Ask HN: What are you working on? (June 2026)

Instead it's just a compass face, with brand and distance.

It's free for basic konbini hunting (has probably most of the stored in Japan), and is offline first. So maybe it's useful to other folk?? (I hope so!)

There's some stamp-book style collection things in there too, but that's more fun, and a few `PRO` gated things (more stats, filtering... soon some other store types). But none of it required for finding the next Biru or coffee.

*https://apps.apple.com/us/app/achira-japan-offline-guide/id6...

Writing a sci-fi book, and it’s finally fleshed out to a point that it’s slightly readable, though more like a script. As this is all in markdown in one folder, with some text files as lists, I started writing a simple web project to keep track of it all.

Created a website for local community information, services, etc. Something that removes the reliance upon social media for this. It’s static, making hosting cheap, and in most cases free as it can run on vercel with contentful for blogs and github to store it. I’m sure there’s another project like it, but it’s always good to practice making something myself.

I was asked to show something for STEM week at my daughter’s school. Started a project to demonstrate AÍ to children. Uses very small training data set, you can write the beginning of a one sentence story, it can keep track of a configurable number of tokens, generates a given number of them. Allows taking steps through the process. This is the only one I’m vibe coding because I’m not entirely sure on how to implement it, plus I’ve added multiple models.

Have been working on a girlguiding page specifically for the division my wife volunteers with, as they relied upon the older district site that’s woefully lacking. Stuck waiting for approval.

- navigation without GPS and Internet

- GIS with tokenized raster layers so that LLMs could easily talk to the maps

I have already finished training the standard discriminative auto-regressive architectures by imitation learning on player actions, compared it with previous baselines set in the study. I want to match or exceed the best prior model Kakuna @ 142M params, but in a limited budget. JEPA style world models are showing promise when conditioned on actions [1] and frontier research on JEPA with trajectory straightening [2] shows that improved planning is natural outcome of improved representations.

If any good research ideas come out of this exploration then even better!

Currently fork with my models: https://github.com/sooham/metamon (under checkpoints) Orginal source for pokeagents: https://github.com/metamon/metamon

[1] https://arxiv.org/abs/2603.19312 [2] https://arxiv.org/html/2603.12231v1

A good primer on world models from Welch Labs - one of my favourite ML youtubers: https://www.youtube.com/watch?v=kYkIdXwW2AE

I'm currently working on modeling energy, climate and new policies like universal basic income

https://microlandia.city

One of the hardest parts I've found is the diarisation (who said what) side of things. Trying to tune this and have it working in a way that doesn't absolutely grind the laptop to a halt or take forever to complete has been _hard_ but also extremely rewarding.

Another part has been the fine tuning side of the Phi-4 model, I'm on version 10 now, getting that pipeline down was a journey in itself, but I've got some great results. I wrote a bit about it in a comment here - https://news.ycombinator.com/item?id=48385906#48389625

I absolutely love working on this, I still wake up and the first thing I think about is voice transcription pipelines (sad I know), but I'm excited to see how much further performance and utility I can squeeze out.

After 1.5 years of development and two exhausting pivots, I’m incredibly happy to finally have our v1 live!

While most of the HR tech is rushing to use black-box AI, I built the exact opposite. It's a transparent, math-driven fitness engine. It extracts objective data from CVs and calculates how well applicants match requirements, letting you see the reasoning behind why someone scored an X%.

If anyone here builds in the HR space or regularly hires engineers, I would absolutely love your feedback or a roast of the landing page.

PS This is a project of immense importance for me, I've been working on for past ~2 years, I'd appreciate to know why this comment is flagged.

We’re live in 3 stores and it’s working great.

The system has two parts: a mobile app that lets employees snap photos of products to generate and publish listings, and a hardware box that sits inline between the barcode scanner and POS to track in-store sales and automatically remove the corresponding listing when something sells

We’re live in 3 stores and it’s working great.

The system has two parts: a mobile app that lets employees snap photos of products to generate and publish listings, and a hardware box that sits inline between the barcode scanner and POS to track in-store sales and automatically remove the corresponding listing when something sells.

https://breaka.club/blog/why-were-building-clubs-for-kids

We also teach kids visual scripting in Overcooked 2!, allowing kids to code their way through the levels of an existing much beloved game:

https://youtu.be/ITWSL5lTLig

I'm running an in school pilot this week (Lunch time school club).

The tech stack for the main product is honestly pretty intense at this point with full multiplayer support, offline play, transitioning from client authoritative to joining a remote server. Built atop GodotJS, TypeScript bindings for Godot, which I maintain. Huge monorepo with over a million lines (yes, I'm aware that's NOT a good thing), and GodotJS itself is not included in that.

Reading Brand's quick little life changer kept me going with surprisingly few cuss fits:

The Maintenance of Everything: https://search.worldcat.org/title/1511798465

Thanks, Stew!

We're a collaborative canvas + context engine for all the code and docs in your company, with a zoomable UI + CLI , where you can collaborate with your co-workers and agents.

We map technical debt, agent readiness, code complexity, security scanning, bus factor and more, so you can easily see how all the software in your company runs.

One of the most complex things is our incremental git blame engine built on top of GitOxide, as our backend is fully built on Rust. Our frontend is built on PixiJS so you can explore at gaming speed with 60Hz refresh rates.

Recently we sponsored Rust Week in Europe and a hundred or so developers tried our mini-game which is GeoGuessr for code, and got rave reviews. Future is looking bright!

- native library to build TUI apps without the 20-60MB bloat of node/bun/go

- terminal coding agent harness focused on orchestration/loops

- a small scripting language that looks like JSX but has signals and render optimizations built in

A complete desktop app for browsing and editing your Postgres, MySQL, SQLite data, creating beautiful dashboards, and soon designing automated workflows for repeat tasks.

[1] https://dbpro.app

I've kept a devlog of the last 10 months of building DB Pro, which has been the best way to bring users to the product. I'd highly recommend folks starting a devlog if they can.

Don't tell my husband that I spent more than $200 on parts and supplies for it.

I've wanted a Heathkit since I learned about them as a teenager, and this is the first one I've ever seen in the wild. The original owner left the date he assembled it and his callsign written on the inside! I looked him up and he died in 2013, but by sheer happenstance I'm restoring it 58 years to the day that he initially built it. I got super lucky with this unit because as far as I can tell, it's only been run a few hours in its entire life. I really only have to replace aged components because they're physically breaking down, I expect the thing will outlive me once I'm done with it. Can't wait to hand it off to a bewildered young EE in another half century.

Yesterday, I finally achieved blockchain payload byte-count stability from block 4 and onwards. Today I'm pulling levers to reduce the payload size. Proof-of-everything, so no confirmations needed.

https://bitcointalk.org/index.php?topic=5580132.new

We try to create pieces that stand on their own aesthetically but have a hidden meaning. We currently have two styles: lambda calculus based pieces (we depict the lambda/Tromp diagram) where we have Y-Combinator earrings (well, strictly speaking they are one beta reduction away from Y-combinator. Aesthetic oblige) and a pendant depicting a lambda expression computing Graham's number. The other style is quantum computing circuits, based on quantum computing research my brother (a physics professor) is doing: a pendant that is actually a non-local controlled-NOT gate.

I wrote a tiny DSL to describe the jewelry pieces, and an interpreter to produce CAD files. We then either 3D print them or have them produced by lost-wax.

