Show HN: Hob - Open-source, on device alternative to Claude Cowork and ChatGPT https://ift.tt/s5phXU2

Show HN: Hob - Open-source, on device alternative to Claude Cowork and ChatGPT Hey HN, I'm Christian and this is a new OSS project I've been noodling on for the last couple of weeks. Hob is a free, MIT-licensed desktop app for Mac, Windows and Linux that works on your files the way the Claude and ChatGPT desktop apps do, except the model runs on your own computer. It uses open-weight models (currently Gemma 4 and gpt-oss). All your chats and files never leave your system. The only things that go online are web search (can be turned off in settings) and any MCP connectors you add yourself. Main features: - Automatically finds the best open-weight model for your hardware on first install (Mac/Linux/Windows) - Writes sandboxed Python code for ad hoc tools (data processing, document conversion, graphs etc.) - Built-in web search (powered by SearXNG) - MCP support - Image recognition - CRUD to your local filesystem - Works fully offline with no internet Before building this I tried every local AI app I could find, but it seemed each one had different problems. The main thing I noticed is that none of them could do what the ChatGPT and Claude apps do with your files, e.g. read and change files in a folder, look at images, run code in a sandbox, draw a chart. The most advanced thing any of them did was paste a plain text file into a small context window, which basically failed as soon as you had a file larger than a few KB. I couldn't find anything that could easily parse images and bigger PDFs. The OSS solutions that were more advanced all seemed to default to hosted models (either OSS or closed), which defeats the entire point of the privacy that on-device OSS offers. They also seemed to push towards an account/open-core style business. I build it using Electron bridged with llama.cpp in its own process. On initial load, the system will try and figure out if you have a compatible graphics card and if not default to CPU-only (I ran into a lot of issues of graphics card beeing seemingly compatible but then crashing on first query, there's a whole saga to this lol). Code execution happens in a sandbox that can only write to the chat's folder and has no network unless asked (sandbox-exec on macOS, Landlock on Linux and a low-integrity process on Windows). Images go to a small Qwen3-VL model. Dictation is a local Whisper model. The hardest part was definitely to get the smaller models to use tools reliably. Ran into a ton of issues of 4B models happily saying "here's your chart" without ever calling the chart tool or get stuck in an infinite call loop. I spent a fair amount of the work building guards vs tweaking prompts (that didn't really work for small models). I understand the pushback that local models are weaker than frontier cloud models. This is true, but for most tasks for most normal people, I'd argue you rarely need the newest frontier model. From my own evals, systems with 16+ GB of RAM do really well, and if you have a GPU or more RAM, you get very close to what you'll get out of the box with a frontier hosted model. Excited to hear what you think and see how it runs on other systems! https://ift.tt/oRc83Xn October 26, 2026 at 09:42PM

Show HN: RPI Pico (RP2040) PIO Emulator Package for Go https://ift.tt/AadQnMv

Show HN: RPI Pico (RP2040) PIO Emulator Package for Go I have been developing a bus communication protocol for a while using the Raspberry Pi Pico (RP2040)'s programmable IO (PIO) hardware (basically- 8 tiny processors that run assembly programs to manipulate GPIO pins) and I needed some way of testing my programs, as the real hardware is undebuggable. Unfortunately, none of the existing emulators I could find supported connecting multiple RP2040s together, and they each had a few different missing bits of functionality, so I decided to make my own! My emulator supports running multiple RP2040s electrically connected together, and emulates the whole of their PIO blocks (all 8 processors each), along with all of the more niche inter-processor interactions and GPIO functionality I could possibly find in the datasheet. I built it as a Go package so that it can be used inside of other programs; I intend on using it myself inside a test suite against my in-development protocol. I've spent quite a long time checking and double-checking most of the logic so it is as complete and accurate as I can make it, but if you spot anything amiss, I'd really appreciate the feedback. Thanks for checking it out! Hope it can be useful to other RP2040 users. https://ift.tt/v6JUrhc October 26, 2026 at 10:59PM

Show HN: Hurston-1 open source TTS that speak Southern American and Afro-English https://ift.tt/719rZzO

