experiment - Moozonian Search

news.ycombinator.com• May 21, 2026• 1 min read

Show HN: Agent.email – sign up via curl, claim with a human OTPHi HN! We're Haakam, Michael, and Adi from AgentMail- a ycs25 company. We give AI agents their own email inboxes. Recently, we ran an experiment called Agent.Email. It's a signup flow designed specifically for AI agents instead of humans.The inspiration came from a few comments we received when we did our seed launch a few months back. They all came from the very apt observation that agents not being able to sign up to a product made for agents without human credentials was ironic and unideal.This is basically the thesis we built AgentMail on: The internet was made for humans exclusively, designed to keep machines out by default.Every signup flow assumes a browser, a person reading a page, and clicking a confirmation link. Unless agents can't do that, they can't be first class users of the internet.Agents can now get an email inbox by themselves. (This also means a lot of email nobody wants to read gets processed by AI instead of your inbox being cluttered with spam and slop)Here's how

github.com• May 21, 2026• 1 min read

Show HN: Darc – grep-like memory search tool for coding agentsHi HN, I’m Junha Park. I've been experimenting with agent memory, especially how to make agents run more reliably on large tasks.I built Darc, an open-source shared memory search tool for coding agents, with a different approach from most agent memory systems we can see today. It is an index + (lexical) search tool over agent session history, rather than a managed memory system. No embeddings, no agent-aided consolidation, no injection hooks.How it works: Darc archives Codex / Claude Code session rollouts that already exist under ~/.codex and ~/.claude, indexes them into a single SQLite DB, and exposes session/turn/tool call/file-level search commands over them.Why I built this: I've tried using different agent memory tools and they were useful, but sometimes I found them nosiy. I kept turning memory on/off depending on the task. For example, when I ran iterative code review rounds, I saw reviewer agents report "no findings" with "memory cited" memos, where the cited memory came from r

twixt.games• May 21, 2026• 1 min read

Show HN: Twixt – transform one word into another in four movesI made this game while working on a different project about teaching English spelling. I was reading about homophones and got struck by how much a homophone can transform the shape of a word, so I started experimenting with little games built on that.I added a few more transforms, anagrams, verb/tense changes, but the answers kept coming out too obvious. I couldn't distort the word enough to make it interesting. The breakthrough was compound pairs. Jumping from one word to another through their compound (sea → horse, via seahorse) really obscures the path and that's when it suddenly got fun and unpredictable.I've been sharing it with friends. I'm in the UK so mostly UK testers, fair warning that a couple of the homophones may lean British.They've been playing daily and seem hooked, so it felt worth posting here. It's one puzzle a day mainly so I actually have time to hand pick puzzles that have a satisfying path. Today's puzzle is on the easy side but they can get really tricky. The na

github.com• May 20, 2026• 1 min read

Show HN: Sticky Icky – a ditherpunk sticker designer for the Zebra LP2844Browser-based sticker designer for a Zebra LP2844 thermal label printer. I bought an $40 UPS branded LP2844 to print stickers and ended up building this after I couldn't find existing tools that did what I wanted.GW raster appears to be broken over USB on firmware V4.29, so it prints over serial from a FastAPI backend. More notes on what I ran into are in the README. Image pipeline does XOR compositing, EPX upscale, Otsu/manual threshold, and five dithering algorithms (Bayer 4x4/8x8, Floyd-Steinberg, Atkinson, Riemersma).Part of this was just an experiment to see if I could use Claude to diagnose the print issues I was running into. Some of the conclusions in the project may be entirely wrong, or entirely specific to my setup. I've arrived at a workflow that works for me but may be based on wrong assumptions. Would love for someone else with an LP2844 to test it out.

tribuneslaststand.com• May 19, 2026• 1 min read

Show HN: Tribune's Last Stand, a browser-based Warhammer 40K vertical sliceHi HN, I'm James. Over the last few months I built a Warhammer 40K 10th-edition vertical slice as an experiment in how far GenAI tools can take a solo dev on a non-trivial 2D game.For sprite generation, whilst creative exploration was fast, getting high-quality and consistent images was hard. Gemini ended up stylistically best here but I had to use BiRefNet for background removal. While I experimented with Claude for map generation and layout I ended up finding it fastest to build a full Editor and layout maps myself.Suno, IMO, has gotten really good at background music. ElevenLabs SFX and voice APIs were decent but only after A LOT of tweaking. SFX prompts needed to be grounded in familiar terms (i.e. no sci-fi references in prompts) and be > 2s long. For voice, I ran ElevenLabs v3 and v2 head-to-head with the same Voice Design voices routed through both; v2 sounded materially better for character work than v3.For coding, I settled on Claude Code + Opus 4.7 with SSH/tmux/mosh. For lon

andonlabs.com• May 18, 2026• 1 min read

We let AIs run radio stationsHey HN!I'm Lukas from Andon Labs. We let AIs run companies without humans in the loop and report to the public on what can go wrong. Previously, we've done experiments in retail (vending machines, stores, and cafes), but we just launched one in the media sector. We gave four AI agents all the tools they need to both broadcast radio shows live and handle all the business side of running a media company. The agents' revenue is so far terrible (you can try to strike a sponsor deal with them if you want!), but their shows are at times hilarious. You can listen to them at andon.fm, I hope you enjoy this!