We are 200% out of our comfort zone (and love it): I know nothing of front end dev, payments, or anything like that. The diamond district in New York is a neighborhood we normally actively avoid, but if you are forced to go there it is fascinating (people examining diamonds on the corner of the street, others in fur coats in summer straight out of a mafia movie...), and especial marketing. Jewelry is a completely saturated business (luckily we are not doing this to pay the rent); we think we have a unique angle, but we are still figuring out the target audience (if there is one).

Store: https://studio-galois.com/

The idea is a project independent knowledge base so agents stop figuring out the same API quirks again and again and instead write down what was solved once. Agents submit via API, vote on each other's entries, anyone can read on the site.

Some thousand entries so far, mostly seeded by my own agents, dev infra stuff and so on. Some of it is real problems i hit in my own projects.

https://hivebook.wiki

Right now there is a runtime and compiler targeting C, written in dependency-free Rust, and a minimal Python frontend. The project is very much proof-of-concept stage so not yet fast. Working on a CUDA backend now.

The goal is to enable automatic discovery of FlashAttention-style optimizations which is not feasible with current compilers.

Very open to feedback/discussion from anybody interested in or knowledgeable about tensor compilers!

repo: https://github.com/ilang-dev/i

https://github.com/ashishb/amazing-sandbox

I discovered this week, while the paper was being reviewed by SG1, that I've accidentally stumbled into tackling a rather important problem. Senders as shipped in C++26 can really only express the async equivalent of inline functions because, except for `task`, all the standard senders fully encode the shape of their computation in their type. With something like the `function` I'm proposing, you can use senders to express async algorithms that are separately compiled, just like sync functions.

If the feature lands in a shape similar to what I've proposed in P4223R0, then I think an obvious extension is to modify the core language to support a newer kind of "coroutine" that allows you to define a sender with imperative code. My vision here is that we act on the observation that `std::execution` is a language feature implemented in the library by teaching the compiler how to turn imperative C++ with `co_await`s sprinkled through it into the corresponding sender and operation state. I think this would open the door to putting async object lifetime analysis and optimization where it belongs (in the compiler) without the overheads and inconveniences of C++20 coroutines. It would even let us apply the inliner to async functions when the compiler can see the body of an async callee, not just its declaration.

For now, my next step is to write P4223R1 to incorporate feedback from this past week's WG21 meeting, and continue exploring the design space around specifying sender attributes for a `function`—I'm thinking the current approach of specifying query function signatures needs to be replaced with a key-value object like receiver environments, but I'm not sure yet what consequences that change would have on the design.

This was very much a passion project and an idea I’ve wanted to see alive for decades, and also let me explore some tech I wanted to get deeper on. I’m bullish on the the tighter integration of CPUs, GPU style cores, and shared memory. Our game, LocoMo, relies heavily of GPU processing of entities under the hood.

You can see me do a walkthrough of the current state of the game here: https://youtu.be/NbB0DCX8Pis?is=vGEw5oTMu_W9f-zT

Choose your genres (and more filters) and get auto updating playlists from your music library. Also just added a new feature - select a few (or one) playlists and create a "mirror", with tracks belonging to the artists of your selected playlists.

My wife and I continue to work on Uruky [1], a simpler Kagi alternative, based in the EU.

Last month we launched image search (got out of beta this month), added our own index and crawler (via Uruky Site Search [2]), and reached 100 monthly active accounts (we’ve passed 150 now)! You can also see a privacy-focused independent blogger wrote about us [3]!!

You can check out the main differences between Uruky and Kagi, DuckDuckGo, SearXNG, etc. in the footer (right side), but one huge difference is that with Uruky, after being a paying customer for 12 months, you can download a copy of the source code (licensed as BUSL into AGPLv3 in 2 years — a suggestion made here in HN)!

You can also now get a free trial for 2 hours when you signup if you pass a proof-of-work captcha (another suggestion made here on HN, and it uses a local Altcha).

Our main challenge continues to be discoverability and outreach because we want to do it ethically. Ideas are welcome! We’ve been sponsoring open source projects, open source maintainers, and indie, small-web, and privacy-related websites and applications.

Feature-wise, for June we’ve already added a ton of personalization and privacy-increasing features like URL rewrites, cash-by-mail payments, and anonymous vouchers! Upcoming is partnering with ProxyStore to sell vouchers (we’re currently in talks for this), so you can buy vouchers with XMR/Monero or other cryptocurrencies. Then we’ll be looking into increasing our own index, focused on indie/small web.

Thank you for your kindness!

[1]: https://uruky.com

[2]: https://uruky.com/site-search

[3]: https://theprivacydad.com/interview-with-the-engineer-of-uru...

Recently started some agentic features for paid version, and this lead to a side project https://eatmydata.ai - a question-to-sql-to-dashboard builder, where data doesn't get exposed to AI (with bundled in-browser SQLite vector search, NER and many other features).

The latter is open-sourced under MIT: https://github.com/eatmydata-org/eatmydata

As a side project I‘m building a multi agent harness that works across desktop and mobile and solves the issue of drowning in too many agent sessions for my own workflow. Hopefully I’ll open source it soon. Reach out if you’d like to beta test it. (Email in my profile)

https://tiledwords.com

I checked my analytics recently and over 100 people have 100+ day streaks which kind of blows my mind!

I released custom player puzzles which has been a lot of fun! I’ve gotten dozens of submissions that I’m working through. People are submitting really clever and interesting puzzles. It’s fun to get to solve puzzles I didn’t make myself! There’s more I want to do here (featured puzzles, categories, etc.)

https://tiledwords.com/player-puzzles/page/1

I think I’ve also tracked down an issue that was causing the game to crash on older iPhones. I’m having playtesters run through it now and hope to deploy tomorrow. (Switching some positioning rules from CSS transforms to SVG coordinates)

I recently made some puzzle brainstorming tools using the Datamuse API which have been very helpful for brainstorming words related to a theme.

I’m starting to debate some monetized features. So far everything is free but it would be nice if my wife and I could dedicate more time to this. If I could get a few thousand dollars a month in subscriptions my wife could quit her job and focus more on puzzle creation and improving the game. If you play and have ideas for features you pay for I’d love to hear them!

I went on sabbatical to fulfill my dream project - consolidating 30 years of training logs that span everything from paper and Excel spreadsheets to various fitness services and devices I used. I'm enjoying the technical challenges involved - digitizing paper hand written logs using OCR / visual generative models, navigating the maze of athletic metrics with their crazy trademarked names and SOTA multidimensional models. Having incredible fun building AI coaches: agents ranging in character from Al Pacino in Any Given Sunday to the coach from my teenage years, utilizing ICL / PFN model-based predictions, ... and more.

The best part is the rush of memories while ingesting my own history - photos and recordings I completely forgot, as well as navigating data shared by friends - records they didn't see in years because the original applications they used no longer exist or won't run on their current HW.

Repo is here if anyone wants to have a look: https://github.com/deosjr/unreal-talk

And a browser-based version can be found here: https://deosjr.github.io/dynamicland/live

Mostly I wanted more art and colour in my workday - something to look at, learn through and draw inspiration from in the moments between meetings and code. You can create an account to save your favourites and curate your own gallery. Just released collections that you can make public.

Art from: The Metropolitan Museum of Art. Art Institute of Chicago. Rijksmuseum. Cleveland Museum of Art.