Show HN: Hurston-1 open source TTS that speak Southern American and Afro-English Hey HN, I’m open sourcing my Hurston-1 narrowband TTS model (Apache-2.0). It is very good with American English and does especially well with Southern American and Regional African English accents. Destined AI is based in Atlanta, and I built this because I’m amazed by what AI can do, and I wanted to help introduce and welcome more people to feel comfortable using and interacting with AI. Since 2021, I’ve been collecting fully consented voice data from over 20,000 people, 1M audio recordings, and over 1,800 hrs of audio captured to date. For this model, I used a small part of the data, 136 hrs, to finetune Maya1 and expand it’s capabilities. Hurston-1 is a narrowband TTS model that focuses the output on telephone use because that is still where most of these voices will be heard. I’ve used it to support customers such as Amazon and Salesforce and generated over half-mill revenue with it. Now I’m open sourcing it to help unlock opportunities for others as well. Limitations: this model doesn’t maintain the exact same voice across prompts and doesn’t support voice cloning. For many use cases, this is fine but maybe not so - if you’re creating podcasts. This is the first release - I’m already working on the next version with improvements. I would appreciate, if anyone is interested in this space, download the model from huggingface, check it out and let me know any feedback. Also, open to connecting with others in this space as well. https://ift.tt/bNuF8pQ October 25, 2026 at 07:00AM

Show HN: Steal My Bitcoin https://ift.tt/RCL3ENF

Show HN: Steal My Bitcoin Made this because I'm curious if anyone out there has figured out how to actually use this PHP RCE we spent the last few days panicking over. Have fun! https://ift.tt/fWIN5OQ October 25, 2026 at 02:00AM

Show HN: A phone lock I can't undo on impulse because I never learn the code https://ift.tt/VTs84Kw

Show HN: A phone lock I can't undo on impulse because I never learn the code https://ift.tt/ZNorWXt October 24, 2026 at 07:52PM

Show HN: Agentsh – A Decentralized, Scalable Multiagent Harness https://ift.tt/vpTMWBt

Show HN: Agentsh – A Decentralized, Scalable Multiagent Harness https://ift.tt/Y1bWmjG October 24, 2026 at 11:49PM

Show HN: MXValidation – Send an email and get an authentication and DNS report https://ift.tt/CbnXe2w

Show HN: MXValidation – Send an email and get an authentication and DNS report https://ift.tt/Jsq5dEY October 24, 2026 at 12:57AM

Show HN: Amgi – Plain-text spaced repetition with FSRS https://ift.tt/gFoLl1M

Show HN: Amgi – Plain-text spaced repetition with FSRS Hi HN! I love Anki and I've been using it for 5–6 months pretty much consistently. It's a wonderful piece of software, but its obvious downside is its UI. I love beautiful things, and spending 30–40 minutes daily on something that doesn't bring me aesthetic joy is kinda urgh... So I'm building Amgi (암기, means memorization in Korean), an open-source plain-text spaced repetition system with FSRS. I started building it only a week ago. Flash cards are simply Markdown files with pretty elegant syntax, so you are not tied to any app. Just open your flashcards folder in your favorite editor and add/edit straightaway: Q: What does FSRS stand for? A: Free Spaced Repetition Scheduler C: The capital of [France] is [Paris|city]. Amgi has a dashboard where you can see your streak & some statistics on your reviews, and a browser where you can search using queries similar to Anki's (deck:Name, #tag, is:due, etc.). It also supports LaTeX, images, audio and code blocks (incl. making cloze cards inside of them). And it has a cute installer! (I hope) It can already import from Anki, but it's still pretty early and the only Anki collection I've tested on is my own, so expect some rough edges. Mac-only for now! My own collection is public too, it's the actual one I use daily: https://ift.tt/5oQ2mhT Would love feedback, especially on the card syntax :) https://ift.tt/gfT24X7 October 23, 2026 at 11:52PM

Show HN: Buildnotes – Turn pull requests in "how it's made" posts https://ift.tt/kRGcdxo

Show HN: Buildnotes – Turn pull requests in "how it's made" posts I've been working on a way to take existing pull requests and turn them into "How It's Made" style content you can use outside of Github. buildnotes is an agent skill that reads a feature's issues, PRs, commits and reviews, then produces a post with real figures. Those include flow diagrams, annotated screenshots with notes in the margin, before/after comparisons, charts and real code. Also supports videos + slides, so you can share them externally too. Open source: https://ift.tt/5Ne6JV2 https://ift.tt/lGO064g October 22, 2026 at 11:57PM

