Live Nov 18 · Cohort 3Nov 18, 19:00 MSK — free hands-on AI agents live workshop · Live Cohort 3 starts Nov 23Join Nov 18 Live →Cohort 3 (Nov 23)
ReleaseCode / SDK · ActionableX: @RoundtableSpace→ original

11 open-source repos chained together for Claude Code, from research to voiceover

11 open-source repositories cover the entire content pipeline: research, text, slides, video and voiceover. You can hand all of it to Claude Code. last30days-skill gathers material from Reddit, YouTube and HN, frontend-slides builds slides from a prompt, and HyperFrames renders video from HTML. Two projects are new: Phonon and ZONOS3-Mini, both for speech synthesis. Check the licenses before a commercial launch, because code and weights often carry different ones.

AI-processed from X: @RoundtableSpace; edited by Hamidun News
11 open-source repos chained together for Claude Code, from research to voiceover
Source: X: @RoundtableSpace. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceX: @RoundtableSpace ↗
Editor-in-Chief
Key metric #1ZONOS3-Mini clones voices in 43 languages, with weights under Apache 2.0
Phononthe code is Apache 2.0, but the multilingual weights are non-commercial only
Key metric #3OpenMontage is released under AGPL-3.0, a serious condition for a product

A content pipeline that once required an analyst, an editor, a designer, a video editor and a voice actor is increasingly built around a single coding agent. @RoundtableSpace, who compiled the list, posted 11 open-source repositories that form a "research → text → slides → video → voice" chain and can be handed to Claude Code. Two projects on the list are brand new. Here is each one in order.

Research and text

last30days-skill (MIT) is an agent skill that researches a topic across Reddit, YouTube, Hacker News and the web at once. It is a quick way to see what people are saying about a topic, not just what search results show.

claude-video (MIT) lets Claude "watch" any video and answer questions about it. It is handy when you need the gist of an hour-long talk or a competitor's demo.

humanizer (MIT) scrubs the telltale signs of AI-written text from drafts. It serves as a final polish before publication: formulaic phrasing gives away machine authorship faster than anything else.

Slides and video

frontend-slides (MIT) creates animated HTML slides from a prompt or a PowerPoint file. The presentation ends up as an ordinary web page.

ppt-master (MIT) takes a different route: it turns documents into editable PowerPoint decks and can generate voiceover from speaker notes. If the client expects a .pptx that can be tweaked by hand, this is the more pragmatic option.

HyperFrames (Apache 2.0, from HeyGen) works on a "write HTML, get video" principle. For a developer, the model is easy to grasp: markup replaces the editor's timeline, so a clip can be generated by code and stored in git.

video-use (MIT, from the browser-use team) edits source footage in a dialogue with the agent. It needs an ElevenLabs key, so the whole process won't be free.

OpenMontage (AGPL-3.0) is the most ambitious entry: a coding agent runs the entire video production. But the AGPL license demands extra care. If you embed such code in a network service, you may end up obliged to open your own code. A lawyer should look at it before launch.

Voice: two new releases

Pocket TTS (code under MIT, from Kyutai) voices text directly on a CPU, with no GPU. For inexpensive servers and local prototypes, that is a major advantage.

Phonon (new, released October 20) is on-device speech synthesis. The code is under Apache 2.0, but the multilingual weights are non-commercial. That makes it a trap: you can take the code into a product, but not the model for several languages.

ZONOS3-Mini (new, from Zyphra) does speech synthesis with voice cloning in 43 languages, with weights under Apache 2.0. For commercial use, its terms are the cleanest of the voice projects on the list.

What to do with this

The main practical value lies not in the individual tools but in how they fit together: an agent can go all the way from research to a voiced video if you give it the right skills and repositories. For a founder or product manager, it is a way to build a demo, training material or a marketing video without a studio.

There are limitations, too. The author of the list warns that you should check each project's license before commercial use, and clone a voice only with the speaker's consent. The licenses on the list vary, from MIT and Apache 2.0 to AGPL-3.0 and non-commercial weights, and that determines whether you can ship the output in a product. The list also includes no quality measurements, so you will have to find out for yourself how much effort manual rework of slides or editing will take. A sensible start is to pick two or three projects under MIT or Apache 2.0 and assemble a minimal pipeline from them, and only then decide whether you need heavier options like OpenMontage.

Story

Key Questions Answered

What are the key details on Research and text?

last30days-skill (MIT) is an agent skill that researches a topic across Reddit, YouTube, Hacker News and the web at once. It is a quick way to see what people are saying about a topic, not just what search results show. claude-video (MIT) lets Claude "watch" any video and answer questions about it. It is handy when you need the gist of an hour-long talk or a competitor's demo. humanizer (MIT) scrubs the telltale signs of AI-written text from drafts.

What are the key details on Slides and video?

frontend-slides (MIT) creates animated HTML slides from a prompt or a PowerPoint file. The presentation ends up as an ordinary web page. ppt-master (MIT) takes a different route: it turns documents into editable PowerPoint decks and can generate voiceover from speaker notes. If the client expects a .pptx that can be tweaked by hand, this is the more pragmatic option. HyperFrames (Apache 2.0, from HeyGen) works on a "write HTML, get video" principle.

What are the key details on Voice: two new releases?

Pocket TTS (code under MIT, from Kyutai) voices text directly on a CPU, with no GPU. For inexpensive servers and local prototypes, that is a major advantage. Phonon (new, released October 20) is on-device speech synthesis. The code is under Apache 2.0, but the multilingual weights are non-commercial. That makes it a trap: you can take the code into a product, but not the model for several languages.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…