🧠 What actually happens when you hit send? You chat with your agent every day — but do you know what it's doing in there? 👀 Most people use Hermes like a chatbot. Huge mistake. In Class 3 we open the hood and follow one message through the entire machine: Gateway → Guard → Prompt Assembly → the Loop → Post-flight. You will never look at your agent the same way again. 🤯 🎁 What you'll discover - 🧠 Why Hermes "forgets everything" every single turn — and the memory-tattoo system that rescues it - 💸 How prompt caching slashes your token bill 50–90% (same answers, fraction of the cost) - 🔄 The operating loop — Perceive → Reason → Act → Iterate — on a 90-iteration budget with 40+ tools - 🛡️ The Approval Inbox: the safety gate standing between your agent and disaster - ✅ Verify-on-Stop — why Hermes checks its own homework before it reports back The LLM is the engine. Hermes is the whole car. 🏎️ After this class you'll finally know what you're driving. 👉 Take the class Class 3 — How Hermes Actually Thinks is live now: https://www.skool.com/tolearnaiautomation-4236/classroom/43a05e49?md=c9f9af8b321442a7bd8252d0ac7be314 Next up: Install in 10 Minutes ⏱️ — you don't need it yet… but you'll want it. 😉

Joined to see if i can get up and running with Openclaw in a somewhat secure fassion :-) I've heard the best approach is to do docker sandbox. and ive tried to find tutorials to follow but everything is changing so fast. I also dont have the possibility for a guest network on my router so I've heard that its possible to use tailscale instead to achieve the network isolation. that way i can have a secure private network that also extends to my Iphone. docker sandbox (the new microVM) encapsulating openclaw and tailscale encapsulating the sandbox. I would also want to I'm pretty new to docker though so need some help getting things put together i think. as far as i can see I cant use the install guide available in the classroom? has anybody tried this setup?

🛑 Read This Before You Add n8n or NotebookLM to OpenClaw A new addition to the OpenClaw Usage Companion Courses section of the OpenClaw course is now live: 🔗 Access the Module Here: https://www.skool.com/tolearnaiautomation-4236/classroom/c8b580f7?md=e94ed8b1387a4536a63d09348bcc25a8 I’ve been seeing a trend in the community where members are bolting on external tools like n8n and NotebookLM before they’ve fully mastered the native power of OpenClaw. In this update, we’re breaking down the "Complexity Budget"—and why adding these specific tools right now might actually be working against your setup. What’s inside the update: - The n8n Trap: Why using it for "simple" cron jobs creates a security bypass in your Soul Fortress. - The Token Fallacy: Why routing tasks to external automation doesn't actually save you money on our subscription-based API. - NotebookLM vs. Native Memory: How memory.md and the native PDF tool handle 90% of what you’re reaching for Google Cloud to solve. - The "Local-First" Roadmap: A guide to keeping your personal research and data on your own machine (using Obsidian). The goal isn't to use every tool that exists—it's to use fewer tools, better. Master the foundation first. Explore the native memory, the native PDF tools, and the internal agents file before you add more points of failure to your assistant. See you in the comments of the module!

Hello! I need to understand how we will use the different LLMs on our OpenClaw setup. So according what i understand the plan is: try to use free/almost free LLM's and switch between them according availability. So in essence, our OpenClaw will : - Priority 1 (paid): if you want use Z-AI (paid). You will get a LLM with a close intelligence to Claude Opus for a fraction of price. Cost: about 9$/month - limited community member link in https://z.ai/subscribe?ic=E71GZDVZPS). - Priority 2 (free): if you run out of credits from option 1, switch to Nvidia kimi-k2.5 (follow the tutorial in this community - https://build.nvidia.com/moonshotai/kimi-k2.5 . Cost: free. - Priority 3 (free): if you run out of credits from option 2, switch to OpenRouter.ai LLM (which will choose between several others LLM's based on availability). Cost: free but if you have 10$ in the account they raise a lot the threshold of tokens you can use. - Priority 4: use another free LLM or paid if you have API keys (OpenAI, Claude, etc) Am I understanding this well? Thanks in Advance! 🙂

Good Morning everyone! We've had four significant releases since v2026.3.8, and this latest one (v2026.3.12) has some important security patches that make this update non-optional. Here's the full picture. 🔐 Security Fixes — Update Immediately For These Alone Two CVEs addressed across v2026.3.11 and v2026.3.12: WebSocket hijacking patch (v2026.3.11 — GHSA-5wcw-8jjv-m286): A cross-site WebSocket hijacking vulnerability in trusted-proxy mode that could have granted untrusted origins admin access. Closed. QR/pairing token hardening (v2026.3.12): /pair and openclaw qr setup codes now use short-lived bootstrap tokens — shared gateway credentials no longer get embedded in chat or QR payloads. Plugin auto-load disabled (v2026.3.12 — GHSA-99qw-6mr3-36qr): Cloned repositories can no longer silently execute workspace plugin code without an explicit trust decision from you. This is a meaningful one if you share or pull configs from external sources. ⚠️ BREAKING CHANGE — Cron Users Read This First If you have cron jobs set up, before updating run this after the pull: bash openclaw doctor --fix Cron delivery has been tightened in v2026.3.11 — jobs can no longer notify through ad-hoc agent sends or fallback main-session summaries. The doctor --fix command migrates your legacy cron storage automatically. If you skip this and have active cron jobs, your scheduled tasks may stop notifying correctly. ✨ What's New Worth Knowing About Dashboard refresh (v2026.3.12): The Control UI got a full overhaul — modular overview, chat, config, agent, and session views, command palette, mobile bottom tabs, and richer chat tools including slash commands, search, and export. Memory: multimodal indexing (v2026.3.11): Opt-in image and audio indexing for memorySearch.extraPaths using Gemini embedding. Your bot can now search across images and audio files in your memory paths — not just text. Telegram model picker fixed (v2026.3.12): Inline model button selections now persist the chosen session model correctly. If you've been frustrated by model selections not sticking in Telegram — this is fixed.