Hacker News

I did not adopt AI coding tools because they were faster or cooler. When I started, it was a chore to use them.

I still have the cursor unlimited plan that gives me unlimited tokens. It expires in May after which Cursor is forcing all of us legacy users onto the new plans where you do have to pay for tokens. So May of last year is when I got my yearly plan, but before that I was paying monthly for a couple of months. It was October 2024 when I started.

I adopted them because I lost the ability to type.

Before all of this, in early 2024, I was curious about AI coding- hearing (probably fake) stories about people building production apps and getting revenue, and I used chatGPT here and there to clean up a function for me, but I never took it seriously.

I downloaded Cursor, tried it once, and abandoned it. It tripped over itself when I asked it to do a simple thing.

Then, two months later, I was using Cursor every day. A change that happened overnight.

It started when my right hand started hurting. I thought I had carpal tunnel.

Then my left hand started hurting too.

The pain got worse. I called my mom asking if I ever had chicken pox, wondering if it could be shingles. I was Googling symptoms nonstop.

Then the weakness started. At first it was subtle, clumsier hands, lower dexterity. My arms were starting to be difficult to lift.

A few more days passed and I could not open my front door anymore.

I was slowly becoming paralyzed. I went to the ER and I was diagnosed with Guillane-Barre Syndrome. (pronounced GHEE-YAWN-BAR-"eh?")

Thankfully, I ended up with a mild case. They let me go home after a week in the hospital. Still, it took months to recover the dexterity in my hands, and the ability to type.

People I work with from that time probably remember my sudden increase in voice notes, speech-to-text, and lots of typos.

When it came to coding, I switched from VSCode to Cursor overnight.

And I was doing it all through voice-to-text.

I tolerated the mistakes.

I rejected code it generated very often.

I got much better at prompting and organizing my thoughts.

I got better at dictating and enunciating so that speech-to-text would stop misunderstanding me.

It was my only option.

It took 6 months to be able to type "normally" again but I will likely never regain my previous speed and accuracy.

I still use Cursor today. Now, I will write code more often, but Cursor is still my primary IDE, and I still prompt more than code directly.

I've since used Claudecode and other tools, but I still like Cursor the best. I wonder how I will feel when when my unlimited tokens are gone. :)

Anyway, that's my story. Feel free to ask me anything.

Comments URL: https://news.ycombinator.com/item?id=47295954

Points: 2

# Comments: 0

Hello everyone! I am working on an open source reporting tool, that was mostly focused on the 'Embed analytics in your app' use case, which I found was either not great or not flexible or expensive, or all three!

However, I decided today to use this library wrapped in an app that makes it work like Metabase (and I use 'like' in its broadest sense here as it is quite early in its life). I have pushed an initial version live this weekend, and am looking for input to help prioritise features that close the gap with Superset / Metabase that would stop actual users using it. I want to avoid adding things that are not necessary. For now it only connects to postgres, but I will add lots of other providers (it is bound atm by databases that Drizzle supports).

I am not targeting big enterprises, more small teams / startups that just want a really user friendly and flexible reporting tool - that includes a simplified agentic analysis workflow like hex.tech (bring your own LLM keys). If anyone has the time to take a look and provide any feedback that would be hugely appreciated! There is a cloud option which for now is free, so if you aren't comfortable running it yourself locally (it only needs docker and a single container), you can also try that and let me know of any feedback. Nothing is paid for now but I was considering that at a v low cost level - e.g. €10 per month - to cover hosting, it is very lightweight and I dont store any data.

The github is here: https://github.com/cliftonc/drizby - I am really looking for input on the roadmap and if this is useful for others outside of my original use case (it is already actively being used in that way by myself and others).

Comments URL: https://news.ycombinator.com/item?id=47295948

Points: 1

# Comments: 0

Article URL: https://twitter.com/alexwg/status/2030217301929132323

Comments URL: https://news.ycombinator.com/item?id=47295947

Points: 1

# Comments: 0

Article URL: https://xcancel.com/0xmitsurii/status/2030451168678457766

Comments URL: https://news.ycombinator.com/item?id=47295946

Points: 3

# Comments: 0

I built Stardial, a customizable terminal clock written in Rust.

It started as an experiment to create a more flexible alternative to tools like tty-clock. Stardial supports animations, color themes, and layout customization so it can fit into different terminal setups and ricing environments.

Features: - Multiple display styles - Custom colors and themes - Animation effects - Adjustable size and layout - Designed for modern terminals

The default theme is designed to look good out of the box, but everything can be customized.

GitHub: https://github.com/USERNAME/stardial

Comments URL: https://news.ycombinator.com/item?id=47295939

Points: 2

# Comments: 0

Article URL: https://github.com/garydevenay/emporion

Comments URL: https://news.ycombinator.com/item?id=47295937

Points: 1

# Comments: 0

Article URL: https://github.com/microsoft/hve-core

Comments URL: https://news.ycombinator.com/item?id=47295935

Points: 2

# Comments: 0

Article URL: https://grimridge.net/blog/solving-compaction-with-lobotomy/

Comments URL: https://news.ycombinator.com/item?id=47295929

Points: 1

# Comments: 0

Article URL: https://jonathan-frere.com/posts/reactivity-algorithms/

Comments URL: https://news.ycombinator.com/item?id=47295924

Points: 1

# Comments: 0

Article URL: https://github.com/ammaarreshi/SkyRoads-Codex

Comments URL: https://news.ycombinator.com/item?id=47295921

Points: 1

# Comments: 1

Hi HN,

I got tired of trusting cloud services with my AI conversations — so I built the setup I actually wanted: encrypted disk, hardened OS, one-command deploy.