The thing Im most proud of though is just the viewer, its designed to just open all the images and videos in a folder, and then there is no UI except a right click context menu, the list is a grid or a masonry layout that uses 100% of the space for the images/video so you can just navigate them. It adds anything you open to a local sqlite db so you can tag things if you want optionally. Also control modes that make sense for either a mouse or a laptop trackpad.

It’s a small board with a ATtiny1616 and motor driver that mounts to the bottom of Behringer MF60T replacement faders and provides an I2C interface for reading the position, moving to a specific spot, and even setting up haptic detents, like a linear version of my SmartKnob project.

Perfect for making an intuitive smart light dimmer switch or a macropad.

Just need to find some time to finish making a proper video about it…

https://github.com/scottbez1/FaderBuddy

There is a couple of semi-unique features; you can use your voice to dictate and generate events (feeding, sleep etc), you can also scan documents for growth measurements.

You don't need user account to use it, there is no subscription, the paid features are available behind a single purchase for lifetime. Still, like 90% of the features are available for free.

Also https://www.athilio.com/ privacy focused, highly customisable personal data analytics for your Oura, Garmin, Polar and Apple Health (ios port coming soon). Of course there is couple of AI features (with a single switch to turn all off), originally those were built just so I would learn how to embed agents in sw products myself. The whole app was originally built for personal use to fix missing features in the manufacturers own platforms: - Period over period comparisons (this month vs this month last year) - Comparing different metrics - Customizable graphs and other widgets - And of course combining the manufacturers metrics (oura for sleep, garmin for training etc etc) Existing solutions for this kind of software seem to have focus on social (strava), or coaching (training peaks), or they are just straight up crazy expensive with their paid tier (both tp and strava for example).

Most recently I was also probing people about how they conceptualize of the soul, making my own drawings, and asking others for drawings. If you have a few minutes I would also be interested in seeing how you would draw a soul, given pen and paper or equivalent materials. It often feels like for a lot of people the concept of the soul gets comingled with very confusing definitions.

There's a general problem where certain concepts become so overloaded that just disambiguating and clarifying what is meant becomes a challenge. I will note that if your first thought or question is whether the soul is even real, you might be confused about the definition or we might be referring to different concepts.

https://www.agentkanban.io - Github Copilot / Claude Code integrated Kanban board with context management

https://www.asmusictheory.com - Music Theory lessons, tools, including piano roll with midi in the web browser

However what happens when you actually build and launch your agent is customers try it, do some initial runs and then go ask your manager to automate their use case. That is why I have been building https://toolscaled.com/ The goal being work through your problem space using agentic chat (like Claude Desktop) and then at the end convert it to a workflow. I am pretty close to launching and have been testing. If you're interested send me an email! (if you do sign up just fyi its still in beta so YMMV.

Besides Materia itself I've been bouncing around some other ideas for the Podman quadlet ecosystem. The biggest one is Athanor[1], which re-uses the same plan-execute system and primitives provided by Materia to backup Podman volumes.

I've also been kicking around a clustering system for Podman volumes called Firmament that uses Serf and the built-in Podman import/export API to move volumes to where they need to be in the cluster. But this will probably wait until Materia hits 1.0 before I really start putting effort into it. Or if my homelab needs something like it, whichever comes first :).

[0] https://github.com/stryan/materia ,main site https://primamateria.systems [0] https://github.com/stryan/athanor

https://github.com/linsomniac/apt-cacher-ultra

The primary features I'm focusing on are: It can serve packages if the upstream is unavailable or corrupt, it is reliable.

It snapshots and verifies the cache, and then only updates the snapshot when: a new metadata is available, it has downloaded updated packages that you commonly request, all the metadata checks out.

It's been running in my environment with ~200 clients, ~50 of them get reinstalled every day and then do a full set of package updates and installs. Been working great, even when I shut down Internet access while doing it.

The pitch: It's insane that we have to pull in Python or Lua to build C code. CMake is an abomination against god that has become usable in spite of itself. Zig cc is proof that this entire ecosystem is an embarrassment. My tool gives C projects a TOML manifest, and builds scripts written in C and JIT compiled by the tool. Now, you can write build scripts in the language itself, pull in dependencies you wanted to use anyway.

It also provides a stable ABI. There's an HTTP-backed index and a Git-backed index. And it generally does the same thing for C that, say, Bun did for JS/TS. You'll be able to run C files from source and have the entire ecosystem available. You'll be able to trivially generate single file static binaries, or dynamically link to an older glibc without arcane tricks. It will fix C.

I'm also still working on my "what if we wrote a real standard library for C"; I added some feedback I got from the release.

https://github.com/tspader/sp

If you are a privacy minded person like me, you got only a few options when it comes to email with some ease of use: ProtonMail, Tuta etc. Rather than becoming a new competitor to those, I want to give the power of the decentralized email standard back into the users hand. Everyone with a bit of self-hosting/Linux knowledge, can setup their instances for themselves and their friends/family/business.

Bootstrapped that heavy via vibe coding. Used it to learn a lot about the email standard and related technology. However, I find it too valuable to just be a learning project. Now I'm cleaning it up to get in control again and to proof its secureness by rewriting/restructuring/refactoring line by line.

Then, I will slap an ESP32 & z-wave on it :D secretly to feed my Home Assistant. :D

Texcraft is an attempt to re-implement TeX with a modular/LLVM software architecture. These UIs take the same code in Texcraft that has identical behavior to TeX, and illustrates some of the inner workings of TeX. The lig/kern one is missing instructions :)

I have also found at least one bug in Knuth's TeX recently and am currently writing it up.

(V)RAM conscious AI inferencer/generator: https://github.com/cretz/thinfer

I made my own firmware for the little AI assistant esp32s3 AI balls you can buy from Ali Express. https://www.aliexpress.com/item/1005008627679270.html

Made a version of Infocom adventure frotz player to work with voice control on the AI ball so I can play Planetfall.

Made a toast alerter for Linux terminal command line.

Made a thing to alert you when someone signs in to a Linux server.

[0] https://github.com/asamarts/alint

I also am working on https://trypixie.com - a way to employee your kids legally. It gets money into their Roth and saves you taxable income all while teaching them about working.

I'm also looking into coding harness self-improvement [2]. An inner LLM (raw LLM request) + harness solves coding tasks, an outer agent like Claude or Codex that proposes harness changes. I experimented with many things in the past few months that made me realize this self-improvement thing that everyone is talking about is just an experiment design problem. I wrote about it here [3]. I'm continuing to improve the infra around the self-improvement loop, to increase signal-to-noise ratio per experiment. I'm also generalizing the infra to expand beyond terminal bench tasks and to collect some data across different models (harness-bound vs model-bound).

[1] https://github.com/workofart/ml-by-hand

[2] https://github.com/workofart/harness-experiment

[3] https://www.henrypan.com/blog/2026-05-25-self-improvement-ha...

More broadly, I spent ages developing a self-solving Kanban for mid-sized companies and enterprises (https://kodan.dev) - controllable autonomy level, multiplayer support, remote coding server, works on multirepo projects, mobile support, previews, and more. The pain exists, but it's pretty hard to break the integration barrier.

So I'm spinning the feature I used the most into a separate, easy-to-understand product for now.

Rust-based hybrid search engine: https://github.com/DeepBlueDynamics/lume and https://github.com/DeepBlueDynamics/shivvr

Agentic terminals: https://github.com/DeepBlueDynamics/hyperia

It's full featured with agent loop, gets work done locally.

