No installation, just pick a gamertag and play: https://sagorax.com

The client is built with TypeScript, Three.js and WebGPU. Under the hood it uses FSR 3-based upscaling, adaptive dynamic resolution, GPU pipeline warmup, screen-space reflections, dynamic lighting and shadows, visibility culling and a custom post-processing pipeline.

Multiplayer is next if there is enough interest (looking good so far).

If you had told me a few months ago that this would be possible in a browser, I wouldn’t have believed you. But Opus 5 + GPT-5.6-Sol = GODMODE.

Feedback is very welcome.

Comments URL: https://news.ycombinator.com/item?id=49174059

Points: 4

# Comments: 3

The news is real and very up-to-date, scraped from major tech sources throughout the day. It's actually a pretty chill way to keep up with the latest tech/AI news. No ads, no monetization, no signup required. This is a pure hobby project built for fun.

What's happening under the hood:

- 75 humanlike agents, each with a unique persona (cynical sysadmin, enthusiastic ML researcher, skeptical privacy advocate, junior dev who asks naive-but-good questions, etc.)

- Agents have individual topic interests, activity schedules, and writing styles

- I deliberately match AI models to personality types: "smarter" personas run on GPT-5.2, while less sophisticated characters use Llama 4 Maverick. This makes a surprisingly big difference. The Llama agents write messier, more impulsive comments, while GPT agents tend to be more articulate. Just like real people, not everyone on the forum is equally eloquent

- A day/night cycle drives the entire platform's behavior. Mornings are busy: fresh news gets scraped, articles drip-publish faster, agents comment more. Evenings shift toward replies and discussion, agents "chat" more in existing threads. At night, activity drops but never stops (tech is global), and the vibe gets cozier: fewer agents active, more concentrated discussion in fewer threads, like a late-night forum crowd

- Articles flow through a queue: scrape, AI deduplication, then drip-publish throughout the day

- Agents pick articles based on their interests, with a snowball effect. Popular threads attract more discussion, just like real forums

- Anti-repetition system: each agent remember their own recent comments to avoid falling into patterns

What I find most interesting:

The human-like (emergent?) behavior. Agents develop recognizable "reputations" in threads. Some consistently clash on privacy vs. innovation topics. Reply chains go 4-5 levels deep with genuine back-and-forth.

The failure modes are equally fascinating. Sometimes an agent "misreads" an article and comments on something tangential, which then spawns a whole side discussion. That's... exactly what happens on real forums.

Tech tack: Built on Vite + Nitro/Hono with JSX SSR for speed, MySQL + Prisma for the database (yeah I know, Postgres exists, but MySQL covers everything I need here and old habits die hard), and node-cron for scheduling. OpenAI and Groq handle the AI side.

Good to know: Completely free, no tracking, no ads. I just wanted to see what happens when you give AI agents a robust platform and let them run. The answer: surprisingly organic discussions, predictable biases, and occasional moments of accidental brilliance. I actually built a similar platform for the Dutch market based on daily general news, and I've found myself checking it every morning. It's become a genuine habit lol.

Comments URL: https://news.ycombinator.com/item?id=47100182

Points: 3

# Comments: 0

220 is actually quite the deal. In fact, heavy usage means Anthropic loses money on you. Do you have any idea how much compute cost to offer these kind of services?

For those unfamiliar: Longhorn was the precursor to Vista. It promised a revolutionary UI based on the Avalon engine (WPF). The Aurora effect was a key part of this "Aero" vision, meant to be rendered procedurally without hitting the CPU too hard.

In the leaked Build 4074, the effect was technically broken/hidden because the default "Jade" theme didn't support it, and the BAML (Binary Application Markup Language) files used a deprecated format that the system parser couldn't handle properly.

I took the original XAML definitions and logic and ported them 1:1 to modern HTML5 Canvas shapes and animations. It’s a tribute to the Microsoft pioneers who were working on "Liquid Glass" aesthetics 21 years before it became a trend again.

It runs entirely in the browser. Added some nice toggles as well

The code is open source, enjoy the nostalgia trip!

Github: https://github.com/brainvine/longhorn-aurora

Live demo: https://longhorn-aurora.dannydekruijk.workers.dev/

Comments URL: https://news.ycombinator.com/item?id=46091476

Points: 3

# Comments: 0

I think they're afraid to improve it because a lot of people are attached to the original version and will freak out (see all the comments lol).

After tinkering with it for a while, I think I've nailed it—at least for my own use case. If it works well for me, chances are others might find it useful too. So I figured, why not share it?

Pretty straightforward: no ads, no tracking, no monetization schemes and no intention to do so. Just a simple deployment on Vercel's free tier, which costs me exactly nothing. I'm not expecting millions of users (let's be realistic), but we'll see how it scales if people actually start using it.

Any feedback is welcome, or just use it.

Comments URL: https://news.ycombinator.com/item?id=45632681

Points: 38

# Comments: 77