Writekin – fine-tune a local LLM on your own writing, on your Mac
Hey Hacker News! I built Writekin over the past week because I was tired of AI writing that didn't sound like me, even though I had just used AI to clean it up, rather than wholesale write it. The usual fixes I found online for this were: - Some sort of SKILL.md, or - A system prompt full of rules to strip the generic AI tells (e.g. no em-dashes, none of the stock phrases, varying the sentence length, etc). While those cleaned up the surface a bit, Pangram still came back as ~100% AI written, which was frustrating, as again it was mainly taking my sloppy copy and tweaking it. So when…
What it does
In the maker’s words, at launch
Hey Hacker News! I built Writekin over the past week because I was tired of AI writing that didn't sound like me, even though I had just used AI to clean it up, rather than wholesale write it. The usual fixes I found online for this were: - Some sort of SKILL.md, or - A system prompt full of rules to strip the generic AI tells (e.g. no em-dashes, none of the stock phrases, varying the sentence length, etc). While those cleaned up the surface a bit, Pangram still came back as ~100% AI written, which was frustrating, as again it was mainly taking my sloppy copy and tweaking it. So when building Writekin I took a different route: Writekin fine-tunes a local model on your own writing. It reads what you've already written (Apple Mail, iMessage, local documents, chat exports), curates it into a training corpus, and runs a QLoRA fine-tuning on-device via Apple's MLX. A Compose screen then drafts and rewrites in that voice. Everything runs on your Mac. Ingestion, training, and generation are all local. The only network calls are: (1) When you download the model weights from Hugging Face and (2) The Sparkle update check. Training on your own Mail/Messages only felt okay to ship if the end user could verify that, so the source is public — so you can read exactly what it does! Quick gut check: It's v0.9 and the output is uneven. Honestly, sometimes it nails your voice, and sometimes it's just completely off. This is more a "this is possible and kind of works" than a finished product. Would genuinely love feedback! Source: https://github.com/scouttyg/writekin
Does the same job
all alternatives →
- IBI built an open-source Apple Intelligence-like Writing Tool for Windows2024 · github.com · ▲9
As a high school student, I love LLMs and the execution of Apple's Appel Intelligence Writing Tools, but was disappointed that nothing like that existed for Windows. So, I created Writing Tools, a better than Apple Intelligence open-source alternative that works system-wide on Windows! Key features: - Uses Google's Gemini 1.5 Flash model for superior AI-assisted writing (much better than the tiny 2B parameter Apple Intelligence model) - Works in any application with a customizable hotkey - Proofreads, rewrites, summarizes, and more - Free and privacy-focused (your API key stays local, no…
- BABuilt an Obsidian plugin that rephrases your writing without takin overJun 2026 · rephrasethis.co · ▲5
Writing is hard, and it's tempting to just let AI do the whole thing So I built an Obsidian plugin that keeps AI in its place Highlight a sentence, get some options, pick the one you like Sharpens your writing instead of automating it
- ADAlpha Demo of NotePad.ai (open source)2024 · github.com · ▲7
Hi all, I threw together a small prototype I am calling “Notepad.ai”. A new take on UIs for interacting with LLMs. While I enjoy using LLM’s in the chat format I wanted to see what it would be like to do it in a more long form style. It let’s you write in a pretty free form, much like Window’s Notepad, but you can choose to hit ctrl+[ to analyze the text with a preset prompt of your choosing. It has a few other small features. It’s WIP and very experimental. I would appreciate any feedback or thoughts. Video: https://youtu.be/ntdlgFmSxQY Live Demo:…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, July 2026
the whole month →- IR
I might be the only SRE on Earth with his own bowling center. It's a more in-depth gig than you'd think. My family and I bought an abandoned 8-lane bowling center in the rural mid-west. In our small town there weren't many recreation options for families. You've heard of a food desert? This is an R&R desert. It had been abandoned for a good reason. The roof leaks, the electrical system was constantly surging, and my 70-year-old bowling equipment (still) doesn't work perfectly. The system that keeps your score is particularly interesting to me. It's the thing you watch during your game, but…
Life & fun · Jul 2026
- EElevators▲1,680
Life & fun · Jul 2026 · john.fun
- 1W18 Words▲1,160
Life & fun · Jul 2026 · 18words.com
- BA
Over the past few months, our team has been building more and more slidedecks using web frontend technologies with coding harnesses like Claude Code, but a common complaint is to make even small edits we need to edit the code either manually or via the harness. To avoid this loop, I ended up creating Bento, a single HTML file with everything you need in a slide tool including animations and shared editing. There's no install or cloud login, everything works offline. The default deck is around 560 KB and it doesn't need to fetch anything once you got it. Open it in a browser and then you can…
Dev tools · Jul 2026 · bento.page
- GG
A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me. But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility. I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context.…
AI · Jul 2026 · github.com
