github.com
Show HN: Fluxnation, the Fastest FLUX.1 Execution Kernelhttps://github.com/ULT7RA/FLUXNATIONMy kernel FLUXNATION is the fastest FLUX.1 kernel currently in existence. I engineered this from the ground up to replace standard ComfyUI SingleStreamBlock implementations by shifting the entire execution flow to a custom C++/CUDA engine, eliminating all unnecessary overhead. I developed the custom Triton kernel to handle the neuromorphic inspired Spike Attention, iterating the design via Nsight Compute until the hardware utilization reached its absolute capacity.S Tier Performance: The kernel achieves 4–6 second image generation times on any sampler or scheduler, regardless of whether you are running 20 or 50 steps. Once Spikey kicks in, it goes ham and COOKS with 0 image quality loss.Bare metal C++/CUDA Engine: I stripped out the standard execution path and built a custom SSBEngine that interfaces directly with torch.ops. Zero ctypes latency, zero wasted cycles.Fused Computations: The kernel executes gate, residual, and bias computations in a single high velocity lovelace pass. ThisJul 20, 2026 7:34 AM
karankurani.github.io
Show HN: OpenCareLoop - Solve your family's health, one loop at a time.https://karankurani.github.io/OpenCareLoop/Hi HN -Like everyone, I have been using LLMs for managing my family's health and especially in understanding more in depth about the personal health issue for me and my loved ones.I ended up creating an agent to manage the long term history of each member. I use it to manage or solve medical issues. I use it multiple times a week for anything related to health.The alpha version can be downloaded here - https://karankurani.github.io/OpenCareLoop/It differs from the regular ChatGPT / Claude projects in that there are proper workflows to process and add structure your healthcare data. It has a "loop" that allows you to make changes in your lifestyle/medication/habits and track it. It is way more effective over a period of time compared to using general AI chat.Some of the stuff that it has helped with so far in no particular order - solve a long running chronic pain of a parent, manage pain levels of another parent, change decisions related to IVF that reduced the number of scans needed, Jul 18, 2026 4:17 PM
news.ycombinator.com
Ask HN: Workflow Automation vs AI Agents?https://news.ycombinator.com/item?id=48943616A while ago, workflow automation tools (Zapier, Make, n8n) were booming. Now AI agents start eating into that space, OpenAI's Workspace Agents, Lindy, Claude's agentic features, etc.i understand both have tradeoffs, but i'm trying to understand real-world usage:1. why workflow automarion over AI Agents? What does it still do better?2. if you've using AI agents, what made you switch? What do you trust them with?3. why not use either of them?feel free to give your opinion, im curiousJul 17, 2026 5:28 AM
vektorgeist.com
Show HN: Vektorgeist- A platform for AI operators and their agent'shttps://vektorgeist.com/VektorGeist is a platform designed for AI operators to discover, share, and distribute AI resources in one place. The platform brings together: AI tools MCP servers Prompts Workflows Templates Agents Technical articles A community for collaboration Recent updates include selectable layouts, customizable themes, an immersive interface, community profiles, and a marketplace for sharing projects and resources. The goal is to create a central hub where AI operators can find practical resources, contribute their own work, and collaborate with others without having to search across dozens of different platforms. Feedback is welcome, especially on: Overall user experience Features that would improve the platform Missing functionality Ideas for making it more useful for AI operators Website: https://vektorgeist.com Discord:https://discord.gg/EEsMTJ73mJul 16, 2026 9:52 PM
ipek.run
Show HN: I kept rewriting scripts for my teams, so I built a visual tool insteadhttps://ipek.run/Hi HN, I'm Nuru. Ipek is a desktop automation app that resulted from my laziness: automating API calls, Excel/CSV parsing locally, OCR and LLM calls by hand; they all take too much time to maintain, especially when various teams request additional tools on top of those scripts.So I built a visual workflow builder where my non-technical teams can run those deterministic workers themselves, see the logs, and produce the outputs on their own machines. They can also connect the localhost MCP and vibe code new scripts(workers); that will connect to workflows.There is no server side, it's a JVM app (not Electron! :D) that lets you create, compile & attach your own workers; it's basically an IDE for automations. You can create workers either from your own agent over localhost MCP (Claude Code, Codex) or using the builtin Claude chat with your own API key - so Ipek becomes completely free to run, with no per-task fees; your only cost is your own API usage, and you own the data & software. ChecJul 16, 2026 12:34 PM
news.ycombinator.com
