返回首页
最新
Hi all,<p>Over the weekend I felt nostalgic for classic anthology-style storytelling and wanted to see if I could create something new in that format. Rather than trying to imitate any specific show, I was interested in the broader idea of short speculative stories built around irony, choice, and unintended consequences.<p>I decided to experiment with AI as a storytelling tool. Going in, I expected the results to be fairly mediocre, but I was genuinely surprised by the output. Some of the stories — and even the generated images — were better than I anticipated and made me want to explore the idea further.<p>The result is Twisted Logic, a small choose-your-own-path anthology story generator. It can use Google’s Gemini models if you provide an API key, but I’ve also been working to make it function with free alternatives and allow people running local LLMs to point the project at their own models. By default it uses free generators and the browser’s built-in voice (which can be turned off).<p>The project is free to use and open source (<a href="https://github.com/anefiox/TwistedLogic" rel="nofollow">https://github.com/anefiox/TwistedLogic</a>). I mainly built it as a hobby experiment and a way to explore generative storytelling and interactive narrative design. If anybody wants some links to the some stories generated as ebups please let me know.
Show HN: Mygs - an Live Preview and Online Code Editor for HTML
I built Prompt Pilot, a browser extension that enhances your prompts with one click. It works like Grammarly but for AI prompts - adds context, structure, and clarity so ChatGPT, Claude, Gemini, etc. understand what you need.<p>Key features:
- Works on any AI platform (ChatGPT, Claude, Gemini, Perplexity)
- XML/JSON output modes for structured prompts
- Privacy-first: prompts enhanced but not stored
- Free tier: 3 enhancements/day<p>Available for Chrome and Firefox. Would love feedback from the HN community!
I built Zone, an iOS app blocker using Apple's Family Controls API. The differentiator is simple: it counts how many times you try to open blocked apps.
Most app blockers just block. But I found the attempt count more revealing than the block itself. Seeing "you tried to open Instagram 47 times today" was a wake-up call I didn't get from blocking alone.
Technical notes: Family Controls API is poorly documented but provides system-level blocking that actually works (unlike overlay approaches). Had to handle some edge cases around authorization persistence and count tracking across app restarts. The API requires Screen Time permissions which adds friction to onboarding but ensures reliable blocking.
One interesting discovery: users seem to prefer seeing raw attempt counts over gamified metrics (streaks, badges, etc). Less is more for this use case.
Built in SwiftUI, local storage only, no subscriptions. Took about 3 months part-time.
Curious if others have worked with Family Controls API and what challenges you faced. Also interested in thoughts on digital wellness apps in general - does tracking behavior change it, or just make you more aware without actual change?
Hi everyone I saw zig is and intresting language I am learning it and also making a transpilied High level language over it I want some help with developing syntax.
in my lang there are three types of var declaration
1) using local keyword this are added in arena of the specific function.
2) using let keyword this are on stack but i am finding solution to make strings easier here.
3) manual memory but my transpiler will automatically use defer keywords so they are safe and delete once block exit
4) using unsafe direct fully manual memory management but still my transpiler will not let compiler till once in code the are freed.
AI memory systems often become a black box. When an LLM produces a wrong answer, it’s unclear whether the issue comes from storage, retrieval, or the memory itself.<p>Most systems rely on RAG and vector storage, which makes memory opaque and hard to inspect, especially for temporal or multi-step reasoning.<p>An alternative is to make memory readable and structured: store it as files, preserve raw inputs, and allow the LLM to read memory directly instead of relying only on vector search.
I recently used Deepseek and when sending another request in "Thinking" mode initially showed activation of "reading" mode, I sent a regular text request without documents so I don't know what that means. Well, I suspect that this is a deeper understanding of the user prompt.