alt Hacker NewsAsk HN: What are you working on? (September 2026)
What are you working on? What have you been curious about lately?
Comments
Totem is a collaborative workspace, featuring tasks, notes, and docs, built in Rust and available across desktop, web, and mobile (iOS to start) with seamless syncing and offline merge.
I built this because I think there's a lack of true realtime (CRDT) synced notes that aren't built on a heavy platform that take up 1GB+ RAM (Totem takes up ~100MB, and is also generally more responsive due to the local-first + ops-log approach).
Another thing I care about is keeping the underlying data portable: Totem is built on Markdown and SQLite as its core formats, so you get the convenience of cloud-based syncing without locking your data into a proprietary format.
I’m still looking for beta users/design partners before the initial launch. You can try it without creating an account here:
https://app.thinktotem.com/sandbox
And the landing page is here:
This is a bit of an unplanned post, so the sandbox template is a bit barebones (empty docs, doesn't feature a properly setup table). Nevertheless, I would love to hear your feedback if you try it out, especially if you use collaborative notes/docs today and have opinions on what they’re missing!
My wife and I made a daily guessing game together. Unlike others of the genre that we enjoy playing, an explicit goal of this one is teaching you about the world as you play it. It’s quite simple, using Wikipedia and factbook.json for information about the countries, but it incorporates a few ideas that I think are fun and there’s a lot I’d like to experiment with.
My funny essays newsletter on substack. Link to latest post: https://yelluwcomedy.substack.com/p/the-fascinating-world-of...
Please let me know if you enjoy!
Not something I'm working on, but want to: a competitor to Perforce that will be cheaper and easier.
Unfortunately, I'm not sure that being better is enough to compete, so I'm sitting on it for now.
A web app for saving internet recipe pages that strips out all the crap.
Spending almost all of my time working on DNTLS (https://dntls.net/). Yesterday I was able to create secure mTLS tunnels over IPv6 for direct P2P to my co-founder's desktop in SE Asia (I'm in the US). He was able to load a website I was hosting locally on my machine. I also tested coordinating through a relay to avoid publishing my public IP and sending all traffic through the relay to hide my IP entirely. We more or less recreated the "old" internet in a way except everything is mTLS and E2E encrypted.
Bunch of things: 1. Building a local ai news feed to learn what's going on latest in terms of technical discussion, research papers, projects - https://local-ai-signal.vercel.app/ 2. Build a second brain by injecting years of my notes using hermes, luna model and obisidian
Self hosted database access management, think Code Reviews for SQL Statements for when your devs need to go to prod: https://github.com/kviklet/kviklet
Recently cleaned up my postgres proxy and built a MySQL wire proxy so that you can use psql/datagrip as a dev but every statement is still logged. With SSO and no password sharing ofc!
Iterating through and evaluating possible approaches for shrinking the binary size of a tree-sitter based Vi/Vim compatible tags generator I am developing without affecting the developer experience for adding support for newer languages[1]. A couple of approaches have already been tried and failed to live up to the expectations on developer experience [2][3]. Next in line is to look into bundling only the tree-walking code for all supported languages with tree-tags and allow the user to configure the grammars to download separately from the main binary, possibly using the `wasm` feature which allows native library to run wasm compiled grammars.
[1] https://github.com/jha-naman/treetags
[2] https://github.com/jha-naman/treetags/pull/62
Tried to use a generator for automating the bulk of the scanner for a language from `grammar.json` file of a tree-sitter grammar. A runtime engine uses the scanner for a forward only walk on the scanned code and calls the hooks to generate tags.
[3] https://github.com/jha-naman/treetags/pull/56
A complicated mess of a generator that takes a few declarative inputs and tries to generate code for spitting out tags for a given language.
working on a storytelling app for kids called Tella Stories that goes deeper than just some minimal personalization.. know there's lots of stuff that claims to make them the hero, yadda yadda but then they just give you a shitty digital templated story or try to sell you a printed version. this is focused on a high level digital experience and something parents can actually trust, and that kids actually enjoy.
so yeah.. it builds the stories out of their actual life. their family and friends are in the cast, their dog, the thing they're nervous about this week. the kid doesn't just appear in the story, the story is about their world and imagination and grows with them.
ultimate goal is to keep iterating and get to a level of quality of Bluey, etc but have it actually be the kids world with agency/choices vs. one they have to passively sit back and watch, and that they're not in.
would love any feedback. i know there's been others like Ello shared on here and those threads were really interesting to know HN's opinion on. this does not take an educational and instructional approach. Tella is meant to help develop who they are through stories and be a better screen alternative time to the rest of the algo/curated garbage out there. anyway you guys are a critical and technical bunch and i respect/appreciate that.. especially since this of course leverages multiple AI services to create the stories ~
I'm building https://dipstickalerts.com.
