Show HN: Flux, A Python-like language in Rust to solve ML orchestration overhead https://bit.ly/49G6gqM

Show HN: Flux, A Python-like language in Rust to solve ML orchestration overhead https://bit.ly/4tfkrLp January 24, 2026 at 03:24AM

Show HN: AdaL Web, a local “Claude co-work” [video] https://bit.ly/4sXVmUQ

Show HN: AdaL Web, a local “Claude co-work” [video] AdaL is the world’s first local coding agent with web UI. Claude Code has proven that coding agents work best when they are local, bringing developers back to the terminal. Terminal UIs are fast and great with shortcuts, shell mode, and developer-friendly workflows. But they are limited in history and image display, and the experience varies by terminal and OS. Many of them flicker (buuuut not AdaL CLI ). Most importantly, they can be quite intimidating for non-technical users. This led us to explore new possibilities for a coding agent interface. What if you could get the best of both worlds: - the same core local agent that does tasks exactly like AdaL CLI - combined with a web UI with no limits on UI/UX This can be especially powerful for design-heavy and more visual workflows Available at: https://bit.ly/49S2fhH https://www.youtube.com/watch?v=smfVGCI08Yk January 24, 2026 at 01:28AM

Show HN: Dwm.tmux – a dwm-inspired window manager for tmux https://bit.ly/3NDp0P9

Show HN: Dwm.tmux – a dwm-inspired window manager for tmux Hey, HN! With all recent agentic workflows being primarily terminal- and tmux-based, I wanted to share a little project I created about decade ago. I've continued to use this as my primary terminal "window manager" and wanted to share in case others might find it useful. I would love to hear about other's terminal-based workflows and any other tools you may use with similar functionality. https://bit.ly/45uTskC January 24, 2026 at 01:15AM

Show HN: Extracting React apps from Figma Make's undocumented binary format https://bit.ly/3NxZfQg

Show HN: Extracting React apps from Figma Make's undocumented binary format https://bit.ly/4qH770s January 23, 2026 at 06:07AM

Show HN: The firmware that got me detained by Swiss Intelligence https://bit.ly/49CzOpo

Show HN: The firmware that got me detained by Swiss Intelligence https://bit.ly/462Wwo7 January 23, 2026 at 05:26AM

Show HN: CleanAF – One-click Desktop cleaner for Windows https://bit.ly/4qxPHDl

Show HN: CleanAF – One-click Desktop cleaner for Windows Hi HN, I built CleanAF because my Windows Desktop kept turning into a dumping ground for downloads and screenshots. CleanAF is a tiny one-click tool that: keeps system icons intact moves everything else into a timestamped “Current Desktop” folder auto-sorts files by type requires no install, no internet, no background service It’s intentionally simple and does one thing only. Source + download: https://bit.ly/46adZuW I’m considering adding undo/restore, scheduling, and exclusion rules if people find it useful. Feedback welcome. https://bit.ly/46adZuW January 23, 2026 at 03:02AM

Show HN: High speed graphics rendering research with tinygrad/tinyJIT https://bit.ly/49NhyZb

Show HN: High speed graphics rendering research with tinygrad/tinyJIT I saw a tweet that tinygrad is so good that you could make a graphics library that wraps tg. So I’ve been hacking on a gtinygrad, and honestly it convinced me it could be used for legit research. The JIT + tensor model ends up being a really nice way to express light transport all in simple python, so I reimplemented some new research papers from SIGGRAPH like REstir PG and SZ and it just works. instead of complicated cpp its just a 200 LOC of python. https://bit.ly/4jSe38d January 22, 2026 at 04:26AM

Show HN: Sweep, Open-weights 1.5B model for next-edit autocomplete https://bit.ly/4qstJla

Show HN: Sweep, Open-weights 1.5B model for next-edit autocomplete Hey HN, we trained and open-sourced a 1.5B model that predicts your next edits, similar to Cursor. You can download the weights here ( https://bit.ly/49JsQO0 ) or try it in our JetBrains plugin ( https://bit.ly/49ElRpq... ). Next-edit autocomplete differs from standard autocomplete by using your recent edits as context when predicting completions. The model is small enough to run locally while outperforming models 4x its size on both speed and accuracy. We tested against Mercury (Inception), Zeta (Zed), and Instinct (Continue) across five benchmarks: next-edit above/below cursor, tab-to-jump for distant changes, standard FIM, and noisiness. We found exact-match accuracy correlates best with real usability because code is fairly precise and the solution space is small. Prompt format turned out to matter more than we expected. We ran a genetic algorithm over 30+ diff formats and found simple `original`/`updated` blocks beat unified diffs. The verbose format is just easier for smaller models to understand. Training was SFT on ~100k examples from permissively-licensed repos (4hrs on 8xH100), then RL for 2000 steps with tree-sitter parse checking and size regularization. The RL step fixes edge cases SFT can’t like, generating code that doesn’t parse or overly verbose outputs. We're open-sourcing the weights so the community can build fast, privacy-preserving autocomplete for any editor. If you're building for VSCode, Neovim, or something else, we'd love to see what you make with it! https://bit.ly/49SL9QV January 22, 2026 at 12:22AM

