Hacker News: PetrBrzyBrzek

New comment by PetrBrzyBrzek in "Anatomy of the .claude/ folder"

PetrBrzyBrzek — Fri, 27 Mar 2026 16:26:41 +0000

Why is this AI slop article first on HN?

Show HN: Almostnode – Run Node.js, Next.js, and Express in the Browser

PetrBrzyBrzek — Sat, 07 Feb 2026 15:46:16 +0000

Hey HN, I built almostnode. It's a library that gives you a Node.js-like environment inside the browser.

It includes: - A virtual filesystem with POSIX-compatible API - 40+ shimmed Node.js modules (fs, path, http, crypto, stream, etc.) - A package manager that installs real npm packages client-side - Built-in Next.js (App Router) and Vite dev servers with HMR - Service worker bridge that makes virtual servers accessible via real URLs

The main use cases are interactive code playgrounds, live documentation, and AI coding agents that need to execute code client-side without provisioning a sandbox server.

It's not a WebContainers replacement, no real TCP/IP or native modules. Think of it as the lightweight alternative when you just need a demo or playground running from a static HTML page.

It's experimental and definitely has bugs. Would love feedback, especially on the module shims and framework support.

GitHub: https://github.com/macaly/almostnode

Comments URL: https://news.ycombinator.com/item?id=46924751

Points: 3

# Comments: 1

New comment by PetrBrzyBrzek in "Ask HN: Share your personal website"

PetrBrzyBrzek — Wed, 14 Jan 2026 19:14:07 +0000

https://petrbrzek.cz

New comment by PetrBrzyBrzek in "Anthropic acquires Bun"

PetrBrzyBrzek — Tue, 02 Dec 2025 21:15:16 +0000

I think it’s an acquihire, and they also like Bun.

New comment by PetrBrzyBrzek in "Claudeskills.cc – Share, Discover, and Reuse Claude/OpenAI Agent Skills"

PetrBrzyBrzek — Sat, 25 Oct 2025 06:53:40 +0000

AI Slop

New comment by PetrBrzyBrzek in "Improved Gemini 2.5 Flash and Flash-Lite"

PetrBrzyBrzek — Thu, 25 Sep 2025 18:14:47 +0000

It’s cheaper and faster. What’s not to understand?

New comment by PetrBrzyBrzek in "Show HN: I built an LLM chat app because we shouldn't need 10 AI subscriptions"

PetrBrzyBrzek — Sun, 13 Jul 2025 12:37:15 +0000

> Couldn't find one, so I built one.

I’m sorry, but I find it hard to believe that you didn’t find any. I personally know at least 5 services that offer this.

New comment by PetrBrzyBrzek in "How to post when no one is reading"

PetrBrzyBrzek — Mon, 02 Jun 2025 06:03:34 +0000

Yeah, I agree. I’m verified and even tried a higher tier, but it didn’t make any difference. The For You tab feels almost like scrolling TikTok – total brain rot.

My strategy now is to repost all content everywhere. But X consistently gives me the worst results.

New comment by PetrBrzyBrzek in "How to post when no one is reading"

PetrBrzyBrzek — Mon, 02 Jun 2025 05:51:49 +0000

I find this especially difficult on X, where almost no one sees my posts. Especially when I compare it to LinkedIn or Reddit, where it’s not that hard to reach thousands or even tens of thousands of views.

New comment by PetrBrzyBrzek in "Figma Slides Is a Beautiful Disaster"

PetrBrzyBrzek — Sun, 01 Jun 2025 18:53:59 +0000

Here is my last one https://macaly-presentation.vercel.app/

New comment by PetrBrzyBrzek in "Figma Slides Is a Beautiful Disaster"

PetrBrzyBrzek — Sun, 01 Jun 2025 16:39:28 +0000

I personally stopped using traditional presentation software altogether. These days, I build all my presentations using vibe coding tools. It might not be the fastest way to create a presentation, but it’s definitely the most fun. Plus, for animations, all you need to do is explain what you want, and most of the time, it just works. I even asked for speaker notes, and I had them ready in three prompts. Magic.

If you’re curious, I use Macaly for this (macaly.com).

New comment by PetrBrzyBrzek in "Show HN: Time travel debugging AI for more reliable vibe coding"

PetrBrzyBrzek — Tue, 04 Mar 2025 20:52:22 +0000

If anyone is wondering why it looks like Bolt, it’s because it’s using Bolt.DIY, an open-source fork of Bolt (https://github.com/stackblitz-labs/bolt.diy). The catch is that it's still using WebContainers from StackBlitz, so it's not really possible to run it commercially. You need to get rid of WebContainers and find something different.

