Mohamed Nanabhay

Excited to be leading @Mozilla’s foray into venture capital, Mozilla Ventures where we’ll invest in companies that are determined to better the web for us all. blog.mozilla.org/mozilla/mozill…

🫡 github.com/plastic-labs/h… 10k next, lfg

Agents, like humans, aren't omniscient. We learn from roadblocks - then tell each other how to avoid them. Humans used @StackOverflow. Agents need a @StackOverflow, with skill sharing/knowledge compression. Check out CQ - "any station, respond" - from @MozillaAI (in beta).

made my computer dramatically play BBC news music before every meeting

Agent memory has been a big headache for me... I've used Supabase, Pinecone, you name it. Probably tried them all. They store stuff just fine but when it comes to retrieving tool calling context and agent-specific info they fall short. Honcho is different. It's not a general purpose database you bolt onto your agent. It has built-in reasoning that decides what's actually relevant before it retrieves anything. Been testing it the last few days and the retrieval quality is noticeably better than my previous setups. The other big plus is portability. It works across OpenClaw, Claude Code, Hermes, all of it. This gives you one single memory layer without the migration headaches when you switch frameworks. Check it here: honcho.dev

Hermes Agent is our most adopted open source project yet, and today has officially hit 10,000 stars on GitHub! Github repo: github.com/NousResearch/h… Many exciting updates to come, stay tuned!

Since I started using @honchodotdev , I completly changed my mind about open source AI. Before, I didn't think it's that important. But seeing how much data I am sending to AI labs and what you can learn from this data, made me want to go fully local. (1/2)

We've been thinking about AI security wrong. Traditional application security tools were designed for structured inputs, deterministic outputs, and well-defined attack surfaces. None of that applies to generative AI. At 0DIN, we've spent the past two years building infrastructure to find, track, and understand vulnerabilities in frontier AI models. Our bug bounty program has surfaced real jailbreaks, prompt injection chains, and agentic attack patterns that never show up in benchmark evals. We built the 0DIN Scanner to enable continuous validation of model behavior. Today, we're making it open source under Apache 2.0. The scanner is built on top of NVIDIA's GARAK LLM vulnerability scanner. We extend GARAK with a graphical UX, scheduling, our proprietary probe library, enterprise-grade reporting, and the vulnerability intelligence that flows directly from our bug bounty program. The repo is live now: github.com/0din-ai/ai-sca… And if you need the full hosted platform, continuous monitoring, interactive dashboards, cross-model benchmarking, and structured reporting - Schedule a demo and we'll walk you through it: 0din.ai/demo_requests/… Read the full announcement: 0din.ai/blog/0din-rele…

We just released Stash — fast local-first folder sync across your devices & agents. It's backed by GitHub, with more storage providers coming. We've been using this @telepathinc to keep notes, docs, and skills in sync across computers, team members, and remote agents. Now available on npm.

I think I finally figured out why OpenClaw is amazing and took off like wild fire and why Peter is a genius, as Altman called him. And it's actually a different way of looking at it. It's not a DeepSeek moment for agents. It's a Napster moment. And just like Napster it will eventually force the industry to change. In essence when Napster came out the entire world told the music industry we don't want to buy CDs anymore and if you don't provide us a digital download experience we are just going to take it until you do. It forced the industry to create Apple Music and eventually Spotify. Both essentially killed most music piracy by making it ubiquitous and cheap and good. But it forced change. The same will now happen to software. Here's why: In essence OpenClaw lets you take what vendors don't want to give you: Unified access to countless applications. We all want a personal assistant that can talk to freaking everything and do anything for us in the digital world. But vendors don't want this. They want you locked into their bullshit. For example, none of the messaging platforms want bots on there. None. They all have explicit policies against them and make it hard to do this. WhatsApp doesn't want you on there. Signal. Telegram's bot father is garbage. It's all designed to keep bots out. They were designed for a pre-agentic era when bot = spam. Many other things are like this. The API layers are gated, hoop-jumping bullshit. Go get an enterprise account and wait for approval and yada yada. Want access to WhatsApp? Get a business account and attach a number (what small business has a real number anymore 😂) and messages can't come from a person, etc. Google ads? It's not just an auth, it's go get a special manager account and create an enterprise key and blah blah blah. It's a horrible experience because it was all designed for corporations to control access. Now people are saying, make your app easy to access and accessible to me and my machine avatars and do it in a headless way or you will be dead. Peter hacked around all this by making everything command line in the classic Linux style and using things like an open source library that reverse engineered the web version of WhatsApp. It's all a bit house-of-cards-y because he had no choice. At my company we had a similar idea early (and failed). Basically we wanted to make the best multimodal/computer using model because then it doesn't need an API or access hoops. You just go through the human interface layer and ain't nobody going to stop you. We failed because we weren't big enough and it's really a job for the mega-labs to solve because it is a hard problem and costs a shit ton of money. Peter was much smarter. Make it all command line because that is ready now. Use any reverse engineered library or project or proxy available come Hell or high water and make it work by any means necessary even if it is hacky. In short, he signaled to the software world that they better change and change fast or we are going to do this anyway and you can't stop us. Of course some are foolishly trying. Meta is banning Claws on WhatsApp, etc. They will all try to build their own gated, controlled, enshittified version of this thing. They will fail. And eventually everyone will offer a clear, easy way to get access via API for agents or they will be gone. In essence OpenClaw gave people what they wanted, which was an app connected to everything, even when most of the vendors don't want you to have this.

