready and you have not picked one, the app uses Claude Code. The picker in the chat header switches the next request. States on each row are Ready, Not signed in, Not installed, or Installed but won't start.

That pricing shape matters. There is no HyperFrames Studio seat price on the Product Hunt listing or the Studio docs I read. The open-source render path does not meter frames. The variable cost is the coding agent. A founder on Claude Pro, Claude Max, ChatGPT Plus, or an API key pays whatever that plan already charges. A founder with none of those accounts cannot finish a generated video in the app until one is connected. Local preview of an HTML composition is a different path and does not need the agent.

HeyGen-hosted cloud rendering and a hosted connector exist in the broader HyperFrames project and are described elsewhere as beta. Credit terms for that hosted path are not on the Studio app pages. I am not treating hosted render as part of this desktop launch.

The first session

Home opens with a prompt box, projects on the left, and a Get inspired block. The first-run card is "Create your first video." It builds a launch video from four parts taken from real launch videos.

Hook grabs the first seconds. Proof shows the product doing the job. Turn flips from problem to answer. Call to action lands the brand and the ask. Each part starts with one clip. A shuffle button swaps the set. Hovering a part shuffles that part or opens the others.

You add a website or an idea. A website gives the agent fonts, colors, and logo. Then you choose "Use this cut." The request goes to the agent. If you are not signed in, or no agent is connected, the app walks through that first and then sends the request.

The app creates a project. Chat shows what the agent is doing, such as Reading files or Editing files. On a first video, a short tour replays how a launch video was made while yours builds. Framey then takes you to the project. When the agent finishes, the video is ready to play, change, and export.

You can skip the card. Back on home, the prompt box takes a plain description. Typing @ attaches files, other projects, or media. "Take the step by step" next to Get inspired reopens the card.

A launch video has a shape. An agent told "make a video" often invents a new shape every run. Four named parts plus a site for brand tokens is a narrower brief, which is how you get a draft you can edit.

Why the file format is the product

Most AI video tools hand you a clip. You can regenerate it. You cannot diff it. HyperFrames keeps the project as HTML and media. The agent writes files. Studio edits the same files. A later agent run can keep going. The README tells agents where Studio edits go.

That is the indie use. A changelog video, a Product Hunt cut, a feature walkthrough, a per-customer card from one template. The template is a page. Personalization is data. Render is deterministic, so a regression is a frame mismatch, not a vibe check.

Skills sit under the app. HyperFrames ships 21 of them. The core set is a router plus domain skills. The README sample is a 10-second product intro with a fade-in title, a background video, and music. Skills load in Claude Code, Codex, Cursor, Gemini CLI, and other agents. The desktop app is the visual seat on that loop, not a replacement.

Limits I would not gloss over. Linux in the Studio docs is the x86_64 AppImage, not an ARM build. Windows is absent from those pages. Agent quality is capped by Claude Code or Codex on your machine, plus the skills they loaded. A website handoff only helps if the site actually exposes the fonts and colors the agent can read. Large inline background images were still blowing preview memory as of the October 6 patch, which is a reminder that this app is moving daily. v0.8.135, also October 5, fixed easing, undo, layer icons, and nested scenes. You are joining a repo that shipped several Studio fixes the same week it put the app on Product Hunt.

What I would try first

I would not start with a brand film. I would start with one 10 to 20 second launch cut for a feature I already shipped. Use the four-part card. Point it at the marketing site so the colors are not invented. Let the agent write the composition. Then change one line of type and one timing mark in the editor, by pointing or by comment, and see if the project files stay coherent.

If the agent bill is the constraint, stay on the open-source CLI and skills and skip the desktop sign-in. The engine does not charge per frame. If the constraint is "I do not want to babysit HTML," the October 5 app is the new surface, and the meter is whichever coding plan you already pay.

Sources: Product Hunt launch page for HyperFrames Studio (Desktop) on the HeyGen product, Studio app docs at hyperframes.heygen.com/studio-app, and the heygen-com/hyperframes releases for October 5 and October 6, 2026.