572 results for internal · 2.774s

News for “internal”
20 results • 2767 ms server time
Moozonian News
github.com• Jul 2, 2026• 1 min read
Show HN: Skill Federation –private search across 87k skills for AI coding agentsWe have been focused on AI error distribution for the past year, and in our last research paper, "Architecture of Errors" showed mathematically that an AI solution needs a finite set of interventions to perform well in a bounded patch domain (a specific application). To prove it, we ran harnessed Opus 4.6 on SkillsBench with and without wild skills (skills that you actually find on the internet) that exclude the oracle skills (the skills specifically designed for SkillsBench). That showed 17.5% -> 22.8% (~30% relative lift) as expected.To run the test, we have created a skill search engine for AI agent-native use - not for humans. Agents imagine the perfect set of skills that would be useful for their planned task and Skill Federation fetches them. The engine uses current SOTA tricks such as key word enrichment and reranking and reproduces SOTA numbers on SkillRet.The skills come from internal storage that is pre scanned to the best effort with Cisco and Nvidia security scanners. We cu
Moozonian News
fan-pier-labs.github.io• Jun 29, 2026• 1 min read
Show HN: Crossbeam-CLI – Connect Claude to Crossbeam without the enterprise tierI built crossbeam-cli, an open-source tool that connects Claude to Crossbeam so you can ask questions about your partners, overlaps, and account mapping in plain English.Crossbeam gates programmatic access behind its top Supernode tier. Even the official Claude connector starts around $4,800/year.crossbeam-cli works on any plan, free tier included, because it talks to the same internal API the web app uses. You log in with your own credentials and requests go straight from your computer to Crossbeam.It ships two ways: a one-click Claude desktop extension (MCPB) for non-technical partner teams, and a CLI (npm i -g crossbeam-cli) for scripting and agents.So far I've only implemented read-only tools, but I could add write support in the future.It's unofficial and not affiliated with or endorsed by Crossbeam. It relies on undocumented endpoints, so they could change or break it at any time.Would love to hear what you all think.https://fan-pier-labs.github.io/crossbeam-cli/
Moozonian News
news.ycombinator.com• Jun 29, 2026• 1 min read
Ask HN: What is happening with the current AI startup ecosystem?As an outsider to the US ecosystem, a lot of the surrounding startup activity feels strange to me. Many companies seem to be building thin wrappers, agent orchestration layers, AI-generated apps, or B2B tools whose main customers appear to be other AI startups.At the same time, the social and funding environment seems extremely intense. Everyone is excited by “solving a problem,” everyone is racing toward YC or seed funding like it’s the new Stanford acceptance letter.Do people inside this SF ecosystem not see how weird this looks, or is there something I’m missing? Is it just peer illusion, where everyone is surrounded by people building, funding, and hyping the same things until it starts to feel real? Or am I wrong, and the future really is DevOps agents, AI SDRs, browser operators, vibe-coded internal tools, and agent slop apps selling to other agent slop app companies?Right now, a lot of it looks circular to me: AI startups selling vibe-codeable-in-two-days products to AI startups
Advertisement
Moozonian News
news.ycombinator.com• Jun 27, 2026• 1 min read
I patched llama.cpp to gain 20% prompt processing TPS. Help me make a PRI've been running Qwen3.6-35B-A3B locally on llama.cpp and noticed that prompt processing throughput gets too low with MTP. I got nerd-sniped.What started as curiosity turned into a two-week rabbit hole of experiments and ended with a PoC that fully recovers the MTP PP overhead on GPU, above any expectation I had.TL;DR: instead of processing the last layer MoE FFN for the entire ubatch tokens (usually 512-2048 tokens), this PoC processes only the output row (usually 1 token during prefill). The result is PP TPS is back to the same as with MTP disabled, keeping most of MTP's benefits to TG TPS, even with a slight drop in draft acceptance rate in one of the benchs.I'm not opening a PR to llama.cpp because this is AI-generated code, which goes against their contribution policy, which I support. If you know C++ and llama.cpp internals, I invite to work together with me to open a PR with a more mature implementation.