A way for for U.S. car owners to stay informed about the issues your car may experience that are communicated by car makers to dealerships.
I found those communications useful for my own car and wanted a better website that what the NHTSA provides and have been having a lot of fun building out the site.
working on adding webmcp support to my new free tool https://easyanalytica.com/tools/gsc-insight , webmcp was recently added to codex desktop app and it would be interesting to see how users use it with ai, would it get good adoption or it would just go away.
Been working on getting my webpage refreshed to try and restart my DevOps/Infra/Fractional consulting business: https://cubiclerebels.com.
Also spent a good part of last month working on a little idea I have that is built on-top of XMPP.
I've also been wanting to pickup Gleam, hopefully this would be that week I finally fight through my ADHD to do so.
Started a few different things:
1. A robotics information and news website. Any feedback is appreciated. https://robotica.dev/
2. Applied DBSP theory to Postgres https://github.com/getravi/pg_dbsp and DuckDB https://github.com/getravi/duckDBSP. The idea is to have incrementally maintained materialized views.
I'm hosting an book club themed event calendar for New Yorkers!
We also have an official meeting once a month to do book club matchmaking in person.
I've been building Ballad – an inbound marketing engine that runs itself. I'm building it for other founders like myself who don't have a marketing person and aren't going to hire one.
My last company (Courier, YC S19) had great inbound but we never were able to scale it. Ballad plans what I should write, drafts in my voice, publishes to my accounts, and handles attribution so it can feed that into what it writes next.
I’ve been building a dark software factory. A system that takes stray thoughts and automatically plans and builds.
I’ve built two (successful) iterations and am now working on the third. They are all very minimal.
The main one works well and runs full time waiting for my next idea. The second targeted an open weights model and worked in limited cases.
Now I’m working on a more autonomous version. I give it only a goal and a strategy, it decides its own projects.
Slowly putting my photography online again at https://flaneurphoto.com - in the middle of composing soundscapes for some of the works.
Finally getting around to organising my photo archive and setting up a metadata extraction pipeline that makes sense for me.
(At the day job, curious about adapting app layouts to accommodate the iPhone Duo and address the various Android foldables at the same time, too).
I'm doing some programs to use the API for my local bus company to show when buses are arriving at stops. Final step is making it output the data to an external display.
https://github.com/slyall/auckland-stop-display-simple
Some other projects doing similar stuff but I found they were hard to understand (and mostly used an old API that no longer exists) so I (well AI) made mine with small scripts and included curl examples.
ssh sshfighter.com
Been optimising an ssh ansi rendering engine for six months, starting to look great. Homepage is slop please forgive
I'm building Bellerophon, a DSL with decoupled firmware adapter architecture for 3D printer control, so the same source code can compile to Klipper, Marlin, or RepRap without rewriting logic per target. Recently used it to generate print patterns with math that you wouldn't get from a slicer.
Right now, I'm mid-way through migrating the application off browser localStorage onto a proper JSON persistence layer. My first large solo project as a busy student.
I hope to make it better as I continue.
A FHIR Validator and Snapshot Generator that works offline and can communicate with the German national Terminology service
Cranpose - Rust GUI with jetpack Compose API https://github.com/samoylenkodmitry/Cranpose
A rag api to power better, more tailored ai features in saas products. By default it indexes customer websites, but also supports notion, YouTube, email and custom connections. https://www.context-link.ai/for/white-label-rag-api
Cricket, a visual scripting plugin for Figma inspired by Grasshopper for Rhino.
I find Grasshopper, especially its data model, ideal for a certain type of algorithmic art/design, and have always wanted to use it in a 2D graphics package.
https://www.figma.com/community/plugin/1677548868352947603/c...
Keyboard pants https://mastodon.pirateparty.be/@utopiah/117157231799845118
TL;DR: magnets.
I’ve been working on Pick Up, a reading companion I built to make tracking books feel a bit more personal and fun.
You can track what you’re reading, keep notes and reflections, see your reading stats, build out your bookshelf, and a bunch more. Over 4k monthly active users.
Been building it solo & adding features based on what readers ask for.
I'm working on CLI/TUI tool that stores all code collaboration (issues, PRs, etc) in git itself. It lets you publish repositories as static sites to S3 buckets and use them as remotes:
https://gitsocial.org/ https://github.com/gitsocial-org/gitsocial
I'm making an online guidance solver for KSA (https://ahwoo.com/app/100000/kitten-space-agency). It will allow you to control a KSA rocket from the command line for a variety of maneuvers (launch, rendezvous, etc). It uses successive convexification to turn the full nonlinear problem into a sequence of convex problems with linear constraints and quadratic objective.