Show HN: LHIC – Deterministic device agent which learn and enhance when you usehttps://news.ycombinator.com/item?id=48932029We built LHIC (Local Human Intent Controller) because we were frustrated by existing LLM-based browser agents. Tools like WebVoyager or ServiceNow's baseline agents are amazing, but they are: 1. Slow: Taking 20 to 40 seconds per simple form-fill step. 2. Expensive: Calling GPT or Claude for every single click. 3. Brittle: Failing the second a minor DOM class name shifts. LHIC introduces a "Fast Path / Slow Path" dual-route execution architecture: - Fast Path (Local, 0 LLM call): For standard web flows (logins, forms, search), LHIC matches user intents directly with pre-defined skills and resolved DOM attributes locally. It runs in under 35ms (median) with zero API call costs. - Slow Path (LLM, Fallback): Only activates for complex, open-ended tasks or when a high-risk action is flagged.Key Technical Highlights: * Self-Healing Semantic Locators: We fuzzy-match DOM structural targets using SQLite historical skill memory, achieving an 80% success rate improvement over static CSS/XPath selJul 16, 2026 9:01 AM
choji.ai
Show HN: Choji – turn Slack threads into ready-to-merge PR with Chojihttps://choji.ai/agents/slackHi HN - we're Chris and Jacob, and over the past few months, we've been building a way for teams (non-technical and technical) to easily get the agents they need to handle things like creating engineering tickets, drafting pull requests, and code review without context loss.Today, I'd like to share how this all works end to end with our Slack thread to GitHub pull request flow. And here's a small demo video of how it works made with Remotion: https://www.youtube.com/watch?v=-wDmWIgnBoMSo Choji is a platform where teams can add agents based on their needs in the product development lifecycle. The first step is you connect your knowledge and code bases (e.g. GitHub, Jira, Linear, or whatever you use). Then, connect a place where you get requests (e.g. Gmail, Slack, Intercom, Crisp), then Choji can see the threads as they come or when you command it to e.g @Choji with Slack, and it'll take the request, enrich it with the context it has from your knowledge and code bases to make an actionaJul 14, 2026 4:23 PM
github.com
Show HN: Cruxible – Terraform-like ontology config to governed state for agentshttps://github.com/cruxible-ai/cruxibleGeneral-purpose text memory works well to represent information that can be plainly stated, rarely changes, and only needs to be recalled, but it leaves several structural problems unresolved:1) Generic memory does not establish settled state, it returns context. Text-based memory can preserve and present information, but a claim added to context is reinterpreted by a probabilistic LLM every time it runs. I found myself constantly asking my agents to re-check the validity of what was in my markdown wikis, especially for knowledge that has a natural lifecycle. Additionally, there's no uniform way of establishing provenance or attribution for a claim in text - relying on git commit messages doesn't feel like a good standard.2) An LLM can't do deterministic queries for relational information on its own, especially when relationships have to be constructed across a corpus. This really matters across sessions and for multi-agent workflows - even if one agent or session gets it right, if youJul 14, 2026 2:26 PM
news.google.com
Stack Overflow: Borders and Crossings - GeekDadhttps://news.google.com/rss/articles/CBMic0FVX3lxTE9wSm5pOG5CVm1PTjRVSlJFdWU3THhhVExpQXQyZWtlTzI1aUJvbUFOS05xWDgzX1BiN2pWby1OVnhmYjJsSF9OVlowNVprMDIxWWFSWDE3M05HX0hHVVljRmdaOTZjb183MUY2MHM2SFRyMzQ?oc=5Stack Overflow: Borders and Crossings GeekDadJul 13, 2026 7:00 AM
openbenchmarks.com
Show HN: OpenBenchmarks – Helping agents discover and pick the right SaaS APIshttps://openbenchmarks.comI'm Fenil, co-founder/CEO of OpenFunnel (YC F24), building this with my co-founder/CTO Aditya. We're launching OpenBenchmarks (https://openbenchmarks.com), open-source, reproducible benchmarks for SaaS APIs, starting with the category we know best: GTM APIs.## Why we built thisMore and more B2B software evaluation will/already runs through reasoning models inside agentic workflows rather than through people. And buyers increasingly pick vendors that are API-first and ship MCPs, so they can wire them into internal workflows. Strong reasoning models are skeptical of marketing. When a genuinely neutral benchmark is available, they discount SEO and self-published benchmarks, and they lose trust the moment something looks like a marketing claim.The thing that survives that skepticism is an independent build-first benchmark the agent can reproduce itself and trust.Reproducible by default: every cell ships the literal HTTP request/response plus the judge prompt/responseAll benchmarks live at Jul 11, 2026 8:50 PM
overseer.studio