Show HN: Yashiki – A tiling window manager for macOS in Rust, inspired by River https://bit.ly/3LMdQaf

Show HN: Yashiki – A tiling window manager for macOS in Rust, inspired by River https://bit.ly/3ZvclQN January 19, 2026 at 06:51AM

Show HN: TopicRadar – Track trending topics across HN, GitHub, ArXiv, and more https://bit.ly/4sNyuY2

Show HN: TopicRadar – Track trending topics across HN, GitHub, ArXiv, and more Hey HN! I built TopicRadar to solve a problem I had with staying on top of what's trending in AI/ML without checking 7+ sites daily. https://bit.ly/4b906AL What it does: - Aggregates from HackerNews, GitHub, arXiv, StackOverflow, Lobste.rs, Papers with Code, and Semantic Scholar - One-click presets: "Trending: AI & ML", "Trending: Startups", "Trending: Developer Tools" - Or track custom topics (e.g., "rust async", "transformer models") - Gets 150-175 results in under 5 minutes Built for the Apify $1M Challenge. It's free to try – just hit "Try for free" and use the default "AI & ML" preset. Would love feedback on what sources to add next or features you'd find useful! https://bit.ly/4b906AL January 20, 2026 at 03:47PM

Show HN: macOS native DAW with Git branching model https://bit.ly/4sQhznR

Show HN: macOS native DAW with Git branching model I am working on building (and have made my first prerelease) for a Digital Audio Workstation with git like branching version control. It's free for local use and paid for cloud syncing or collaboration. https://bit.ly/4jMDBDI January 21, 2026 at 01:05AM

Show HN: Automating Type Safety for Mission-Critical Industrial Systems https://bit.ly/49x8Cs7

Show HN: Automating Type Safety for Mission-Critical Industrial Systems https://bit.ly/4b3FmdA January 20, 2026 at 10:43PM

Show HN: E80: an 8-bit CPU in structural VHDL https://bit.ly/4pMdPAU

Show HN: E80: an 8-bit CPU in structural VHDL I built a new 8-bit CPU in VHDL from scratch (starting from the ISA). I felt that most educational soft-cores hide too much behind abstraction, eg. if I can do a+b with a single assignment that calls an optimized arithmetic library, then why did I learn the ripple carry adder in the first place ? And why did I learn flip flops if I can do all my control logic with a simple PROCESS statement like I would with a programming language ? Of course abstraction is the main selling point of HDLs, but would it work if I tried to keep strictly structural and rely on ieee.std_logic_1164 only ? Well, it did and it works nicely. No arithmetic libraries, no PROCESS except for the DFF component (obviously). Of course it's a bit of a "resource hog" compared to optimized cores, (eg. the RAM is build out of flip flops instead of a block ram that takes advantage of FPGA intermal memory) but you can actually trace every signal through the datapath as it happens. I also build an assembler in C99 without external libraries (please be forgiving, my code is very primitive I think). I bundled Sci1 (Scintilla), GHDL and GTKWave into a single installer so you can write assembly and see the waveforms immediately without having to spend hours configuring simulators. Currently Windows only, but at some point I'll have to do it on Linux too. I tested it on the Tang Primer 25K and Cyclone IV, and I included my Gowin, Quartus and Vivado projects files. That should make easy to run on your FPGA. Everything is under the GPL3. (Edit: I did not use AI. Not was it a waste of time for the VHDL because my design is too novel -- but even for beta testing it would waste my time because those LLMs are too well trained for x86/ARM and my flag logic draws from 6502/6800 and even my ripple carry adder doesn't flip the carry bit in subtraction. Point is -- AI couldn't help. It only kept complaining that my assembler's C code wasn't up to 2026 standards) https://bit.ly/3ZrvXoV January 17, 2026 at 10:39PM