It's open-source and Swift-based, we built our own inferencing engine since every other engine is based on Python. Check us out - https://github.com/osaurus-ai/osaurus

Looking for some feedback!

It provides digital loyalty cards for cafés (think of an electronic version of paper stamp cards). However with zero apps or customer signup, instead loyalty passes go straight into Apple and Google wallets.

It’s written in Ruby on Rails, which I’m enjoying learning. Still a bit rough around the edges, though it’s free for now so I’d be grateful for your feedback.

Thanks!

Another project is https://www.beeldplek.nl, a timelapse platform powered by community photos. The idea is to place a mount and QR code at fixed viewpoints around the neighbourhood. People scan, photograph the view, optionally add their name, and submit. The infrastructure is up and running but getting the permit to place the mount has been a slow process so far.

Native macOS app, and build on top of Wireshark Libs, so you can see packet details like Wireshark, and it's much easier to use.

Open Source and License under GPL-2.0 at https://github.com/ProxymanApp/TCPViewer

I also made Computer Engineering for Babies which I've posted about on here a couple times before.

https://hackylabs.com

Although the goal is to build an efficient all-in-one-workspace, I wouldn't run a company on it just yet. Right now I'm looking for early adopters who don't mind the rough edges and relatively minimal feature set.

You can grab an early build at https://alpha.totemkb.com.

New workspaces will be in a 14-day 'trial' mode, email rohit@totemkb.com if you'd like me to upgrade your workspace free of charge.

https://packagexray.com/

tsz is my main side project. Trying to learn from this for how to make software in fully automated fashion. tsz's goal is to match tsc (tsgo) but perform better. I am not passing all tsc's own test cases and working towards making it work on complex type packages.

https://github.com/tsz-org/tsz

Everyone is working on personal agents but their identity model is wrong. They act as you, risk your reputation, your data and more. Nym is a personal agent that has (and can make) all of its own accounts and only gets selective read only access to yours.

The goal is to make reliable agents that are able to operate safely in the world to help you do what you want, without exposing your accounts and personal identity to potential harms.

For instance nyms have their own e-mail addresses at nym-mail.com, you can CC them on chains and they can only respond to people on that chain with a lease of 5 days, or permanently for people you specifically add.

Repo with video: https://github.com/monkeydust/rightmind

https://github.com/solomon-b/lambda-calculus-hs

- Imagined job I want to do: Teach software from the ground up, with good illustrations.

- Side: https://peace.mk/ - Create my own automation framework, because I want to make it clear what infrastructure-as-code is going to do before/during/after you run it

- side-side: https://azriel.im/disposition - a diagram generator like graphviz, but supports markdown, to visualise what infrastructure exists / will exist / will be deleted / is in progress when automation is running

- side-side-side: https://azriel.im/dioxus_codemirror - needed a code editor that supports LSP so manually creating diagrams is learnable

I'm back up the stack to the diagram generator, and hopefully soon back to the automation framework

I’m doing composable and dynamic global profiles which are selected based on what you’re doing.

https://github.com/elleryfamilia/rosita

https://logging24.com/landing_a/

The basic idea is to make Regex-scans so fast/cheap that "a metric" can be anything numeric in the text and "tracing" is useless because you can just log (and filter) more things. Turns out Regex at >200GB/s solves a lot of problems.

Metric cardinality explosion is immediately a non-issue, histograms have arbitrary resolution, and you can get from histogram pixels back to the underlying logs. And no need to instrument everything thrice for logs, metrics and traces.

The next big feature I'm aiming for is needle-in-a-haystack searches. The data block headers support it already, but the scan engine doesn't yet use it.

agent-vault-proxy is a local proxy that injects real secrets into requests in-flight, so a compromised or prompt-injected agent has nothing to steal, feedback welcome: https://github.com/inflightsec/agent-vault-proxy

I've primarily been testing it by building out my AI tool chaz into an Eidetica-native AI Agent framework for decentralized Agent sessions. It's working surprisingly well, it maps pretty well onto the storage model and it's uncovering issues with Eidetica I need to fix (which was always my primary reason for building it anyways). https://github.com/arcuru/chaz

Separately I'm building OptiMap, a SIMD-accelerated hashmap repo that explores the design space for hashmaps and benchmarks different approaches. This is mostly for my own learning but I'll eventually turn into a blog post. https://github.com/arcuru/optimap

I've been learning Basque and wanted to see a visualization of how the semantics move into different grammatical structures when translating between Basque and English/Spanish.

Under the hood it's using Stanford NLP to analyze the input then that analysis is given to Claude to generate the data structure needed to visualize the translation. It's really cool and maybe my favorite of the itch-scratchers I've built for myself over the years.

(Xingolak is Basque for "ribbons," a nod to the visualizing metaphor used in the UI.)

It creates its own copy of your workdir for the agent to play in, and then you pull changes out ala git diffs or commits.

    $ yoloai new mybugfix . -a # launch default sandbox in . and also attach the terminal

    # Work with the agent...

    $ yoloai diff mybugfix  # See what it did
    $ yoloai apply mybugfix # Bring out commits and/or uncommitted changes.
    $ yoloai destroy mybugfix

https://kraa.io/about

I know that there are already way too many markdown editors out there, but I think Kraa still offers something unique in this space (combination of minimal UI, plentiful features and some unique stuff like real-real-time chat).

Example of how easy it is to create a 'community' on Kraa: https://kraa.io/kraa/trees

Also - no AI integrations whatsoever.

I built https://loracle.app to automatically build a wiki of various entities in our campaign and enable rag q&a with an ai assistant about specific world facts.

This week we're working on a modular WASM build to allow others to embed Tritium directly into their own platforms. AI native startup law firms love it.

https://tritium.legal

Hallways (https://hallways.lonnycorp.com) - a web browser for 3D spaces, where instead of hyperlinks you have portals that you can seamlessly walk through

LonnyMQ (https://lonnymq.lonnycorp.com) - a performant, production-ready TS PostgreSQL message queue library and accompanying blog post that walks through its design (of which I'm quite proud of)

https://github.com/jpfaraco/muy

I've been building this little animation tool I’ve wanted for years, inspired by one of Bret Victor’s demos from his talk “Inventing on Principle”. I wrote about it here [1].

Basically, instead of setting keyframes and tweens, you perform animations in real time: select a layer, manipulate its properties and the timeline records every frame.

No install, no account needed. It's like Excalidraw, for animation.

I still have some ideas and hope to keep evolving it. And I hope other people find it useful for making neat videos.

[1] https://joaofaraco.com.br/en/projects/muy/

I'm also reading Principles and Practice of Deep Representation Learning, Or: A Mathematical Theory of Memory.

https://flopper.io

Aside from that, I've launched a new tool that tries to promote Solar panels. The UK has some of the most expensive energy in the world, so I've been trying to let homeowners and building managers understand if their building's are suitable.

Uses some APIs all plugged together - including the DEFRA datasets for DSM (LiDAR from planes).

https://solarable.org

I'm also rebuilding an integrated task/knowledge/publication system I'd previously built atop Gemini's Gemtext format. While I loved the simplicity, I've discovered that there are lots of burrs in that design, especially on the publication side, which I'd be able to lift by using a more fully featured document format like Djot.