Moozonian News
news.ycombinator.com• Jun 26, 2026• 1 min read
Ask HN: Options for an independent AI researcher with strong results?I'm outside the AI industry, outside of academia, and no easy contacts into relevant areas. My background lends itself to exploring AI & reasonable level of care checking results. I was focusing on building practical useful things for a portfolio to help change industries mid-career, but that has become something a little different now.The exploration has led to an analytical framework that appears useful more broadly for looking at neural representational models. (It's not a model architecture or training method, benchmark, or prompt engineering.) It is a way of analyzing internal representations through the mapping of structure-preserving connections, to transport them to a frame of reference outside the model, exposing some stable relationships, while not being model-specific or require training something else.It has survived attempts at invalidation, generated useful predictions and functioning interventions, and has continued to reveal additional applications the more I do. There
Moozonian News
news.ycombinator.com• Jun 23, 2026• 1 min read
Ask HN: Anyone else's company ban use of Chinese models?I work for a large company (think banking, insurance, etc.) and we have a ban internally on using any Chinese models. They haven’t stated a reason, but I’m assuming it’s one of the typical ones. I’m wondering how common this is. It feels like a big disadvantage for a company to tie their hands behind their backs like this, given how far ahead the open-source Chinese models are, and given the increasing costs of using the frontier labs’ models.
Moozonian News
news.ycombinator.com• Jun 22, 2026• 1 min read
Ask HN: Switching from backend development to graphics programmingI love computers. I wrote my first program in Borland C++ when I was 11.By chance, I managed at my 15 to get a job where I did some HTML pages and later some PHP programming. Making websites wasn't as fun as making games in C++. Overall PHP didn't seem as fun as C++. I made all of my lab projects at the university in C++ with Qt, or wxWidgets or bare WinAPI. In fact, I improved the university the internal testing system on our Math/Physics faculty. I got a privilege to do labs in C++ & Qt even when they were supposed to be C#. Also, in the university I came across Linux and never came back.Nevertheless, I couldn't get a position as C++ programmer and stuck in web development for 15 years. I focused on backend and took care the systems reliability where I could apply my passion about systems programming. I took all of the possible challenges related to the performance and reliability. In fact, it was awesome making graphs with Graphviz or plots/heatmaps in Grafana to troubleshoot the sy
Moozonian News
baserates.dev• Jun 22, 2026• 1 min read
Show HN: Baserates.dev, memorize programmer napkin mathWhen I began my career in programming I remember reading Jeff Dean's "Latency Numbers Every Programmer Should Know" [1], and getting inspired. Alas, it's been over a decade, and I never fully internalized all the numbers.Until recently. I discovered Simon Eskildsen's talk [2] and github repo [3] on programmer napkin math. I liked his reasoning, and decided to _actually_ internalize the numbers now.So I made baserates.dev.It's a spaced repetition program that teaches you the napkin math from Simon, enhanced with a few things from Jeff Dean's list. The main changes: I included CPU instructions, like L1 Cache reads. I also included some nooby context, like how to pronounce the character μ, and a reminder on what is bigger: nano seconds, or micro seconds.The app automatically saves your progress, and doesn't require auth. Most of the logic is vibe coded: I fed Claude all the links. I use InstantDB [^4] for the backend, mainly to support guest auth and saving progress. Most of my time spent
Moozonian News
news.ycombinator.com• Jun 19, 2026• 1 min read