Show HN: Artificial Ivy in the Browser https://bit.ly/4qCQ4Ne

Show HN: Artificial Ivy in the Browser This is just a goofy thing I cooked up over the weekend. It's kind of like a screensaver, but with more reading and sliders. (It's not terribly efficient, so expect phone batteries to take a hit!) https://bit.ly/4jRSVz8 January 20, 2026 at 04:14AM

Show HN: Whirligig – Tinder for Gigs https://bit.ly/4qZPc4G

Show HN: Whirligig – Tinder for Gigs https://bit.ly/4oRzecA January 19, 2026 at 11:33PM

Show HN: Pdfwithlove – PDF tools that run 100% locally (no uploads, no back end) https://bit.ly/4qYFBLB

Show HN: Pdfwithlove – PDF tools that run 100% locally (no uploads, no back end) Most PDF web tools make millions by uploading documents that never needed to leave your computer. pdfwithlove does the opposite: 1. 100% local processing 2. No uploads, no backend, no tracking Features include merge/split/edit/compress PDFs, watermarks & signatures, and image/HTML/Office → PDF conversion. https://bit.ly/3Z67TYV January 19, 2026 at 06:04AM

Show HN: AWS-doctor – A terminal-based AWS health check and cost optimizer in Go https://bit.ly/3NyXvGk

Show HN: AWS-doctor – A terminal-based AWS health check and cost optimizer in Go https://bit.ly/3NsbQED January 19, 2026 at 05:35AM

Show HN: Auto-switch keyboard layout per physical keyboard (Rust, Linux/KDE) https://bit.ly/49KHB3i

Show HN: Auto-switch keyboard layout per physical keyboard (Rust, Linux/KDE) https://bit.ly/4quHK1S January 19, 2026 at 01:16AM

Show HN: I quit coding years ago. AI brought me back https://bit.ly/49GCfWC

Show HN: I quit coding years ago. AI brought me back Quick background: I used to code. Studied it in school, wrote some projects, but eventually convinced myself I wasn't cut out for it. Too slow, too many bugs, imposter syndrome — the usual story. So I pivoted, ended up as an investment associate at an early-stage angel fund, and haven't written real code in years. Fast forward to now. I'm a Buffett nerd — big believer in compound interest as a mental model for life. I run compound interest calculations constantly. Not because I need to, but because watching numbers grow over 30-40 years keeps me patient when markets get wild. It's basically meditation for long-term investors. The problem? Every compound interest calculator online is terrible. Ugly interfaces, ads covering half the screen, can't customize compounding frequency properly, no year-by-year breakdowns. I've tried so many. They all suck. When vibe coding started blowing up, something clicked. Maybe I could actually build the calculators I wanted? I don't have to be a "real developer" anymore — I just need to describe what I want clearly. So I tried it. Two weeks and ~$100（Opus 4.5 thinking model） in API costs later: I somehow have 60+ calculators. Started with compound interest, naturally. Then thought "well, while I'm here..." and added mortgage, loan amortization, savings goals, retirement projections. Then it spiraled — BMI calculator, timezone converter, regex tester. Oops. The AI (I'm using Claude via Windsurf) handled the grunt work beautifully. I'd describe exactly what I wanted — "compound interest calculator with monthly/quarterly/yearly options, year-by-year breakdown table, recurring contribution support" — and it delivered. With validation, nice components, even tests. What I realized: my years away from coding weren't wasted. I still understood architecture, I still knew what good UX looked like, I still had domain expertise (financial math). I just couldn't type it all out efficiently. AI filled that gap perfectly. Vibe coding didn't make me a 10x engineer. But it gave me permission to build again. Ideas I've had for years suddenly feel achievable. That's honestly the bigger win for me. Stack: Next.js, React, TailwindCSS, shadcn/ui, four languages (EN/DE/FR/JA). The AI picked most of this when I said "modern and clean." Site's live at https://bit.ly/3NpNjA2 . The compound interest calculator is still my favorite page — finally exactly what I wanted. Curious if others have similar stories. Anyone else come back to building after stepping away? https://bit.ly/3NpNjA2 January 19, 2026 at 01:50AM

Show HN: WebGPU React Renderer Using Vello https://bit.ly/4pKI1ww

Show HN: WebGPU React Renderer Using Vello I've built a package to use Raph Levien's Vello as a blazing fast 2D renderer for React on WebGPU. It uses WASM to hook into the Rust code https://bit.ly/45jCR3b January 17, 2026 at 10:27PM