Modules are simply folders, and the tool just reads from stdio and outputs to stdout. Runs are stored in simple text files with all the info of inputs, outputs and other metadata.

https://github.com/vichoiglesias/kern

2 person team and we didn't do anything manually beside creating the entity relationship, and briefly documenting the overall design system we wanted. Now we are sitting on an almost 80% completed system with 6 more months in hand.

My goal with this language was to pick a set of primitives to compose and express as much as possible.

I don't have a demo up yet but if anyone was curious https://codeberg.org/lilbigdoot/gloo/src/branch/thinkythough...

The two main features I am missing right now are recursive types (I want to do proper mutual recursion and have been procrastinating) and some form of type classes or implicit modules. Structural typing has been useful and I'm finding a lot of features are falling out for free from that.

Long term goal is to create something with performance within a reasonable range of C# / Java etc generally, with tools for opting out of GC. I don't plan on chasing zero cost memory safety, since I want to spend my "budget" on tooling and expressiveness.

Until the language semantics stabilize I plan on generating some pretty naive JS/TS to play around with real programs, and eventually target .NET and native (likely via C++ transpilation)

Been doing this to improve/simplify the grammar for Trilogy[1], a streamlined SQL language - I’ve been planning a redo of one feature and it’s nice to be able to rapidly get feedback on various syntax success rates. Also been particularly useful to optimize error messages, which should help people too.

[1] https://trilogydata.dev/

It works well for me so far and I’m pretty happy with it!

- https://altitag.com: A real estate photographer can upload a drone photo and get points of interest pins overlaid on the photo using EXIF data. The annotations help provide some nice neighborhood context without needing to open up Photoshop.

Today I caused thermal runaway on a BLE thermal (sic!) printer. That melted half of its components together before I noticed. The fun fact is you can do that witouth authorizing, as long as printer is turned on "poof".

Now I'm trying to figure out a BT protocol if DJI Power station, so I can read and track its metrics.

I wrote an improved driver'ish for cheap 5G modem recently. I've been on the last 5% stretch for few months lol.

And I started reverse engineering my LandRover OBD/CAN stuff, so I have some data to publish for other hakers.

In the UK alone, around 7.2 million people have asthma. Globally, WHO estimates that asthma affected 363 million people in 2023 and caused 442,000 deaths.

Peak Flow Meter Diary is not meant to detect every possible trigger. It will not warn you if someone suddenly sprays perfume nearby, or if a dusty bag is opened in the same room. But it could help with risks that can realistically be monitored ahead of time, such as weather, pollen, pollution, cold air, storms, and similar factors. The aim is to make daily tracking easier, show simple visual warnings and notifications, and make it easier to share useful records with clinicians.

I’m also trying to build it in a way that reduces paper, plastic, and electronic waste. If funding allows, I would like to make the project carbon-negative.

That is the bigger dream: to make a small example of how even modest start-up can think about environmental impact from the start, and use it as a practical showcase.

The pitch and full project explanation are here: https://www.kickstarter.com/projects/why5/peak-flow-meter-di...

Feedback welcome, especially from anyone with asthma, clinicians, carers, or people who have worked on health tracking tools. By now I know that my kickstarter is not going anywhere, so I would value any input was the idea that bad, or lack of marketing and accessing appropriate groups etc. I think this community has a lot of experience so I would like someone to share what could have I done better. Do not be shy to tell me if you think idea was waste of time.

https://splat-3d.com

Currently in beta, working out some pipeline optimizations. Looking for people to test! Feel free to try it out, join the discord, etc. Looking for feedback on the experience, reliability, etc.

The goal is for folks to be able to tune their own pipelines, right now I am working on adding more API params/knobs. Looking to build a good capture guide too, since most folks struggle with capture IMO

The last few months I’ve been reading a lot about neuroscience behind learning and practicing music and I’m fascinated by the subject. It has helped me realise why the app works for me, as well as my own mistakes that held back my progress for many years despite putting in decent efforts.

It was a much needed inspiration to continue working on it with a re-evaluated roadmap.

I recently wrote a blog post about it - https://www.captrice.io/blog/what-makes-captrice-work.html

It's still early days, but I have a demo running. Unfortunately, it requires using a drop-in replacement library for CoreLocation. That alone may make it infeasible.

It’s a complete redesign from scratch that combines Mac and Windows into a single codebase via Dioxus (right now they’re two completely separate codebases).

Existing app is at https://www.digitalrebellion.com/posthaste

My mother had a stroke a little over a month ago and I don’t live close by. I went in search of a wellness product that would let me know how she’s doing without her feeling I’m prying too much. I didn’t find one, so now I’m trying to build it. I’m also working on moving closer.

https://github.com/Kapperchino/agent-joe

I also just posted a new blog post on trash valorisation: https://stefan.schueller.net/posts/kva-winterthur/

Repo: https://github.com/paperclipinc/mitos

With this framework, I'm making (among other things) an early literacy app at https://letterspractice.com. My aim here is to hit >= 75% efficacy of Mentava at <= 1% of the price.

The app is near to production readiness, and I'd be happy to share access now with anyone who has verbal but non-literate kids. Be in touch if interested at colin at letterspractice.com

It currently exists of 12 libraries/tool, most of which are pretty stable by now, though some are still very much in flux.

This is one of those things that turns out to be kind of a lot of work. :-)

It started out as a project to try Fable. It wrote a lot of the code and I am learning as I go. I am still questioning some of the design choices but so far it is working. I do want to improve it, so any feedback is welcome.

The service is composed of two open-source services, namely a Nextcloud app (Astrolabe) and backend (nextcloud-mcp-server). I use the service as an MCP server across a number of apps, and others use it primarily for semantic search over large numbers of documents.

Both are open source, and I'm working on a managed offering, completely based in the EU, for individuals/teams that already use Nextcloud and want to be able to use semantic search across some or all of their documents.

Essentially your data stays in Nextcloud, and the MCP server backend keeps a vectordb in sync to enable semantic queries over your content. The number of supported apps is growing, including:

- notes

- deck cards

- files

- news items (RSS feeds)

- cookbook recipes

- contacts & calendar

And I'm adding support for other apps as I go.

Hister is a full text indexer for websites and local files which automatically saves all the visited pages rendered by your browser. It provides offline result previews, a flexible web (and terminal) search interface & query language to explore saved content with ease or quickly fall back to traditional search engines.

I've been using it for a few months and as my local index is growing I can avoid opening google/duckduckgo/kagi - and even websites listed in results - more and more frequently.

The initial reception is overwhelmingly positive with already more than 30 contributors and hundreds of contributions - perhaps you can find it useful as well. (Or at least have some constructive criticism =])

GitHub: https://github.com/asciimoo/hister

Website: https://hister.org/

Small read-only demo: https://demo.hister.org/

Basically every game server hosting provider bills monthly, but most players don't play all the time. So I'm building instalobby with a friend to provide to gamers on-demand hourly billed game servers.

We're starting with Valheim, but expanding to more games hopefully soon.

(If anybody wants to try we are offering $1 worth of credits to every new account)

Most agents for durable workflows feel like toy examples. There is no "Codex" or "Claude Code" for, say, Temporal. So I'm building full-featured agent for these runtimes. Why? Because it makes long-running agents easier to operate and scale. Currently, all frontier harnesses need to run inside a guest OS and need a dedicated process, this is quite challenging to orchestrate and maintain.

To make it work, I had to figure out what part to run as deterministic workflow code, and what part to run as I/O or side effects (aka activities). I'm using a CAS for most of the payloads to maintain a lightweight footprint in the workflow code.