New comment by PetrBrzyBrzek in "Show HN: Mastra – Open-source JS agent framework, by the developers of Gatsby"

PetrBrzyBrzek — Thu, 20 Feb 2025 07:29:24 +0000

I created a similar library for orchestrations, but it’s more explicit and lightweight. https://github.com/langtail/ai-orchestra

New comment by PetrBrzyBrzek in "Show HN: Simple demo tool for Gemini 2.0 Flash object detection"

PetrBrzyBrzek — Tue, 17 Dec 2024 13:24:35 +0000

Gemini 2.0 is the first model that can return precise bounding boxes. Other providers and models can't handle this.

Show HN: Simple demo tool for Gemini 2.0 Flash object detection

PetrBrzyBrzek — Tue, 17 Dec 2024 13:06:08 +0000

I built a straightforward demo to showcase Gemini's object detection capabilities. Upload any image, get bounding boxes and clean JSON output. It's currently the only model I've found that can accurately return object bounds.

Demo: https://langtail.com/gemini-bounding-boxes

The tool is pretty simple:

- Drop an image or use example ones (apples, llamas)

- Hit detect

- Get visual bounding boxes + JSON output

Happy to answer any questions!

Comments URL: https://news.ycombinator.com/item?id=42441038

Points: 2

# Comments: 4

Show HN: White Ninja – Conversational AI agent for prompt engineering

PetrBrzyBrzek — Tue, 10 Dec 2024 16:28:31 +0000

I built a free tool that approaches prompt engineering differently. Instead of writing prompts manually, you interact with an AI agent that:

1. Understands your intent through conversation 2. Generates optimized prompts based on your requirements 3. Battle-tests prompts against common failure modes 4. Iteratively improves based on test results

Technical details: - No registration required - Built-in testing framework - Handles context window management - Implements safety guardrails automatically

The motivation came from observing that manual prompt writing is error-prone and time-consuming. The agent helps bridge the gap between intent and implementation.

Demo video: https://langtail.com/prompt-improver Try it here: https://youtu.be/_56TpCmtYS4?si=HX_GSpVnxbdsfNYA

Feedback and suggestions welcome.

Comments URL: https://news.ycombinator.com/item?id=42378358

Points: 4

# Comments: 0

Show HN: PromptNinja – Test your prompts against adversarial attacks

PetrBrzyBrzek — Thu, 05 Dec 2024 21:34:29 +0000

I built this tool to help anyone writing prompts test them against various types of adversarial inputs and edge cases. It helps identify:

- Prompt injection vulnerabilities - Unexpected behaviors - Edge case failures - Prompt stability issues

Whether you're building AI app, using ChatGPT, or just want to write better prompts, PromptNinja simulates different attack patterns and gives you a battle score based on how well your prompt holds up against various challenges.

Try it here: https://langtail.com/prompt-ninja

Feedback and suggestions welcome!

Comments URL: https://news.ycombinator.com/item?id=42333027

Points: 3

# Comments: 0

Langtail 1.0 – Spreadsheet-like interface for testing LLM apps

PetrBrzyBrzek — Wed, 30 Oct 2024 08:52:38 +0000

Article URL: https://langtail.com/blog/introducing-langtail-1-the-best-way-to-test-your-ai-apps

Comments URL: https://news.ycombinator.com/item?id=41993025

Points: 15

# Comments: 3

New comment by PetrBrzyBrzek in "Use OpenAI Voice mode to practise for YC interview"

PetrBrzyBrzek — Thu, 10 Oct 2024 15:04:37 +0000

Thanks, I understand. This is a free tool and doesn’t have much to do with the main platform at the moment. Langtail is primarily for testing LLM apps.

New comment by PetrBrzyBrzek in "Use OpenAI Voice mode to practise for YC interview"

PetrBrzyBrzek — Thu, 10 Oct 2024 13:29:35 +0000

Hi, I'm a non-native English speaker, and YC interviews are known for requiring quick and concise answers. This is a much bigger challenge for me than for native speakers. The best way to improve is, of course, practice. Thanks to OpenAI's real-time API, I can train anytime, knowing it's just AI. It's far more comfortable for me to make mistakes this way, and when properly prompted, the LLM can be quite critical. By default, it tends to be overly optimistic, calling every silly idea brilliant. I'd appreciate if you could try it out. This app was created in just one day over a weekend. The images are generated using Flux 1.1-pro. Regarding the OpenAI key, it's only on the frontend and isn't sent to our backend.