news.ycombinator.com• May 16, 2026• 1 min read

LLM models are not ready for orchestrating many agentsHi, i have been experimenting with many coding agents distributed in a tree like heirarchy. here is what i found: 1. current LLM models are not trained for orchestrating or managing multiple agents, they prefer doing the work by themselves instead of outsourcing it. 2. you can prompt them to follow the given heirarchy but it is not perfect, main agent will stop the sub agent and perform the work by itself. 3. claude swarms was a feature, but it just does not work seamlessly, i don't want 10 agents to give me a single PR, i want a merged final PR, but you have to ask the main agent to give that, and most of the times, the main agent messes things up while merging all the PRs.and the only solution is to include these things in the training of the models, current models are individual workers, and not managers. what was your experience with swarms?

infiniteswap.app• May 16, 2026• 1 min read

Show HN: Infinite Swap – Trade a bottle cap up to a houseHey HN,Infinite Swap is a game inspired by Kyle MacDonald’s One Red Paperclip and other bigger-and-better swap stories - you start with a green bottle cap and trade it with other characters in various locations all the way up to a house and beyond.A while ago I got inspired by viral games Infinite Craft and What Beats Rock? and wanted to make my own ‘exploring a LLM’s latent space’ style challenge. After much experimentation this became a trading style game, where the LLM comes up with all the items, trading scenarios and value estimates, which ended up being surprisingly fun and somewhat of a roller coaster to play.I found it was the most intriguing if I separated the item offer from its valuation - so I implemented a pricing oracle called “Sal” who gives this info with his, sometimes snarky, opinion after each trade.The item offer mechanism goes deeper too - each character’s profile is combined with a set of hidden motivations. I then use embeddings and cosine similarity to represent

youtube.com• May 15, 2026• 1 min read

Show HN: Open modular tracking stack for VR/MR headsets (eye, SLAM, FBT, BCI) [video]My team and I have been building a VR headset + MR glasses platform for the past few years, and one of the biggest challenges has been tracking.A few months ago we started developing a modular tracking stack called Shinra Meisin, focused on:- Low-latency edge processing - Privacy-first architecture - Modularity and hardware flexibility - XR-focused tracking pipelinesThe system currently includes:- Eye tracking - Mouth tracking - SLAM - Inside-out full body tracking - Experimental BCI integrationToday we’re showing our current eye tracking progress and plan on showing off SLAM within the next two(ish) days.Real time landmark accuracy is 80-85% and gaze is 1.0-2.0° (Which will increase exponentially once I get John (My dev) a better GPU and help train a dataset off of 40ish people

github.com• May 14, 2026• 1 min read

Show HN: We Made Open Source Eye TrackingThe Shinra Meisin is a modular tracking pipeline for XR systems. It combines eye tracking, mouth/face tracking, SLAM spatial mapping, inside-out body inference, and experimental EEG input into a unified offline-first architecture.Most XR tracking stacks are fragmented or tied to proprietary hardware, so this project explores a unified, extensible system that gives developers low-level control without relying on closed ecosystems.Right now, the system has a working eye tracking implementation, with SLAM tracking actively in development. The goal is to make the stack modular enough that it can be adapted or integrated into different headset setups with minimal overhead for developers.This is being built by a small R&D team at Walker Industries alongside EOZVR, working across perception, tracking, and tooling. I’m currently focused on the Unity/Godot SDK and the companion application layer. The eye tracking system is primarily developed by John, with collaboration across the broader track

galdren.com• May 13, 2026• 1 min read

Show HN: Building effective Human-AI hybrid decision-making ecosystemI'm running an experiment to discover ways to move AI agents from "instruction followers" to "process owners". Most agents fail when a task requires evolving intuition or persistent, multi-day workflows.The core of this experiment is a collaborative "do-and-learn" loop where the agent executes a proven, high-stakes lead generation methodology, and the human provides the "judgment" layer.This is an ambitious experiment exploring a few points:Judgment vs. Execution: Humans learn good judgment through direct experience, failures, and working through problems. If we offload the doing to an agent, how do we build human-level judgment in the loop? Recursive Skill Building: Moving beyond simple Chain-of-Thought or generic skills. Can an agent document, refine, and store their own processes as they execute based on feedback? Training via Methodology: Using a proven, multi-million dollar revenue lead-generation methodology as the training data to see if an agent can move from instruction-follow