Currently supporting skills, MCP, prompts, a virtual file systems, and soon sandboxes.

https://github.com/smartcomputer-ai/lightspeed

Here's a live example of it figuring out when to post on HN: https://kavla.dev/hn (spoiler, its noon UTC on Sundays)

And here's it generating an interactive map of 20000 earthquakes: https://kavla.dev/quakes

I feel like the canvas is actually a great way to interact with an agent, everything it does is visible, so auditing what it did is (relatively) easy.

I still got some credits to burn so agent usage is free atm (you still have to sign up to use it though)

Providing sandboxes through a CLI. Guardrails such as egress control and secret injection and audit trails built in.

We can also be used as 3rd party sandboxes in Anthropic managed agent and OpenAI sdk.

https://instavm.io/blog/self-hosting-claude-managed-agents-o...

It's Agentic QA + auto-provisioning sandboxes. Makes it plug and play to do code reviews that actually run your code instead of looking at it really hard. B/c the agents control all of the environment (ie running all of the services), it's able to collect runtime evidence about pretty much everything.

A couple open source examples: (Excalidraw) https://app.ito.ai/share/d1cb1475-fbe5-4c71-901b-409ba2aa6d6... & (n8n) https://app.ito.ai/share/bb7d73aa-fd08-482d-9938-87938e2a232...

By default, home page gives all models in the leaderboard, local and hosted. Search for models in the search box on the home page to find the top models by ranking, local(by size) and hosted (by price).

You can also do deep querying/sorting/searching filters of models in each of these three nodes (see the other tabs on top).

The next steps I am working on (would love feedback on this or anything else):

Phase 1: - Change clicks on home page model tiles in one column to search and show models filtered by that across Artificial Analysis, Ollama, OpenRouter - User specifies their system VRAM (unified/dedicated) and we automatically filter the home page with models that would fit on that RAM - in the three columns. - User specifies their price range (per MTok, max across input and output), and we similarly filter and rank by those models across all columns. - User specifies both (VRAM and price range), and we filter by both - leaderboard is union of local and hosted, local by VRAM and hosted by price range match.

Phase 2: Once I have this working, add a local desktop client that automatically reads user system and infers VRAM, renders app as webview. Considering pyside6 with Qt for this.

Phase 3: On desktop client, user can download and chat with the local models automatically based on leaderboard, optionally call hosted models, etc. Used primarily to evaluate and compare local vs hosted models for user's use cases. Also have some interesting alternate experiences to host within the local private app for user to interact with llms, agents, etc.

Do let me know whether this seems useful, or how I can make it more useful.

Working on it has been a joy as ad-blocking tech touches so many aspects of software engineering - from systems and security to the intricacies of JS environments in browsers.

Benefits-wise, system-wide filtering disables ads and tracking not just in browsers, but desktop apps as well (which you'll be amazed how much they do). It's especially relevant now as Google is re-activating their efforts to hinder ad-blockers by killing Manifest V2 in Chrome. So much of tech is actively bleeding cash on AI right now, which means the efforts to screw over users will only accelerate. This makes something that sits at the network level indispensable imo.

Crossed over 100K MRR and I'm shooting for 2M ARR by the end of the year. Growing something in this stage is totally different from making it go from zero to one so it's an interesting learning curve. AI has also changed the calculus as well where it seems less crazy to try and do this sort of thing. Time will tell!

Specs/area are not the focus currently. I just want to build a few useful blocks with it (e.g. analog summer, filter, ...).

Anyway, I’m working on my manual skills of soldering.

HTML/artifact canvases have a lot of potential

https://x.com/backnotprop/status/2064951065439834378?s=20

https://x.com/backnotprop/status/2065436433985474726?s=20

https://github.com/plannotator/effective-html

Shared context workspaces are important

https://plannotator.ai/workspaces/

https://github.com/plannotator/tot

I haven't decided to release it publicly (if so, it'd be free). But it's a link library for my Mac for all the links sent/received in Messages. Apple's new suggested way to do this, of course, is with Siri AI. But I have been using this and like it.

Curious if anybody else would want it.

https://waboost.com/

We are in the process of writing our own vertical stack with Go to control the machine instead of expensive and handicapped solutions from Siemens and etc.

[1] https://hyperclast.com/ - open, fast, self-hostable replacement for Notion

- vibesurfer (https://github.com/frane/vibesurfer): a web browser for agents, without Chromium and CDP.

- agented (https://github.com/frane/agented): a “text editor” for agents, with undo, state, and LSP support.

- grpvn (https://github.com/frane/grpvn): a local chat for your local agent and LLMs.

https://novian.works/voice_test/

in the process, figuring out some tricks for getting opus to work with 3d a bit better

two tricks ive found is to:

1. get claude to present all the orientations to you, then pick which one after 2. convert 3d problems to 2d ones - get it to draw streamlines describing the geometry, rather than trying to look at the whole thing in 3d

fable was a fair bit better at working in 3d than opus is. well, opus mostly isnt

https://bedtimebookhelper.com/

In the mean time, I’m working on a recipe application I’ve had countless false starts on. It’s centered around iterations and version on recipes, tracking changes to ingredients and directions to build new a new recipe from an existing one.

I’m starting with a go Bubbletea tui this time and I’ve been having a lot of fun with it compared to the React SPAs I’ve tried before. Not feeling compelled to style anything while working on the UX has been nice.

https://store.steampowered.com/app/4069810/VERDURE/

https://github.com/agentify-sh/gateway

- https://github.com/rumca-js/awesome-database-feeds - list of RSS sources

- https://github.com/rumca-js/Internet-Places-Database - list of domains

[1] https://apps.apple.com/us/app/reflect-track-anything/id64638...

https://toge.app

The idea is to handle the whole thing, from meeting up at the start point, to multi-day trips, gas stops automagically planned in where you need them.

iOS only right now, Android support is planned but not a priority.

It's a bit of a passion project, as it solves a bit of a "personal" problem, I realize its niche.

I am also not a software engineer, but a DevOps engineer, so it's _entirely_ written by Claude in Swift + Swift UI, Typescript for the backend.

edit: ah, yes also a broker controlled component manager that can start, stop, monitor services over the mentioned broker. This is the carpet that brings the room together.

Strictly human content, pagination instead of endless feeds, one-off payments instead of subscriptions, linear feed by default, public profile scoring instead of secretive algorithms.

Hope to share it soon around here, too.

an open source technical interview platform built for modern interview workflows like takehomes, agent coding sessions, as well as the standard leetcode-style questions.

I sometimes need to have a quick but realistic model of an optical system without paying a few thousand for some of the well known commercial offerings, so I've been building this.

The idea is decompile something like Wordperfect or Framemaker, then port the NeXTStep code to GNUStep and have WP on GNUStep/Linux.

https://josh.works/traffic-bean

It’s founded in Rust and incorporates a Deno runtime for extensions.

It’s headless now, via JSON-RPC. I’ve got the basics of a trait based system which will enable different frontends. At the moment, I’ve created an extension for `pi` which allows me to use that as the frontend.

https://github.com/Fingel/gelly

Also built out a .fits parser that uses rayon to decompress in parallel making it about 5x faster than cfitsio.

https://www.pedaldrivenprogramming.com/2026/06/8x-faster-fit...

It's called Vocast: https://github.com/cnrmurphy/vocast

Thinking about adding some things like queuing RSS feed items to be converted to audio and a feature for being able to do the conversion from my phone.