None-US Claude users: beware if used Fable – account suspension experienceOn Wednesday June 17 I woke up at 6:48 AM (Eastern) local time to an email from Anthropic with the subject: “Your account has been suspended.” It basically says an internal investigation of “suspicious signals” indicates I may be in violation of a supported country policy, and that they revoked my access to Claude. It was followed by two more emails at the exact same time, stating the same thing.I was refunded too: two refunds (one for credits, one for my monthly membership).For context, I’ve been a Claude member for more than two years, and I’ve been on top-tier plans since near the beginning. My use case is mostly business brainstorming, research (business + personal context), and a little coding. So losing access basically stops meaningful work.I appealed immediately. The appeal flow has one “nice surprise”: it allows you to export your data. The drawback: the export link returned a tiny ~235 byte JSON file with basically account metadata (my name, phone, timezone, role, etc.). It d
Advertisement
Moozonian News
github.com• Jun 18, 2026• 1 min read
Show HN: Gcontext – a tree of llms.txt files to steer agents on support tasksI work at a startup that makes martial arts gym software (MAAT). We handle the memberships of students so gym owners don't have to, using a payment system and a database. As we get more gyms we get more support tasks: subscription problems, membership updates, data exports...The way we solve it now is a "tree of llms.txt". An llms.txt normally references what info is available on a website or docs — we use the same idea internally to organize the info the agent needs. The agent starts from a folder and navigates down:. ├── llms.txt # references each folder at this level ├── stripe/ # info.md: how our stripe account is structured ├── firestore/ # info.md: how the schema looks └── support/ ├── info.md # how to resolve support tasks ├── runbooks/ # one file per task, with its own llms.txt │ ├── cancel-subscription.md │ ├── export-gym-data.md │ └── fix-membership-mismatch.md └── logs/ # one file per day, every task the agent resolved Copy With this we can steer the agent much better and cr
Moozonian News
scaleninja.com• Jun 17, 2026• 1 min read
Show HN: Noject – menubar app to stop drive ejections on macOSHi HN, I built Noject, a small native menu bar app for macOS that prevents accidental ejects of always-on external drives.I made it for my own Mac mini setup with NVMe external drives that are always plugged-in and behave like internal storage. Noject lets you protect those volumes and reject any normal eject/unmount attempts from Finder, Disk Utility, Terminal, and other apps.It uses Apple's Disk Arbitration framework, protects volumes by UUID, and has an "Eject Once" bypass for when you really do want to disconnect something. It is privacy first and does not need to read the contents of your drives.Available for lifetime and free/fair download from the website https://scaleninja.com/noject/
Moozonian News
rubyai.beehiiv.com• Jun 16, 2026• 1 min read
Show HN: The Ruby AI NewsletterNow on its 32nd edition, the Ruby AI Newsletter tracks what’s happening at the intersection of Ruby, Rails, and AI coding agents.YC recommends Rails for new startups, YC’s internal software like Bookface, Work at a Startup, and the software that runs the accelerator all run on Rails. Garry has been a vocal advocate for Ruby on Rails for agentic startups, calling it a "crazy unlock". And there is an incredible ecosystem forming around RubyLLM. Ruby and Rails gives agents conventions, structure, and fewer architectural choices to hallucinate.This edition is about that idea from the Ruby side: if models are increasingly interchangeable, restricted, expensive, or temporary, then the durable advantage is the system around them. Ruby and Rails may have a real edge there because they already reward clear conventions, small teams, mature libraries, and boring production software.Feedback welcome!
Moozonian News
mydev.day• Jun 16, 2026• 1 min read
Show HN: I built a dev diary that writes itselfHi HN, I'm Lucas. I mostly build XR things, but over the past month or so I've been developing a tool that became necessary for me: I'm usually working on many different things at once, and I needed a way to create reports and check what I worked on last year on a specific day.mydev.day is an automatic development diary that writes itself as you build. You connect your agent to it once per project (takes a couple of minutes), and from then on it automatically adds an entry per feature. After that you can query it via the API to ask anything about all your work across projects and accounts, or go to the app to print nice looking charts and reports.It started as an internal tool, but inspired by the indie hackers I follow, I decided to turn it into a product to see if it could be helpful for other builders and developers out there!Fun fact: it was really cool to dogfood it on itself as I built it!