Demo: https://github.com/congzhangzh/your_openclaw/raw/main/demos/...

The idea is simple: if you're going to self-host AI, do it right — from the bare metal up.

Layer 1 — The disk: LUKS encryption + Btrfs compression (or ZFS native encryption). AI logs, API keys, model configs — everything at rest is encrypted. Someone pulls your disk? They get nothing.

Layer 2 — The OS: Debian Trixie. Stable, predictable, full toolchain. No surprise updates breaking your gateway at 3 AM.

Layer 3 — The container: Docker with Tini as PID 1 (proper signals, no zombies). Data lives on the host as plain files (~/.openclaw) — ls, cp, rsync. No opaque volumes.

Layer 4 — The gateway: OpenClaw with token auth + device approval. Connect Telegram and more. Guided onboard walks you through everything.

The whole setup:

git clone https://github.com/congzhangzh/your_openclaw.git && cd your_openclaw ./shell That's it. `openclaw onboard` inside the container does the rest.

Built-in monitoring (btop, nload, iftop) in the container. Ctrl+P, Ctrl+Q to detach — gateway runs 24/7.

Repo includes VPS disk encryption guides and provider recommendations. MIT-licensed. I use this daily on a cheap European VPS.

Feedback welcome: - Is the layered security overkill or just right? - Are you encrypting your VPS disks? - What AI backends are you running?

GitHub: https://github.com/congzhangzh/your_openclaw

Comments URL: https://news.ycombinator.com/item?id=47295889

Points: 1

# Comments: 0

Article URL: https://adlrocha.substack.com/p/adlrocha-money-and-collateral-in

Comments URL: https://news.ycombinator.com/item?id=47295879

Points: 1

# Comments: 0

I am exploring ways to work with my PC that doesnt involve always sitting at a desk and typing with hands like a cave man.

Testing out using Wispr Flow and similar voice inputs -- seems to work fine for some use cases.

I also place the laptop on a treadmill sometimes and try to to get some research / browsing / work done. Mouse (trackball) and typing are the current weakest link.

are there decent handheld input gadgets that allow simple trackpad / click / scroll up&down / next&previous type of navigation and a push-to-talk voice input? I am looking at cheap remotes for FireTV stick and other streaming boxes that seem to have voice input -- anyway one could hack one of those to do our bidding and pair with a PC?

Comments URL: https://news.ycombinator.com/item?id=47295871

Points: 1

# Comments: 1

I build a lot of small AI tools locally, mostly on top of Ollama, and the thing I keep running into is that every session starts from zero. Whatever context I built up the night before, whatever the model learned about how I like things structured, whatever half-finished reasoning we were working through together, it is just gone when I open a new terminal.

For a while I was just manually pasting in context at the start of every session which is exactly as painful as it sounds. Eventually I built a small proxy that sits between my client and Ollama and tries to solve this. It embeds recent interactions, stores them locally, and injects the relevant chunks when a new session starts. It works well enough that I actually use it every day now, but I built it the way someone with no formal CS background builds things, which means I patched it into shape and I am not totally confident the architecture is right.

The part that still bothers me is scoping. I work on a few different projects at the same time and I do not want context from one bleeding into another. Right now I am managing that by hand, basically just keeping separate directories and being careful, but that feels like a workaround not a solution.

Genuinely curious what other people have landed on. Are you using a vector DB for retrieval, or plain files, or something MCP based, or have you just accepted that local sessions are stateless and built your workflow around that? And if you have solved the scoping problem cleanly I really want to know how.

Comments URL: https://news.ycombinator.com/item?id=47295866

Points: 1

# Comments: 0

Article URL: https://github.com/kivancadiguzel-design/Spadyum-Genesis/blob/main/README.md

Comments URL: https://news.ycombinator.com/item?id=47295864

Points: 1

# Comments: 0

Article URL: https://github.com/gcv/julia-snail

Comments URL: https://news.ycombinator.com/item?id=47295855

Points: 1

# Comments: 0

Article URL: https://notes.brooklynzelenka.com/Blog/Notes-on-Writing-Wasm

Comments URL: https://news.ycombinator.com/item?id=47295837

Points: 3

# Comments: 0

Article URL: https://substancewiki.org

Comments URL: https://news.ycombinator.com/item?id=47295726

Points: 1

# Comments: 0

Article URL: https://en.wikipedia.org/wiki/Therac-25

Comments URL: https://news.ycombinator.com/item?id=47295712

Points: 1

# Comments: 0

Article URL: https://github.com/NERVsystems/llm9p

Comments URL: https://news.ycombinator.com/item?id=47295708

Points: 2

# Comments: 1