Show HN: Overdesigned Simple Solitaire https://ift.tt/8nUEcrN

Show HN: Overdesigned Simple Solitaire I got irritated by the maximalism and lack of aesthetics present in modern software when trying to find a nice solitaire. A solitaire does not need a score, levels - and most of all it does not need to download anything, have user log in or do anything else except maintain game state. So since software comes automatically nowadays, I decided as a weekend project to create the last solitaire I'll ever need. The application is built on Rust so I can compile it to native. But it runs just as well as web app! I geeked out completely on design (it may look like "default" but that's the point) to create as non-intruding yet aesthetically pleasing experience as I could. I might have accidentally put in a complete card design system with a content pipeline (maybe I'll do default blackjack next) - but as a player you only see the one pack of cards. Anyway, hope someone here appreciates the minimalistic aesthetics. The game state is saved in the url so if you want to save a nice game you can do it by copypasting the link. The game engine is versioned so if I ever change the engine or state schema there will be support for the current version as well. There _is_ a novelty. If you click the "teach" button it shows all the available free moves as arrows. Hope this is fun for someone! https://ift.tt/y4aM1kx October 22, 2026 at 02:15AM

Show HN: Renor – private memory for coding agents, shared across tools over MCP https://ift.tt/kmK7uYU

Show HN: Renor – private memory for coding agents, shared across tools over MCP https://ift.tt/kWMI71c October 22, 2026 at 12:52AM

Show HN: I built ESP32-S3 LED tubes for live coding https://ift.tt/7pPqMdF

Show HN: I built ESP32-S3 LED tubes for live coding https://ift.tt/sAwjq5z October 22, 2026 at 12:25AM

Show HN: MCP Census – a live change feed for 48,797 MCP servers https://ift.tt/Ye1moB3

Show HN: MCP Census – a live change feed for 48,797 MCP servers https://mcpcensus.com/ October 21, 2026 at 03:18AM

Show HN: Rouler – Steer Windows traffic per app or IP across multiple interfaces https://ift.tt/u4YOfVk

Show HN: Rouler – Steer Windows traffic per app or IP across multiple interfaces Hi HN, Rouler is "rules for your routes". It allows you to steer network traffic across multiple network interfaces on Windows. For example: - Run Tailscale, Mullvad and OpenVPN official clients simultaneously. - Use Rouler to send Slack via the OpenVPN connection, Chrome + an IP range via Tailscale and a torrent client via Mullvad. Rouler is VPN agnostic, so you can "split-tunnel" with anything that sets up a windows network interface. I'm an independent developer from Australia and have been working on Rouler for a bit over a year. It's been a challenging project to say the least, but I'm really happy with how it's turning out. I have spent a great deal of time building and throwing things away, starting over, belting the initial poc into something that I find intuitive... hopefully others do too! I've done all the testing, fixing and improving I can, so now it's open for public beta. I'd love to hear your feedback... and there are some specific points I'm interested in hearing about listed on the Rouler website. Please reply with any questions you might have. Thanks for reading! Brendan p.s. Rouler is "built in Rust" TM ;-) --- On AI (skip if you're sick of hearing about it): Since AI is the hot topic and people are naturally cautious about it, I'll add some quick notes about it, specifically about the driver development as that's the key thing to get right. The driver is hand written and every detail has been poured over many times over. I spent somewhere around 6-9 months manually building and testing just the driver, staring at Dbgview logs, wireshark logs, windows route tables, restarting VM's back to their snapshot and manually copying things across to start again. That's not always the most efficient way of doing things, but it's the way it had to be done. If I had tried to vibe code it Rouler simply wouldn't exist. I did try to lean on AI at times, but it constantly sent me in the wrong direction, told me things weren't possible, tried to take shortcuts and implement poor performing trade-offs. It also loved to hallucinate the windows api. Keep in mind this was around the Opus 4.1 - 4.6 days. Outside of the driver, in user mode land, AI has crept more and more into my workflow. I've had a lot of success utilising the latest models in the app in particular, and I purposefully separated the UI and core logic so I'm able to quickly iterate on the UI. I've also found AI very useful for scripting and refactoring. It has built some great scripts that would've taken me days or weeks to build and fine tune. https://rouler.io/ October 20, 2026 at 09:21AM