@AlexFinn convinced me to run Hermes locally on my Mac. That was the right move. But I didn’t want my AI system to exist only when I’m sitting at my desk. So I built a hybrid setup: local Mac cloud droplet Hermes on both And honestly, this feels like the right architecture. Running Hermes locally on the Mac gives me the best experience by far: - full local context - direct access to files, tools, and workflows - fast iteration - lower friction - more natural day-to-day use It feels personal, immediate, and powerful. But local-only has one obvious limitation: when the laptop sleeps, disconnects, or isn’t with you, your system goes quiet. That’s fine for a toy. Not fine for something you want to rely on. So I added a droplet and turned it into the always-on layer. Now the architecture is simple: - Mac = primary local brain - Droplet = always-on remote brain - Hermes = the operating layer across both That changes everything. The Mac is where I do the high-bandwidth work: - active chats - building - debugging - testing - steering agents - using the full local environment That’s where local Hermes shines. The droplet handles the part I never want to lose: - uptime - persistence - remote access - background automation - continuity when I’m away - resilience if the local machine is offline So the system is useful even when I’m not at my desk. This is the real unlock for me: I’m not choosing between local and cloud. I’m using both for what they’re best at. - local for speed, context, control - cloud for durability, reach, reliability That’s a much stronger setup than either one alone. And once Hermes is running across both, it stops feeling like “an AI app.” It starts feeling like infrastructure. A local operator when I’m in flow. A remote operator when I’m away. Same system. Different environments. No dead zones. That’s why I think the serious architecture for personal AI is: local-first cloud-backed Not local-only Not cloud-only Hybrid Because the goal isn’t just intelligence. It’s intelligence with uptime. So yes, @AlexFinn was right to push local Hermes. But the real magic, at least for me, was taking that idea one step further: Mac for the best working experience. Droplet for always-on presence. Hermes tying both together. That setup is ridiculously good. <3 @NousResearch

Unreal Engine 5.8 ships today with experimental MCP server support: Your sources, your pipeline and your workflow—simply configure the MCP plugin and connect to any agent. Get familiar with the MCP server and the PCG Primitive Plugin today and see what teams can build together: epic.gm/ue-5-8-blog

browsers can block videos, they can't block text. this repo converts any video into real-time ASCII art and streams binary-encoded frames through WebSockets → 30 FPS playback on HTML5 Canvas → B&W to 16 million color modes → no GPU needed in ASCII mode it's just characters on a screen → autoplay restrictions don't apply it's not a <video> element the browser doesn't even know it's watching a video.

chinese tek is actually amazing GLM is BARELY behind Fable 5 (which is so strong the US gov banned it)

JUST IN: Anthropic faces lawsuit over limits on its $200 Claude Max plan

OH MY GOD its happening @MistralAI has officially confirmed the upcoming release of Le Chaton Fat - 30T MoE with 256 experts - 1M context window - multimodal and multilingual - outperforms every other model (including Fable 5) on every benchmark