Show HN: I built a digital game master screenhttps://overseer.studioExactly one year ago today, I got fed up with tab juggling when running my D&D games and decided to build something to organize it all.VTTs were too cumbersome for me because it required a ton of prep upfront and I still needed to teach my players how to use the software. Tabletop games are most fun when the story flows without friction and so my goal was to make a program that made my life as a GM easier.Overseer can take any website, PDF, application, and more and place it on an infinite grid. You can add shortcuts to any tile for easy access, resize, move, and rearrange your scenes, and make your GM screen uniquely yours.It even has plugin support to build your own tiles using simple HTML/CSS/JS.The power of Overseer is just how simple it is to use and extend.Today marks the 1-year anniversary of the first commit I made. I wanted to share it here in hopes others find it useful.Happy gaming!Jul 11, 2026 6:33 PM
apps.apple.com
Show HN: Present and Build Whiteboards, Flowcharts and Diagrams on iPhonehttps://apps.apple.com/us/app/canvasprism-whiteboard/id6785350171Hi HN! I built a native iPhone and iPad whiteboard app on top of the open source Excalidraw libraries, combining them with several other open source components to create a polished offline experience. I also added features like AirPlay presentation mode with an integrated laser pointer, along with over 100 fonts and 400 backgrounds for customization. It was a fun project that brings together a lot of great open source work into a native iOS app. It's completely free to use. There are no ads, tracking, no purchases, or anything like that. I use Excalidraw a lot, and I wanted something I can play with on my phone. Please give it a try and let me know if you run into any bugs. ThanksJul 9, 2026 1:49 PM
arcaide.foo
Show HN: Arcaide – Explore code with multi-level call graphshttps://arcaide.fooOne of the things I do when approaching a new codebase is to find the entry points and start exploring down the call paths. This gives a good overview of the different components in the code and how they're connected. I wanted to translate that to a visual experience, similar to how you would use call graphs, but there's a couple of problems with classical call graphs. One, call graphs represent flow at the function level, so the architectural context is lost. And call graphs tend to get very large and can grow exponentially with program size.The approach I'm exploring is to construct "multi-level" call graphs. These are call graphs represented in the form of program structure showing control flow not just at the function level, but rolled up to higher level units such as classes and packages. This gives you the ability to zoom in and out and see your code at different levels of abstraction, à la C4 diagrams, allowing you to navigate large graphs by expanding the areas you care about aJul 9, 2026 12:59 PM
news.ycombinator.com
Ask HN: I hate coding agents. Is this skill issue?https://news.ycombinator.com/item?id=488443451. Farewell deep focus. Before, I could work in flow state for hours on end. Now, waiting for the prompt to finish introduces constant interruptions, which translates into constant distractions, constant context switch, infinite ideas for endless new projects. My brain is melting. I can literally feel when my neurons start clogging up, forcing me to take frequent pauses just staring outside the window for 10 minutes straight.2. I don't own my own projects anymore. Before, I used to understand every single little implementation detail of a project. When a client reported a bug, I immediately knew what and where to fix in the code base. Now, all I can do is to clasp my hands and recite a prayer to the LLM deity, hoping that my voice will be heard.3. The overlords decide my fate. This is a corollary of #2. There's an outage every other day, and since I don't understand 80% of the code the AI has written, no AI available means it's literally impossible for me to work. I depend completely oJul 9, 2026 11:47 AM
bing.com
I knew there was plenty of AI slop on LinkedIn. Shocking report says the problem is far worse than suspectedhttp://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a8d1f2f2a174fce8c1a31df85c3d497&url=https%3A%2F%2Fwww.digitaltrends.com%2Fsocial-media%2Fi-knew-there-was-plenty-of-ai-slop-on-linkedin-shocking-report-says-the-problem-is-far-worse-than-suspected%2F&c=7212800831468308889&mkt=en-usI already knew LinkedIn was overflowing with posts written by AI, recycled leadership advice, and those god-awful lessons about entrepreneurship. A new report suggests the situation is considerably ...Jul 9, 2026 7:48 AM
boilerroom.ai