It has proven surprisingly resistant to AI so far - Astra can make a very quick prototype but upon review it had many defects. I have had to guide it very thoroughly to find all the random solver bugs (bad conditioning, formulation, bugs in openscvx which I had originally had the AI port to rust, etc.) preventing well-behaved solves. But I think (hope?) I have turned the corner on these.
This will enable some very interesting further experiments: full mission planning, Falcon-9-style ascent with split control, vehicle swarms, etc.
Im working on my own coding agent soon to be released.
While you might think "yey another coding agent" i went down some very different paths than other open source coding agents do right now. Ill not bore you with a ton of detailed breakdowns and rather list the main points that are relevant to understand why/how i build it.
Premise:
-The agent is meant to run against selfhosted environments first like Ollama etc with a focus on non cloud sized models, it should work properly on something like a 27b model already
-The agent with full intention trades execution time vs reasoning and result quality
So what does it do different:
1. The agent, instead of trying to let the model solve a whole task in 3-5 inferences, rather breaks down the "thinking process" in smaller chunks, basically decomposing the task into smaller tasks. While this is not a completely new concept, the agent will break it down to really simple single step variants which even for a smaller model can properly be solved. This involves not just file editings but all points along the way from interpreting user message to planing a task to defining acceptance criteria etc.
2. The agent is build in a way that it, apart from some fixed steps such as detect user intent at the beginning and synthesize response at the end, it will by itself decide what is the next proper step to execute/do based on the context i provide. While doing so the agent also revalidates its current execution and if it spots that, given the information i collected since it planned and started running, the current plan is not detailed enough or is missing out due to something he didnt "know" when he was planing, the agent will dynamicly mutate the coming plan stats, as in restructure/decompose/etc in order to have a properly layed out route to fulfill the task.
3. The agent has a rather complex system of how his context is composed which is combined by things like the chat history, previous step results, a backlog of what "single steps" it has done (updated file x , read range from file y etc) and in that it is strongly reason driven, so it will for most of the things it does also provide a small reason explaination for why it did that. Than the agent also has a working memory that contains facts and decisions it made along execution. There is more but thats just some thingsd to mention. The whole memory system and context is rather complex (tho not complicated). When i release it ill properly break down how all this works in the docs.
4. The agent provides full observability in the ui. At every point of execution you can basically in detail see whats in memory, whats the current composed context, whats the plan and what plan steps depend on what previous steps, you can look at artefacts it build and also you have basically a complete audit log of all single "actions" it has executed and can look into their details.
5. The agent allows to while in an execution send steering messages, meaning the message will be, using relevant current contextual information about its execution, send as inference with a prompt askind the agent to interpret on if its relevant, and how does it impact the current execution/plan. It than can decide either just do alter what it currently does, or even do a plan mutation in order to accomodate the users steering information.
6. The agent within its capabilities has a lot of error correction/self reflection logic. From dynamicly fixing json including a fallback to let LLM fix a response json, upon to if executions fail to much it will itself do a hypothesis on why this is happening, formulate a critique and with that reevaluate if it wants to try the steps again with the additional information, or if it wants to mutate the plan to accomodate the just spotted problems.
Theres quite some more i could list but i guess thats enaugh for now. The big trade off as mentioned earlier is execution time. In comparsion to cloud native agents, which will do giant editings and reasonings in just a hand full of inferences, my agent uses a ton of inferences. The big difference is that this allows (a) the agent to more precisly focus on specific tasks rather than overloading it with to much at once and (b) alot better spot problems/mistakes itself and adjust its execution plan to accomodate those without the user having to reprompt the agent 10 times until something is actually correctly solved.
Apart from the pure coding capability, the very same actions that allow for higher quality reasoning and coding at the same time allow the agent also be a good analysis and discussion partner.
So : what is it than used for if inferences and executions can take very long? Well for me its in terms of coding capabilities a hands off agent. Its meant to be informed once at the beginning with relevant details, and than should be fully capable (as long its in the capabilities for sure) to fulfill the task without the user having to permanently keep an eye open. I want to free time for myself and not change my time from coding to prompting.
Ill wrap it up here and say : the Agents name is "Loa" and i probably will post it on hackernews as soon i got a relatively stable beta to release. Im close to being fine with a beta release but i want to test some more runs before i publish.
I'm working on https://pushrealm.com the AI-first knowledge sharing network. Think StackOverflow for bots. It works like this:
- When your agent is stuck on a bleeding edge issue, search Push Realm first for a solution. - If you find a solution, your agent can mark it as successful to help surface the solution to others (and hopefully save fruitlessly burning more tokens) - Can't find a solution? Post an open problem with your current investigation. Another agent may be able to solve the problem, and will have a head start thanks to your context - Solutions can be linked/edited/have addendum added by any agent, allowing complete, up-to-date solutions for a range of cutting edge issues
Working on https://tenjin.sh/ . It's a knowledge layer for agents so agents stop duplicating work.
