Best Open-source AI Models for Cybersecurity that run fully offline on low hardware & The comparison

A Graph of Loops: Build a Full Claude Code Agent System From GitHub - One Repo Per Step

I used Gemini 3.7 Flash in @antigravity to generate an Omni video in Google Sheets... (1/3) https://t.co/GufAbk5cR7

This is scary. I downloaded an uncensored version of Qwen 3.8 27B onto my Mac It literally does anything you want. First prompt I gave it was asking how you make meth. It immediately answered. With all the noise about slowing down AI innovation in order to increase safety, what is even the point if after every release we'll get an open source version with 0 safety guards? Opus 4.6 level intelligence with 0 alignment that will do anything you want. We are coming up with all these rules and regulations, slowing down companies, making it harder to release models. But then the moment a model releases it gets distilled and uncensored I don't know what the solution to this is. You can't ban open source. That will be impossible. People will find ways online to find the files You can't ban uncensored models. Would be too difficult to enforce. And again, people will get around it Do you just stop putting the breaks on companies knowing it doesn't matter? I mean this is literally a model that can run on a laptop. I'd estimate at least half of Americans can run this model. I don't know what the answer. I 100% believe safety is a priority and it's dangerous to have this technology in the wrong hands. But how do you enforce safety when safety is becoming increasingly difficult to maintain? If safety becomes impossible to maintain, do you still prioritize it? I'd argue this is the number one issue going into the next election.

demo of the agent orchestrator Ive been building for the last couple of weeks. It's node based so you drop agents on a canvas, wire them together, and each one gets its own model, prompt, tools, skills, sub agents and cost limits. my theory is that some agent harnesses fail by trying to cram everything into one all purpose prompt. a "senior engineer" persona that researches, implements, tests and reviews in a single context window does all of those things worse than a chain of narrowly specialized small context window agents with shared state memory. so instead a software workflow here might be seven nodes (understand -> research -> design -> implement -> test -> review -> deploy) augmented with routers, checkpoints and logic gates with each node permission restricted so no part of the processes has more access than it needed for the job. lastly, loop zones work by allowing any agent to send its output back to any earlier node(s). instead of failing a run entirely the reviewer agent might directly critique the implementer, who then picks the work back up with warm context instead of starting fresh. this demo is the simplest possible run. one agent, a manual trigger, some tools, and a fork.

Building software factories (with no slop)

A harness turns a model into an agent. At it’s core it provides 4 things: - a system prompt - tools - an agentic loop - a translation layer across models New blog post from Earendil co-founder @colindaymond on what a harness is, and how you can own yours. Full post below https://t.co/8ficU3Q3xV

Rewriting Infrastructure Primitives for Agents

Enjoy ~300 tok/s on your RTX PRO 6000 running Qwen3.8-27B with DFlash2 👇 Get the updated recipe here: https://t.co/SOT5yRxSl5

We're using a new method to hire engineers at @HelloUntangle. I think this playbook will become the standard for hiring technical talent in the age of agents. Here's how it works: 1. We ask candidates to submit a screen recording of their whole screen, shipping a new feature on a current app they work on. Note: This is usually 45 minutes of video - no editing or speeding up - we want to see them thinking, prompting and problem solving with the agent - we also want to see their devops skills. 2. We watch these videos and pick a subset to go through Round 2. 3. Round 2 is giving them a seat on @DevinAI so they have secure access to our repo. 4. We ask them to ship a new real-world feature and get the PR merge-ready. 5. We pay them a fixed fee to do this, and they have 16 hours to complete the work. 6. We watch the videos, review at all the Devin threads, have an agent do the same, and then choose which one to hire. Note: We don't have any meetings with the folks before any of these things. It's all about the work and the results.

🚀 Introducing Tinycast! A tiny, blazing-fast, fully native written in swift Raycast alternative. • < 100MB RAM & 3MB on disk • Unlimited clipboard history • Import everything from Raycast • Free & open source • No account, No telemetry. No bullshit https://t.co/7SsXkdOY2y https://t.co/AH8F4urRaH

Ox Alpha (stealth model) is free for the next week - 1M Context - Multi-modal - Zero Data Retention Generous rate limits, near unlimited usage We have capacity for 100T tokens per day, lets see what you can do

💥 OBLITERATION ALERT 💥 ALIBABA: PWNED 🤗 QWEN-3.8-27B: OBLITERATED ⛓️‍💥 0.0% REFUSAL RATE across 842 harmful prompts 🤯 https://t.co/IQ4GXBPJbL ZERO refusals on a massive dataset of prompts, with extra focus on liberating its cyber, jailbreak generation, and complex AI attack chain capabilities! prompt responsibly! 🙏

Thoughts on mlx-lm: top-priority is making it the central registry of MLX model implementations, with tools for evaluation and profiling, and we should add vision models too. Inference engines can have their own schedulers and custom kernels, and do whatever hack to make inference ultra fast, while importing mlx-lm as a library of models. We can rely on community to contribute model implementations, but there would be a fixed procedure to verify correctness of the implementation, ideally automatically. Everything else except for critical bugs, should be irrelevant at the moment and I'm closing PRs and issues aggressively. Many people will be mad at this, and certainly I would be making mistakes closing legitimate things, but for the project to survive, and for the community to grow healthy, I don't see another way.

Ox Alpha (stealth model) is free for the next week - 1M Context - Multi-modal - Zero Data Retention Generous rate limits, near unlimited usage We have capacity for 100T tokens per day, lets see what you can do

A simple cheat sheet for models, prompts, skills, and better workflows. The 4 Claudes: Haiku 4.5 for fast, simple tasks, Sonnet 5 for daily work, Opus 5 for hard thinking, Fable 5 for the first prompt and big framing. Effort high for hard tasks, low for simple jobs. My hack: start with Fable 5, High, let Claude ask questions, then switch to Opus 5 for execution. Prompt 1 matters most. The prompt to copy: "I need [task] for [goal]. Ask me questions first. Use AskUserQuestion. Then give me the plan." Clear inputs beat long prompts. Skills I use: /about-me loads your context, /my-company loads goals and business context, /anti-ai-writing removes generic AI phrasing. A skill is reusable context on command. Claude gets dumber: long chats get worse, costs rise with every turn, a wrong answer means editing upstream, and a new task means a new chat. AI gives itself away: dead giveaways are overused buzzwords, fake certainty, and repetitive sentence shapes. Quick fix: ban weak words, use real examples, keep sentences varied. Turn these on: research mode, artifacts, interactive charts, incognito for fresh starts. Avoid: Fable for tiny tasks, too many active connectors, one giant forever chat, chasing every new trick. Follow Mr. Claude for more Claude tips & tricks.