Buy A Modem

Launch HN: Relvy (YC F24) – On-call runbooks, automated

Hey HN! We are Bharath, and Simranjit from Relvy AI (https://www.relvy.ai). Relvy automates on-call runbooks for software engineering teams. It is an AI agent equipped with tools that can analyze telemetry data and code at scale, helping teams debug and resolve production issues in minutes. Here’s a video: [[[https://www.youtube.com/watch?v=BXr4_XlWXc0]]]A lot of teams are using AI in some form to reduce their on-call burden. You may be pasting logs into Cursor, or using

Show HN: QVAC SDK, a universal JavaScript SDK for building local AI applications

Hi folks, today we're launching QVAC SDK [0], a universal JavaScript/TypeScript SDK for building local AI applications across desktop and mobile.The project is fully open source under the Apache 2.0 license. Our goal is to make it easier for developers to build useful local-first AI apps without having to stitch together a lot of different engines, runtimes, and platform-specific integrations. Under the hood, the SDK is built on top of QVAC Fabric [1], our cross-platform inference and

Best Open Source Offline AI Agent

With the rise of AI agents, I've been looking for an agent that can use a combination of local models or models that are hosted on servers I control.For the local models, the agent must be able to run completely offline. When using remote models, it must only send network traffic to my servers. No checking for "updates" or pulling any type of dependencies from elsewhere.When trying Opencode, it was a failure.See: https://github.com/Chetic/opencode-offline"

Show HN: I built a local coding agent using Apple Intelligence

Hi HN! I built an on-device coding agent called Junco, designed to explore what's possible using the AI (Apple Intelligence) you already have on your Mac.Junco is a ~9MB Mach-O binary written entirely in Swift using the LanguageModelSession API. It's primarily an exploration and learning exercise for me, but it's also exciting to see what's possible. A clear pattern emerged: deterministic scaffolding is critical to guiding a small model along. Whereas Claude Code can defer ta

Show HN: QVAC SDK, a universal JavaScript SDK for building local AI applications

Mythos, Glasswing, and the hardware disclosure problem nobody is discussing

Coverage of Anthropic's Claude Mythos Preview and Project Glasswing has focused almost entirely on software vulnerabilities. That is where the demos are and where controlled release maps cleanly onto existing disclosure practice. I have not seen anyone engage with the next obvious question: what happens when a Mythos-class model is given detailed hardware architecture documentation and asked to do a security-oriented review? My intuition is the hardware case is meaningfully worse, for reaso

Show HN: Clusterflock: An AI orchestrator for networked hardware

Hi HN!We built Clusterflock to solve our own headaches with managing AI agents across distributed setups, different VRAM and RAM allowances, and the need to easily try out new models.While we focus on infrastructure (we built this specifically for networked hardware) it does ship with a powerful mission runner (or orchestrator), which is multi-session and asynchronous.Here is what it does best:Hardware-aware auto-downloading: It profiles your networked hardware and automatically pulls down the b

Ask HN: Is it still worth making "Huge" Language Models for dev tools?

I just want to ask the frontier builders and developers who are working on the flagship models a few questions. Is it still cost-efficient and worth it to keep making huge language models, when smaller, specialized models should be enough?Meaning that, when a user is working in a codebase with a certain framework, should the agent/model also know the complete chemical composition of an element, world history, and other random facts? Or should it only know the related and needed things? For

Tq-KV – Rust implementation of TurboQuant that works on GGUF models

TurboQuant came out at ICLR on March 25. We tried every available implementation on GGUF models. None of them produced usable output. Perplexity goes from 5.18 to 3,556. The model starts mixing languages mid-sentence, hallucinating citations, losing coherence entirely. It's compound quantization error. GGUF models already have quantized weights. Quantize the KV cache on top of that, and the errors multiply through softmax. Nobody was handling this. So we wrote our own from scratch. 13.7K li