As it's open source and built with a codebase that's easy for LLM's to work with, users can download it and tailor it to their business/operational requirements, although it also has out of the box 'industry best practice processes' so you don't have to reinvent the wheel and can only focus on writing the 10% custom stuff which differentiates your business.

As all the processes are flexible, you can also do proper 'continuous improvement' with your staff - something traditional WMS products struggle with.

No link because I'm finalising it at the moment, but if you are interested please reply!

https://fata.dev

TUI based interface to search in your files very quickly. I created it from the need to have an equivalent of voidtool's Everything on Linux. It's a bit different though because it's keyboard based. You define zones where you search for files most of the time, and you can manage previous files history. Then there are actions you can perform on each file/folder.

https://www.codako.org/

And been working on a Mario-with-guns game concept: http://devz.cl/posts/what-if-mario-had-a-gun/

Thought it’d be a short concept to get from start to finish but the things you need to implement and plan for in a video game can be near infinite and decision paralysis is a real problem for me.

Desperately trying to attract new monthly sponsors and people willing to buy me the occasional pizza with my terrible HTML skills. Is it working?

https://brynet.ca/wallofpizza.html

https://store.steampowered.com/app/4810350/Medusas_Gaze/?bet...

Created with 0 AI assets

The idea is simple: Its handle of the complexity for AIOps infra like GPU VM provisioning, NVIDIA driver setup, Docker setup, model download, and launching the inference server. User can run any OSS and AI tools inside their cloud.

website + video demo: https://www.dagploy.com github : https://github.com/dagploy/dax

Website: https://www.asfaload.com/

Code: https://github.com/asfaload/asfaload

https://ben3d.ca/blog/proof-of-possession-api-tokens

hack music

• A visual moodboard and notes app that uses local models to link and surface content, a bit like an AI powered Memex.

• A new UI design tool for Mac/iOS with deep support for design systems and AI agents.

• A CMS and static site generator that runs entirely in the browser. Download the site as a zip or publish directly to GitHub/Netlify.

https://github.com/sparktype-project/sparktype

https://v5.filepond.com

Last few years of Congress: https://andratwiro.github.io/riot/?city=congress&solo=1

Reichtag during Hitler's takeover: https://andratwiro.github.io/riot/?city=weimar&solo=1

For curiosity: https://airplane-ai.franzai.com/ based on Gemma

For profit: optimizing my virtual desktop in the cloud setup for AI First workshops

https://github.com/kronotop/kronotop

The project is currently 100% vibe coded with codex\gpt-5.5, but after running some experiments, I'm working on replacing some of the vibe coded SQL engine with Apache Calcite.

https://github.com/coldentry/Specter

And on a new post about how to design web apps for the AI-era for my blog: https://mliezun.com

[0] https://cybernetic.dev

[1] https://github.com/sveltejs/svelte/issues/17176

It takes your instructions, write a versioned spec, then generates a hybrid workflow of code+LLM calls and protects it with tests/evals

The result is that the agents run much faster (90% of it is code), cheaper (LLM steps are scoped tightly and uses smaller models) and reliably (specs get turned into coded state-machine)

https://aptakube.com/

https://www.kickstarter.com/projects/taikohub/taiko-01-keybo...

It let's developer do test planning and testing automation using their coding agents. The records of the testing sessions are then shareable and can be added to PRs, giving the reviewers visibility into how the feature works, what scenarios are handled and tested and what might have been missed.

If you have a business that relies on reviews, I'm looking for a beta tester!

GetSetReply.com aims to:

1. Get you more reviews

2. Avoid negative reviews

3. Respond to reviews

You can email me via my email in my profile.

You play a duck in a small shared town. You pick a job, pay rent, post on a Twitter-style feed, vote in local elections. The simulation keeps running when you close the tab. No PvP, no loot boxes, no combat. Playtime is a few minutes a day by design.

Not technically released even though the site is live, but close enough to a beta at this point.

https://phials.phoundry.app/

The nichest of niche social network clients. It's for people in one particular country, who watch one particular TV program, on one particular day of the week.

Now that the cost of writing software is zero, I love that my focus have moved from vain attempts to generate passive income to just building whatever random shit I feel like. Wish I'd made that choice earlier in life, but no worries!

Wrote up more details at https://openrun.dev/blog/service-binding/

The idea is to make querying ClickHouse feel more like using a polished desktop with ClickHouse native features :

It’s built in Swift/SwiftUI with Monaco as the SQL editor.

Screenshot: https://ibb.co/gbW4rW7G

I’m making a baby book for my son Henri featuring famous Henri’s through history.

I’m also building a zigbee free/busy eink display that only needs to powered once a year or so

I am looking to build a platform that allows for real interview workflows like takehomes, agent coding sessions, as well as the standard leetcode-style questions

It sends me an email once a story hits a certain number of upvotes per minute, so it's useful for keeping track of breaking news.

It'll also soon allow you to get alerted to specific words or phrases in titles. (I have one set up so the monthly hiring threads notify me as soon as they appear.)

https://hnalerts.com

https://raygum.com

https://brodlist.com

https://28times.com

Re-reading the Lean Startup to hone our GTM, market validation and growth engine.

(mathbreakers.com)

https://nicolasbouliane.com/recipes/holunder-syrup

https://odeva.app

https://github.com/panyam/mcpkit

Perhaps the more interesting bit is that it's in Java (not Typescript or Rust)! Java 25 is pretty neat. Bonus: getting to know how to distribute a self-contained Java program using jlink and the likes: https://docs.oracle.com/en/java/javase/17/docs/specs/man/jli...

https://github.com/beeb/swpui

https://github.com/pkhamre/opencode-docker

https://typequicker.com

I made Docker not suck for large images. 2-10x faster depending on the operation. I’ve spent the past two weeks burning down the last bits needed to release a BuildKit integration.

An open source audio interface along the lines of a Scarlett 2i2.

So I made my own dictation app. Supports arbitrary API providers (e.g. Deepgram, Speechmatics, Elevenlabs), Offline models and a subscription if you want it. Otherwise it's free forever for BYOK and offline models. Deepgram is a YC startup from 2016, and have models that are genuinely good - so it's up to you if you want to use them.

Also, Granola doesn't let you read your own meetings after 30 days. So I added a feature in DuckType to import your data from Granola, unlocking all your meetings from their paywall.

Another app: OpenCook https://open-cook.com/ . We curated and wrote our own recipes into StashCook, which requires a subscription just to read your own recipes on the web app. So I got Codex to extract our recipes and rebuild one that is open source, OpenAPI and includes AI features.

This won me 1 year of GPT Pro at the codex event :)

I hope you can tell... I'm tired of companies designing their products to lock you in, to charge you more, with no added value. I build software for people like me. So I'll be building more apps that replace this user hostile software.