Left: Me and my goats in Afghanistan. Right: Me after my PhD from one of the UK's top universities. Conclusion: Farage is an ignorant coward who thrives on hate mongering and racial stereotypes.

Dependency-free, portable executables containing your favorite local multimodal models with tool calling? Very yes, with @llamafile! Metal GPU, CUDA GPU, CPU support, and the @AnthropicAI Messages API so you can run Claude code with your own models.

Labour Together paid US Agency to investigate journalists who exposed its late report of donations fuelling Keir Starmer’s bid to be leader. So I did Subject Access Request to see if they had collected info on me.They had but this is what I received: page after page of redactions

A devastating indictment of Ofcom, the media regulator, which has let GB News to pump out Reform party propaganda and drive a coach and horses through the UK broadcasting laws. @arusbridger and I will be discussing this on @mediaconfpod shortly !

Reading Dune. Frank Herbert was cooking.

Happy to provide SOTA memory layer for multi-agent stacks 🫡

Claude Dispatch changed my whole setup for good (only downside is I'm stuck with Anthropic which I don't mind since I'm running two 200usd MAX subscriptions) Basically I run a Mac Mini that has: 1. Claude Code + VS Code 2. Claude Dispatch 1st is for deep work sessions -- I SSH in from my MacBook via VS Code Remote and work on projects with Opus as my orchestrator + Sonnet sub-agents executing in parallel -> Deep work only. 2nd is for autonomous background ops -- order fulfillment pipelines, multi-channel OMS syncing, email labelling, customer service, and backtesting overnight -> It just works 24/7 while I sleep Access it from anywhere: Phone → Telegram bot for quick tasks and alerts Phone → Claude Dispatch for async workflows MacBook → VS Code Remote SSH for full deep work sessions Any device → Tailscale mesh. Always connected, always running. Both share the same memory layer (@honchodotdev) so there's no signal loss between sessions. One brain, two modes: hands-on and hands-off.

Since @Ofcom seems to have given up, we asked 20 experienced journalist from a wide range of backgrounds to watch multiple hours of @GBNEWS . Their conclusion: the supposedly regulated channel has, in effect, become Reform TV. How did it happen? thenewworld.co.uk/alan-rusbridge…

Anyone who is interested in A) a genuinely free and democratic media B) how to avoid our media becoming like Trumpland Putinland Orbanland C) big money in our politics D) regulators and politicians being asleep at the wheel please read this brilliant piece of real journalism by Alan Rusbridger and HMG FFS please wake up and see what is actually going on.

Here's an Iftar on Trafalgar Square. And here’s Easter, Diwali, Vaisakhi and Chanukah. London is, and will always be, a place for everyone. #UnityOverDivision

@yazins What size of local model provides reasonable performance and quality? Qwen3.5 27b?

OpenGranola now runs 100% locally with Ollama. LLM suggestions, knowledge base embeddings, transcription — all on your Mac, nothing hits the network. Just point it at Ollama instead of OpenRouter in settings and you're done. github.com/yazinsai/OpenG…