Hybrid Attention

TLDR: Forked pytorch and triton internals . Changed attention so its linear first layer , middle quadratic layer, last linear layer Inference got much faster with a low perplexity hit in tests .Full attention O(n²): 17.96s / 5.6 tok/sHybridAttention O(n·W + n·D): 0.35s / 286.6 tok/sI have been building a small Rust focused language model from scratch in PyTorch. This is not a finetune. It is byte level, trained from random initialization on a Rust heavy corpus assembled here:

it's not Ai if the LLM is not in control

I always thought that the frontend of "Ai" is awful, but now I know it for sure:OAI5.1+ is good, but chatgpt sucks, it doesn't have gmail integration and barely able to do anything but basic retrieval from the integrations it actually has.Opus is amazing, but claude web is mediocre at best. It has a very limited set of integrations even after 2 years, some don't even work (clay), and it uses way too many tokens to do basic stuff.XAi is ok for social queries but grok is very b

Show HN: Fabro – open-source dark software factory

Hi — I created Fabro to free myself from supervising a fleet of Claude Code tabs running in a REPL (read-eval-prompt-loop). REPLs are great for exploration, but once I know what I need I want to be able to walk away while the agents get it done. (Before building Fabro, I looked for something off the shelf but couldn't find anything that was open source, hype-free, and full featured / ready.)Fabro helps experienced engineers evolve towards a “dark” software factory where average time be

Show HN: Build queryable packs for AI agents from videos, podcasts, and files

Hi,This started from a pretty personal use case.There was this very technical person I follow who would go live on YouTube from time to time. He has a ton of experience, and would casually drop really good insights about software architecture, engineering tradeoffs, and just general "you only learn this after years" kind of stuff. He also posts shorter clips, but I wanted something else: I wanted that knowledge to be always there, queryable whenever I needed it.At the same time, I was

Show HN: I built an open source multi-agent harness in Go

Hey HN. I built an AI agent harness over the past few months and I'm open sourcing it today.Some context on why. I've been building with Claude Code daily using this harness. It orchestrates multiple AI agents as a team, with a dashboard, chat, kanban board, the works. I used it to build a full SaaS product (MyUpMonitor, https://myupmonitor.com) in about 24 hours of focused coding.Then yesterday Anthropic announced Mythos and decided to keep it behind closed doors. Meanwhile

Show HN: 2500 vision benchmarks / evals for Vision Language Models

I love reading benchmark / eval papers. It's one of the best way to stay up-to-date with progress in Vision Language Models, and understand where they fall short.Vision tasks vary quite a lot from one to another. For example:- vision tasks that require high-level semantic understanding of the image. Models do quite well in them. Popular general benchmarks like MMMU are good for that. - visual reasoning tasks where VLMs are given a visual puzzle (think IQ-style test). VLMs perform quite

Toilet Paper Health GIF by Bonny Fiber

Best modem

You can’t have a home Wi-Fi network without a reliable modem. This is why most ISPs (internet service providers) give you one. The catch is, the modem isn’t free; you’re paying for it each and every ...

Show HN: Building your first ASGI framework – step-by-step lessons

I am writing a series of lessons on building an ASGI framework from scratch. The goal is to develop a deeper understand of how frameworks like FastAPI and Starlette work.A strong motivation for doing this is because - I have been using AI to write code lately. I prompt, I get code, it works. But somewhere along the way I see I stopped caring about what is actually happening. So, this is an attempt to think beyond prompts and build deeper mental models of things we use in our day to day lives. I

AI overly affirms users asking for personal advice

<a href="https://arxiv.org/abs/2602.14270" rel="nofollow">https://arxiv.org/abs/2602.14270</a><p><a href="https://www.science.org/doi/10.1126/science.aec8352" rel="nofollow">https://www.science.org/doi/10.1126/science.aec8352</a>

Show HN: Epismo CLI – Make human-AI workflows reusable, like GitHub did for code

Hi HN, I'm Hiroki, founder of Epismo. Just released the Epismo CLI (https://npmjs.com/package/epismo). 380+ downloads right after launch. Thank you.The problem: I got a great result in a Claude Code thread. A week later I couldn't reproduce how I got there. The real workflow lived across chat histories, tabs, tool settings, and tiny followup prompts. Prompts copy easily, but multi-step processes don't.If GitHub made code reusable and Hugging Face made models re