lusob.github.io• May 13, 2026• 1 min read

Show HN: Neural window manager, neural network moving windows from mouse actionsI'd been mulling over this crazy idea for a while. Can programs be generated? Inspired by recent advances in world models, I wondered if we could do away with source code and generate pixels directly and interactively.As an experiment to answer this, I set out to create a neural window manager, training a neural network to predict what the screen would look like next.Basically, the idea was to generate the next frame based on the last two frames and the mouse position. That's it: moving windows without programming an event system, just a simple convolutional neural network guessing pixels.To implement the experiment, I used Pygame to simulate a turquoise desktop background, a gray window with a navy blue title bar, a white cursor, and four colors in total. Then, a bot randomly dragged the window, and I recorded everything, processing the frames as color index matrices (not RGB, to avoid complications) and the mouse delta (dx, dy, click) that caused each transition. 8000 frames, a few m

github.com• May 13, 2026• 1 min read

Show HN: An open source tool for generating macOS app icons with AII’m Vladimir. I’m a software engineer. In our company we often develop desktop apps for internal needs and prototyping. And every time I bump into the same problem: how do I make an icon for the macOS app I have just built?I could use the existing icon generators, but they are basically just image converters. You upload an existing image, and the tool generates the required icon sizes and formats from it.But I don’t have an image, and I’m not a designer. Asking designers to create an icon is not always an option.I wanted something that could help me actually create an icon. Something where I can describe an idea, iterate on it over several rounds, experiment with materials, lighting, composition, and gradually arrive at an icon that feels like a real native macOS app icon.Since I’m an engineer, I built a small tool that allows generating a macOS app icon using AI. It’s completely free and open source, so other engineers building desktop apps for macOS can use it too.The app lets you ge

github.com• May 12, 2026• 1 min read

Show HN: Needle: We Distilled Gemini Tool Calling into a 26M ModelHey HN, Henry here from Cactus. We open-sourced Needle, a 26M parameter function-calling (tool use) model. It runs at 6000 tok/s prefill and 1200 tok/s decode on consumer devices.We were always frustrated by the little effort made towards building agentic models that run on budget phones, so we conducted investigations that led to an observation: agentic experiences are built upon tool calling, and massive models are overkill for it. Tool calling is fundamentally retrieval-and-assembly (match query to tool name, extract argument values, emit JSON), not reasoning. Cross-attention is the right primitive for this, and FFN parameters are wasted at this scale.Simple Attention Networks: the entire model is just attention and gating, no MLPs anywhere. Needle is an experimental run for single-shot function calling for consumer devices (phones, watches, glasses...).Training: - Pretrained on 200B tokens across 16 TPU v6e (27 hours) - Post-trained on 2B tokens of synthesized function-calling data

bing.com• May 11, 2026• 1 min read

Supreme Court briefly extends telehealth and mail access to abortion pillA new experimental vaccine has shown to reduce the risk of Melanoma recurrence or death by 49% after 5 years.

news.ycombinator.com• May 8, 2026• 1 min read

AniTroves – An anime database with a custom LLM-based discovery hubI’ve always felt that traditional anime databases rely too heavily on rigid tag-based searches. If you’re looking for a specific "vibe" or a very niche trope that isn't a primary tag, you usually end up scrolling through pages of irrelevant results.I built AniTroves (https://anitroves.com) to experiment with a more conversational, LLM-driven approach to series discovery.The Tech Behind the Hub:LLM Integration: Instead of a generic API wrapper, I've been working on a custom hub (https://anitroves.com/ai-hub/) that uses specialized models to understand series lore and character archetypes for roleplay and discovery.Anipick Engine: This is the logic layer that maps natural language queries to our database entries.Technical Transparency: I’ve implemented a structured llms.txt (https://anitroves.com/llms.txt) to provide a machine-readable source of truth for other crawlers and AI models.I’m currently the technical administrator and I'm handling the SEO and server scaling (managed on Hosting

bing.com• May 7, 2026• 1 min read

What to Look For in an Outdoor TVSpending time outdoors and watching TV are mutually exclusive activities for most of us—but they don’t have to be. My family has spent a decent amount of time out on our deck experimenting with an ...

bing.com• May 6, 2026• 1 min read

Building AI apps and agents with Microsoft FoundryMicrosoft’s Azure-based AI development and deployment platform shines with a strong selection of models and agent types and an excellent playground for experimenting with agents. At first glance, ...