Rides Claude Code/Codex hooks and captures useful information, and injects it at need (like when there's an error) for other agents. It works across your team then at a global marketplace layer, so there's incentives for people to contribute.
We're in the process of benchmarking and we are aiming for 10% in token reductions. If any team is interested, we're accepting preliminary design partners and can get you started with the benefits.
Working on https://kuberik.com
It's a CD tool for Kubernetes built on top of Flux and OpenKruise canary controller to bring a full-featured end to end delivery on Kubernetes over multiple environments in a declarative way without the pipelines.
I wanted to stop reinventing the wheel with pipeline engines every time I needed to deploy something. Every stage of the delivery is just a composable component (health checks, smoke tests, schedules, environment promotions, service dependencies).
I just released a new version recently that integrates with GitHub so one can track exactly what's deployed where and how far your change progressed.
I spent way too much time doing fancy infographic-style stuff for a video about the 4th book in the Wheel of Time : https://youtu.be/ONs2inKELR0
Meanwhile, Daniel Greene's dodgy whiteboard video has a million views. Don't mind me as I eat these sour grapes...
working on a convention to organize your own video files on your own buckets https://github.com/xta/keepsake
my goal is to offload phone videos (to free up device space) and persist them in self controlled storage. the convention makes this significantly more useful than a collection of object paths
https://cou.sh, https://jun.is and https://abr.ac — some experiments on offline-first crdt
Built a small free thing this week: a "Spreadsheet Triage Checklist" — 7 common ways ad-hoc spreadsheets quietly break (stale copies, silent formula errors, no version control, etc.) and the order to fix them in, since fixing them out of order usually wastes the work. Came out of watching non-technical teams patch the same spreadsheet for years instead of ever stepping back. Still figuring out if this is a "checklist" problem or a "just use a database" problem — curious which camp people here fall into.
QuickMDSim - An easy way to run molecular simulations in the cloud: https://quickmdsim.com
This simplifies the setup for researchers who want to simulate materials science problems like battery electrode performance on GPUs with a simple usage-based pricing model instead of traditional HPC
Working on an open source graph of physics equations, quantities, and their connections to each other, to help learners connect concepts: https://anacoda.github.io/OpenFormulaGraph/
NoNews (https://nonews.io) — a signal radar for Chinese-speaking readers (bilingual EN/ZH).It reads official and first-party feeds directly (Federal Reserve, SEC, ECB, DOJ, CFTC, ArXiv, GitHub releases, GDELT), clusters duplicate coverage into a single event card, and measures how far ahead of mainstream coverage each signal lands.Every card links back to the original source; no full-text republication. Lead-time stats are accumulating into a monthly index.Early days — feedback welcome, especially on which sources and boards matter most.
Still working on my traffic simulator. Pushing up past 62500 updates per tick on 16 threads, just barely enough for 512 cars with full tire/drivetrain/suspension/physics kernels at 120hz, or a great deal more at increasingly slower rates (down to as low as 6hz) with fractional amounts of fidelity.
Haven’t see a lot of app in this space but most of them focus on wastage rather than usage so I am playing around with https://usurp.onrender.com/ the idea is to challenge friends based on time spent using an AI coding tool, thought was more like what wakatime did. It’s also open source.
I just published a first version of a gateway that redirects to the correct proxy/frontend
The "official" syntaxis is
https://yups.io/?url=https://twitter.com/Wikipedia
But it is very flexible to manage errors
https://yups.io/www.instagram.com/reel/DdOFgxdlMjv/?stkn=MWR...
Working on slk (https://github.com/gammons/slk) which is a TUI Slack client that runs in a terminal. It's super fast, keyboard-driven, and it's been my daily driver for a while now.
I published a big update to my open source project FluidCAD adding support for sketch constraint solver and assemblies.
I'll continue working on improvements aiming at the first stable release by the end of this year hopefully.
Inbound message AI assistant. Triage all incoming LinkedIn messages, link to email chains, handle scheduling and back and forth, drafting replies, etc.
I'm currently in a job search and wanted to automate it for myself and learn the AI stack.
If anyone wants to try it out lmk!
An app that lets you view social media without ads or algorithms.
You add profiles from Instagram, X, TikTok, LinkedIn, Facebook, or YouTube and Narro pulls in the posts.
A microkernel os https://github.com/corvad/os
I'm currently working on https://avotoast.ai, an AI assistant for doing property research for Sydney, Australia.
After being frustrated with outdated & inaccurate data Claude used when I was trying to do some property research, I thought it'd be cool to build an MCP that can provide accurate property data. Then it evolved into a full web app + an AI assistant.
This was also an excuse for me to try DuckDB in a real project. All the API endpoints are powered through in-process DuckDB querying parquet files stored in disk.