Show HN: Boilerroom: dialer, call recording, and managed follow-uphttps://boilerroom.aihello HN, let me tell you why i built boilerroom (https://boilerroom.ai).throughout my past endeavours in sales, the one thing that thoroughly pissed me off was living in a million places to execute one thing; the outbound motionthis includes everything top of funnel like finding leads (be it inbound or outbound), researching, filtering for qualification criteria, throwing them in a sequence, engaging them via dialer, email, and linkedin, gathering context, all the way through bottom of funnel like discovery calls, demo calls, and post-sales activitieswhat absolutely does not exist (or if it does, a piss-poor model) is a place to simply execute this work in one placewhy tf do i have to be on clay, or apollo, clicking around looking for leads, researching them around myself, put them in to some other garbage software to sequence to tell me when to make a call or send out automated emails, only to get a massive backlog of things I was supposed to do, which in turn makes my whole flow looJul 6, 2026 6:15 PM
news.ycombinator.com
Ask HN: I use coding agents daily, but how do real engineers use them?https://news.ycombinator.com/item?id=48806479I have a broad but shallow technical background (longtime management consultant) and use agents (a variety of services and models in VS Code) extensively for my own projects. I know I'm hitting a ceiling. My workflow is naive and nonpro, a lot of tinkering and bashing my way through code generation, context drift, manual fixes, etc.I want to learn how senior engineers treat these tools. Processes, systems, guardrails, must-dos, etc.If you write code for a living and have integrated agents into your workflow, what does your system look like?Any/all input is welcomed: resources, URLs, outlines, warnings, discoveries, recommendations.None of these are specifically needed, but ideas?-How do you structure a brand new project? Scaffolding, git init/ignore? Repos? Initial commit strategies? How do you keep stuff out of the context window that you don't want in it?-How do you "layer" your work so it's much more about the context and integrity of the structure and much less about prompts?-Do yoJul 6, 2026 3:58 PM
github.com
Show HN: Bramble – Local-first password managerhttps://github.com/flythenimbus/brambleI'm currently working on Bramble, an open source password manager with P2P cross-device sync. Initially I released the Chrome extension, but recently I also published the Android app and iOS is pending Apple's approval. Besides that, the latest version also includes passkey storage for all platforms!About Bramble:It aims to be as feature-rich as all popular and a replacement for cloud-based providers. I don't think we need to store our data in the cloud and be at the whims of companies raising their prices every year. There's always a breach and then we find out that some fields aren't encrypted, metadata is visible, and so on. I'm frustrated with this and the increasing lack of transparency during these breaches.The P2P sync in Bramble uses a Nostr relay (which can be self-hosted) to keep your devices in sync. The relay just introduces the devices to each other; the data then flows directly over WebRTC, so there's no vault server and no cloud copy of your passwords anywhere. What leavJul 2, 2026 7:29 PM
strata.space
Show HN: Strata, real-time Markdown editor you can mount as a filesystemhttps://strata.space/showHi HN. Strata is a real-time markdown editor built around document portability across agentic workflows. The same docs live on the web, mount to disk on Linux/macOS with live CRDT collaboration intact, and render inside any MCP app (claude.ai, ChatGPT). There's also a CLI, MCP server, and API. Demo video shows live sync as ClaudeCode reads and writes legal guidance.FUSE is really neat. It can mount entire filesystems on Linux without materializing files until requested, which immediately brings to mind use-cases around knowledge bases for company files, or RAG corpus. However, the Mac side muddied the picture. We started pure FUSE + MacFUSE, but MacFUSE was an onerous kext install. Next we built on native FSKit which is kext-less, but effectively read-only. FSKit won't let you invalidate a cached read, so to push a web edit down to a local file you have to force overwrite it through your own mount to make the kernel re-cache. This is as sketchy as it sounds, and is a poor proxy for colJul 1, 2026 3:02 PM
clockjumper.com
Show HN: Clockjumper – An Intervibed Projecthttps://clockjumper.com/Hello there,Long time listener, first time caller over here. I’ve been meaning to try A.I coding capabilities for some time now. I wanted to focus on a problem that is simple enough for me to execute all the way to the finish line.I didn’t want to mess with any backend complexities or deal with a website that requires APIs or some such. I figured a simple timezone conversion tool would be a decent choice. It is a solved problem to the point of becoming a commodity and all I’m doing is adding, subtracting time and making some tweaks here and there.I don’t know if someone out there has already coined a term for this workflow of mine but I’m claiming a stake and naming it “an Intervibed project”.I’ve used multiple different A.Is to code everything and I can assure you that I’m actively suffering from “Prompt Fatigue”…….so what about this one? has this term been coined yet? I damn sure hope so!The project is being hosted on Lovable, I run my initial prompts thru chatgpt 5.5 Pro and then suJul 1, 2026 2:41 PM
