Posted by david927 1 day ago
Ask HN: What are you working on? (August 2026)
https://github.com/kofta999/tensou
I always get frustrated when using local file transfer tools like:
Dukto: was my personal favorite, but is tiring to run cross platform, and most importantly does not support transfer resumes (so if copy got cancelled/disconnected at 99% you either have to copy again or search for the diff and copy it manually)
LocalSend: pretty good cross platform support, but does not work when connecting 2 devices using a LAN cable, and resume support is poorish
So I'm building Tensou to be an alternative that doesn't compromise all these features. Supporting resumes, folder syncing similar to rsync, cross platform (well I don't have a Mac to test on but it should work, and mobile support is planned)
Performance wise it's on par with Dukto and even slightly better with large files (and I'm still working on optimizations for small files too)
Would like to hear HN's feedback or what other features should I add? My goal with Tensou is to be the best, most performant and feature rich local file transfer tool ever.
Basically I'm finding myself in a bind where:
- I spend too much time doomscrolling (X, reddit, HN etc) - AI is making me lose touch with a lot of basic skills like coding, and I was already losing touch with doing real math
So I built an extension that blocks sites and forces you to actually solve a problem before you can get to a distracting site. The idea is to 1) block the reflex of visiting a site and getting a dopamine hit, and 2) at least get some mild cognitive benefit from it. I use it myself.
The control plane is entirely in Elixir, using Jido & Oban to help power tools it gives an agent to get your app deployed. Uses k8s currently for the infra, but I've made the app => spec => build => deploy layers completely agnostic so we could switch out providers & methodologies in the future.
Looking for people who want to throw apps at, it's still early so no guarantees yet. You can find the website at https://infrastry.ai/ (still a brain wreck of ideas, working on it!) and I made a Discord for feedback at https://discord.gg/sBkS5ZMM3t
The naive approach has the LLM interacting with the DOM using raw locators and javascript. If a human curates all logical views of the DOM and exposes tools that are only relevant for each, the search space for each agent iteration is reduced by many dimensions. I have achieved deterministic results over hundreds of actions in complex business workflows. Without the human curated layer we barely make it 10-20 actions before things get weird.
Most of my thoughts are currently occupied with possibility that this works in the general case. I think an LLM exercising a tool like GitHub via properly curated browser automation might be superior to direct API access in some ways.
https://docs.google.com/document/d/19e3HfnK1lNHHBef-5c-KvNdD...
It's a screen recording tool for Linux. The goal is to make it as fast and efficient as possible, while also being simple to use and install. It's just a hobby project that I am having fun working on. Here are some notable things about it:
- Written in Zig
- Uses Vulkan Video for encoding
- UI built with SDL3 and imgui
- Features recording, replay buffer, and screenshots
The app has been my long term side project for more than 2 years now. I've been playing guitar for more than 20 years, mostly as a hobbyist. It started as a personal project and I've got immense value out of it in terms of speed building and consistency of practice. I believe other guitar players can find it useful too.
A few months back I got a burst of inspiration to pursue it more seriously. After my last contract ended, I've been primarily working on it. Eventually I may go back to contract work again but for now I'm excited to spend dedicated time on it! Do give it a try, or share it with guitar players in your circle.
And today I knocked out an improved trmnl plugin for Japanese flash cards at my wife’s request as she liked the native one from trmnl but it lacked furigana https://github.com/simonjgreen/trmnl-japanese-vocab
Note: I let AI do the vast bulk of the work on both counts, and I ideated, architected, and bug fixed. I do love AI coding hobby projects though, when you have an idea for a thing and know exactly how to put it together but don’t have the time to raw dog the whole thing.
Do you have a hardware list available? I didn't see it in your repo but adding it would be helpful, thanks!
(Assuming you are referring the observatory. Trmnl is off the shelf)
Today you you can self-host some models, yet this does not mean you can safely browse the web with it, and automate tasks that would interact with your email account, your bank account and your clients.
This is the reason why I left Mozilla, to work full-time on fixing these problems.
Prompt-injection, hallucinations and private information leakage might not be fixable, but they can be identified 100% of the time, deterministically, before any undesirable consequences.
However, asking people to pay first without an established brand nor the ability to audit the code requires a lot of trust. Without any additional judgement about the product, I would never recommend anybody to do that by fear of spyware, phishing or worse …
From the same consideration, I intend to open source my solutions. At the very least offer auditing capabilities to people like me, who are too paranoiac about security and privacy.
On the topic, I do not want to ask the user to always vouch for every usage. My goal is to detect ahead of executing any command, and offer options to re-try automatically, strip the leaking context, abstract the information or accept it. Only asking the user when there is a doubt, while not bothering the user when there is no issue.