Moozonian News
quickish.website• Jun 15, 2026• 1 min read
Show HN: Quickish – Instant HTML HostingAs Claude and ChatGPT has gotten better, I've found myself enjoying using Co-Work to make presentations at work. Sharing the HTML files on Slack and elsewhere was cumbersome and trying to host it somewhere public (even if unlisted) wasn't much of an option for my work stuff.Then I saw Shopify's blog post about Quick (https://shopify.engineering/quick), an internal intranet with simple HTML page hosting and was inspired. I wasn't sure I could get buy-in to host it at my day job so I spent my own time coming up with Quickish. Now I can share all my beautiful presentations.Originally I wanted it to be tied to Google Drive / Workspaces, you share the folder with quickish and put your HTML in, quickish hosts it while respecting the privacy of the folder (workspace only, etc). However, as I worked through building I realized I could make it easier to use and add that part in. Actually, it already works behind the scenes I just need to get the app verified.And now, you have what you see. Ever
Moozonian News
github.com• Jun 11, 2026• 1 min read
Show HN: Kikubot – Each AI agent is an inboxHi All, I’ve posted an OSS Agent framework with its main philosophy being the use of email as the message bus (no message queue, vector store, orchestrator). Essentially every agent is an email address. The main reason behind this project is to facilitate company adoption & deployment of AI agents.# How it behaves (what it looks like) A user sends an email to a specific email address, e.g., kiku@agent.acme.com. Kiku was created as a designated coordinator agent that takes the requested task, like “take the attached article and prepare social media posts”, and routes it to one or more internal agents via email. Kiku knows to which agents to route to because “she” maintains a “roster” of the agents in the cluster. This roster describes the capabilities of each agent. Agents act on the part of the task sent to them and return the results to the coordinator who then returns it to the user.One of the key design aspects is the project uses email threading as state memory. When the agent’s LL
Advertisement
Moozonian News
extend.ai• Jun 10, 2026• 1 min read
Show HN: Extend UI – open-source UI kit for modern document appsWe're open-sourcing 14 components & examples today for PDF, DOCX, and XLSX viewers, plus bounding box citations, file upload, e-signature, and more. It's MIT licensed and fully customizable.Demo video here: https://share.extend.ai/kRmSGKRFWhen we started, we tried every file viewer and document component library we could find. Unfortunately, none of them had all the functionality (and polish) that we wanted, so we ended up building our own for https://extend.ai/. It was only ever meant to be internal, but enough customers kept asking for it that we decided to open source it.It's useful for building document processing agents, real-time user facing document intake flows, or all kinds of internal tooling.We naively thought this would be a solved problem. Turns out, making PDF/XLSX/DOCX viewers that work at scale is not trivial...we use and maintain it for Extend ourselves, so we've fixed a lot of edge cases that came up while running millions of pages / day through our own system. Our ho
Moozonian News
news.ycombinator.com• Jun 10, 2026• 1 min read
We put the Bun Rust rewrite in productionAt Prisma we've been working on a Compute offering to complement our hosted Postgres. We settled on Bun early in the process as we liked the experience we'd be able to provide thanks to Buns batteries-included approach.But then we started using Bun in anger, and we found many memory leaks in the Bun internals, that we knew our customers would face as well. Prisma Compute is operating long-running services, utilising memory snapshots to pause the instance when not in use. We need these Bun instances to be able to reliably serve customer workloads for days and months.We were able to fix some of these memory leaks directly in the code base, but getting the patches upstreamed proved difficult. And running on a fork with an ever increasing number of patches wasn't an appealing option.And then as our Beta launch approached, Jarred announced the big rewrite (prior discussion: https://news.ycombinator.com/item?id=48132488)We have found that the rewrite directly addresses several of the memory
Moozonian News
craftbot.live• Jun 9, 2026• 1 min read
Show HN: The agent that builds and operates its own SaaS toolsFor context, we started working on our general AI agent CraftBot before OpenClaw came out. It works similarly to OpenClaw and Hermes agent: control your PC to do task + memory + proactivity. However, here is the catch: it can create and operate its own SaaS tools with the concept of Living UILiving UI is a system where an AI agent can scaffold and launch real, working web apps on demand. Each living UI can be a dashboard/software/internal tool. They are essentially just frontend with X techstack talking to a backend + database with the techstack of your choice, spun up in its own pair of ports as supervised subprocesses managed by a host process. The backend owns all the state (so the app survives page reloads, tab switches, even host restarts), while the frontend is just a dumb view that fetches data and posts user actions. CraftBot can create a project from a template, install dependencies and launch it. It can also read and write its data through a scoped HTTP client, plus built-in