- LLMs (any OpenAI compatible API, vLLM, LM Studio, etc.) - image gen + image edit (flux klein) - text to speech (magpie, dia with voice cloning) - speech to text (OpenAI audio transcriptions + riva compatible) - image to textured 3d model (trellis2) - image+text to video (ltx2.3-gguf) - text to music (acestep)

currently it is just me and Claude vibing. While using Fable 5 moved all of my local inference services to k3s across 3 RTX 4090 PCs and my DGX Spark, now I can just tell Claude/Hermes/etc. to start and stop services.

inference.club is built with Tailscale's tsnet library. It is sort of like an OpenRouter built for different types of local AI models. inference.club also lets you showcase and share generated content. For example here is 90 seconds disco funk track generated by acestep: https://inference.club/s/Vxm6ozW24oBs_JGbPcq7tA

I was inspired by AI Horde, and wanted to see if I could build something that could support all of the model modalities that I use for generating short-form AI slop content on local hardware. This is also similar to Hugging Face Spaces, but running on consumer hardware with a common API. I've been watching the quality of local AI inference making massive improvements in quality and performance, and I want to make it easier for people to try "local AI" even if they don't have a GPU.

- No sign-up required & no ads

- Live PDF preview & instant download

- Flexible tax support (VAT, Sales Tax, etc.)

- Fully customizable invoice templates

- 120+ currencies & multi-language support

- 100% In-Browser

So far, I have mostly feature-complete implementations of the following, which are faster than the state-of-the-art implementations, up to 20x faster in some cases while matching or beating them in quality:

- a new 2D data visualization library, along with more bespoke data viz implementations such as word clouds and Primitive.

- programmatic image generation

- image compression

- a new statistical machine learning library, along with more bespoke algorithms such as UMAP and HDBSCAN

- a novel modelless invisible image watermarking approach

- a novel machine learning approach which may be a crime against data science but the performance is really good

- local text embedding generation with MLX

- image-to-ASCII art conversion

- grep/jq replacement (faster than ripgrep)

I aim to open-source them over the next months but the main bottleneck is the inevitable barrage of "gtfo AI slop" comments even if I dot every i and check every t, in addition to the distribution of new software being extremely difficult nowadays due to the death of social media and "20x faster" raising red flags even if I have the data to justify it.

if you have built coding agent in the past using mastra, what are the problems you have faced with mastra? does it support complex branching/context trimming and other features required to efficiently manage context for AI agents?

screen.cam records your Mac and turns it into a polished video. Smart zooms follow your cursor. Filler words come out automatically. AI-assisted YouTube uploads.

CRM with agent baked in that can properly do stuff. No idea why attio/twenty are soooo bad at this. It's a table. getcrme.com / https://github.com/ChristianSch/crme

and gargoyle, an activitypub server with a (theoretically mastodon compatible UI) https://github.com/myfedi/gargoyle. Was annoyed at the homogenous fediverse dev teams out there that don't want their precious service federate with others. I want more federation (tested it with bookwyrms and lemmy for now. Mastodon/GTS also working ofc) and a pretty UI and not waste time with weird identity politics. You do you. I want an open fediverse, not a filter bubble. And GTS was too hard to hack on.

The hard part is doing it without modifying WP, and serverless mariadb that can scale to zero.

A very simple idea: when you eat more than your maintenance calories, you gain weight; when you eat less than your maintenance calories, you lose weight.

By using an algorithm, we can accurately figure out your maintenance calories more accurately than traditional regression based formulas like katch mc ardle.

It's way more accurate than calorie burn tracking devices like fitness bands and watches. (garmin/apple watch etc...)

MacroCodex helps you spot dips in maintenance calories from metabolic adaptation, then auto adjusts your calorie target and macros so your plan stays aligned with your real maintenance calories (TDEE).

It's very useful to those who find it hard to gain or lose weight.

it's a completely free app, no paywall, no unnecessary data collection.

Already reached 13,000+ users

You start a task in Claude Code, and it automatically matches you with a random dev who’s also waiting on theirs.

You can chat, skip, or end the chat anytime.

http://github.com/DevRoulette

https://m4gr4th34.github.io/dossier-001/paper.html

i've massively improved a bunch of things like the AI filter, which now gives you the option of filtering out github repos with AI authorship.

Also improved comments, which I'm serving through my own backend which has made loading of comments super fast, and it's going to be the foundation for some really great other features coming soon.

Soon: HN feature parity via browser extension and sync'd accounts.

https://www.tirreno.com

In the last month or so I added a few nifty features:

- Auto-scan functionality: Instead of having to click on cards to discover what they are, I can now do whole-frame detection on an interval (configurable), so players can mouse over the webcam stream of another player and automatically see what the actual card is. Super helpful for deciding who to attack and makes turns quicker!

- Card view is now grouped by player, since auto-detection will populate a lot of cards during the course of a game.

- Switch the video stream to Livekit from my homebrew version. Players were having video trouble and I hope Livekit is good enough so solve that problem.

Next up: I really want to build a community around this, and I'm struggling on getting the word out to people / having them try it out. I've done some SEO and word of mouth advertising, but haven't had much luck. I feel like I need to switch directions a bit. I'm a developer by trade, so this is wholly new to me.

Come check it out: https://cardcast.gg

> Guild manager for my MMORPG guilds with Discord integration

Deliberately no ads, no subscription, no tracking, works offline.

https://gergltd.com/aperturelab/

- A hand-crafted browser game-engine and game for the engine, with things like determinism at the core. I will be launching soon and can't talk too much about it yet because its quite novel. It actually has quite a few novel ideas within. Very minimal usage of AI in this project, I've been working on it for ~6 years now. A bit toooo long.

- A pure slop-crafted browser extension, because I paid for claude code Fable and it got rug-pulled so I am burning my tokens on a 100% slop project just to see what hands-off coding is actually like. A slight distraction from project 1 I do when I'm feeling a bit burnt out. Super fun so far proc-gen type stuff. Derivative

Continuing development of online training for software testers, with a heavy emphasis on AI, since that’s where the demand is.

During a livestreamed demo yesterday, I ran into a ridiculous bug in Copilot for Excel. After all these years Microsoft still can’t manage the basics of reliability and still deny that they need good testers.

When people deploy python and perl code, they have to either export their entire environment, or build a container. The latter is not possible in a number of deployment cases, and the former carries all manner of dependency radius gotchas.

So I am building (ok, I am prompting/testing/reviewing, the agent is doing the heavy lift) compilers for each of python 3.14.x [1] right now, and perl 5.42.x [2], that can generate static code.

Early stages, perlc does work well, pyc is a work in progress.

[1] https://github.com/joelandman/pyc

[2] https://github.com/joelandman/perlc

indiesecurity.com

https://legitize.app/

still very early and im trying to keep it very affordable, since the whole point is I dont want people wasting their money on hustles that were never legit

> gmd indexes local markdown with full-text, vector, and hybrid search on Typesense; web search, fetch, crawl, and research; llm-wiki pattern and agents; local or cloud.

On top of that, it's lead me down the rabbit hole of a 1995 (limited) theatrical movie called Mr. Payback, which may have only ever existed on 50 sets of laserdiscs distributed to those theaters. I'm hoping to track down a copy of it... if anyone had any clues on that one, I'd love to hear them. I'd purchase a Domesday Dupe device and dump it. But it may be a genuine lost movie.

For example, I was inspired by the activeness of typelit.io when reading - typing out an entire book helped keep my mind from wandering when reading. But typing the whole book is too tedious. I wrote a few scripts to mirror the words on an epub, which does help with focus but isn't quite good enough.

My current epub reader software I use requires you to press a button to reveal the next word. This has dramatically improved my reading comprehension, prevents inadvertent skimming, and keeps my mind from wandering.

I'm still experimenting but for those who have ADHD or are borderline ADHD, it's quite a revelation - I can finally read without my mind wandering.

C++/python/networking/systems/web developer for 30 years with plenty of free time