News for hackers. Stuff that matters.

wassimgr

Show HN: Weightlift – the missing package manager for downloading model weights

https://weightlift.dev/

If you’re building an on-device web-based AI app, you’ve probably built this 50 times already:Something that tells your users which models are available to download or already cached, then shows progress as they’re being downloaded.While building Rescript (see https://github.com/wassgha/rescript ), I had to build this exact same pattern.So I built it into a tiny library that streamlines that...

( Read More... | no comments | on HN )
Hey @dcreater, I’ve just pushed a big update that adds a lot of manual controls over the start/end of words as well as the ability to manually split clips and insert skips. I also deployed a larger, more accurate speech model (whisper small); you can use the model selector to try it out! https://github.com/wassgha/rescript/
This is now on the roadmap!
Follow https://x.com/wassgha for updates/new releases.

Currently on the roadmap:

- Saving/re-opening projects – done

- Importing transcripts (SRT / VTT / JSON) – done

- Support for audio/podcasts – done

- Importing plain text transcripts (auto sync)

- Manually editing word boundaries

- Splitting / reordering of scenes

- Desktop app

- Support for Apple SpeechAnalyzer to remove the need to download a speech model

Hmm, I don’t think that’s exposed through a web api yet. Will probably have to do it when this becomes an Electron desktop app (soon).
one peanut / month is the current price
Also just added the ability to save/restore projects! :)
Just pushed a PR that fixes hallucinations and lets it handle longer form video + the option to select a larger (~600mb) model that has much better transcription accuracy.

Please give it another try! https://github.com/wassgha/rescript/issues/13#issuecomment-5...

Please ignore the attention-grabbing title, the novelty here is that it runs fully in the browser with local/offline models that showcase some of the new browser technologies. This is not meant to be as replacement / fully-featured product, but rather a tech demo and an argument for what "personal software" can look like in the future.
The title is largely for attention-grabbing + relating the project to a product people know already; obviously this is not a replacement, the goal is to experiment with local/offline models and test the limits of the browser and showcase some of the new technologies like SharedArrayBuffer. I work at the limit of javascript performance and making projects like these is how I teach myself new browser APIs.
It took 4 hours and 20 commits :)
Now live on ProductHunt! Please support the launch :) https://www.producthunt.com/products/rescript-edit-videos-li...

Show HN: Descript wanted $24/mo, I built an open-source alternative in a weekend

https://github.com/wassgha/rescript

Descript costs $24/mo, so I built this over a single weekend with Fable!Introducing Rescript: edit videos by simply editing the transcript text. Drop in a video and it is transcribed locally with per-word timestamps and speaker labels. Delete words in the transcript and the corresponding clip is cut from the video.Runs fully in the browser: Local, free, offline and open source.→ Github...

( Read More... | 43 comments | 3 points | on HN )
Re:Denoising Voice Recordings On-Device on Monday June 29, 2026 @11:37AM (#48720680)
Is it available for node? would love to add it to https://news.ycombinator.com/item?id=48715439
You’re basically building https://coreviz.io though
Re:I rebuilt Siri AI from scratch and open sourced it on Monday June 29, 2026 @02:09AM (#48715440)
Announcing OpenDex: A fully open-source, extensible and cross-platform voice-first agentic harness that can control your computer and do tasks on your behalf.

- Fully open-source, completely free (bring your own API key) - Themable, change anything about the UI - Extensible, build your own skills, AI tools and widgets and more - Model agnostic, connect it to Claude, ChatGPT, Ollama or any LLM

Open to contributions!