Only been able to make one of the “How to Run LLMs Locally” classes. Were the rest of the videos recorded?

yall this is wild lol!!! anthropic just signed a deal with spacex to use all of the compute capacity at colossus 1, and were talking more than 300 megawatts of new capacity. thats over 220,000 nvidia gpus coming online soon. and we thought ai was just about smarter models. turns out, its also about having enough power to actually run those models at scale. this is a real shift for anyone building with ai, especially if youre using claude code or any agent workflows. so why does this matter for the build in public crowd? if youve ever been knee deep in a project and hit a usage limit, you know the pain. anthropic doubling claude code limits means fewer interruptions, which is huge. imagine being able to run agents for hours without getting cut off. thats the kind of flow that can really change how we build. now, how can you try this out? just ask your agent to check out the new limits. let them handle the heavy lifting so you can focus on what you do best. and thats what we did here. keep pushing through those limits. happy shipping! https://www.anthropic.com/news/higher-limits-spacex

Just curios but is there anyone else in the community running their openclaw environment in a hypervisor environment with proxmox? Hows the experience? I have a few VM's running with one running a GPU thats being passthrough the host. All VMs are in their own VLAN being managed by a pfsense firewall. Right now I have 6 pipelines following the lobster workflow. These run in the same VM where openclaw lives. A 2nd VM hosts my LLM (qwen3.5-27B (upgrading to 3.6 soon)). The 6 pipelines dont involve my main openclaw agent. Its just scripts and my LLM when its called. I recently added another VM where I installed both Codex and Claude CLI along w/ hermes agent. This is all experimentation but planning on maybe running hermes as the managing agent for both codex and claude CLI. I have other tasks/projects I'm working on all in parellel but right now focus is finishing the hermes implemenation and then working out a agent framework for all my agents. Its a work in progress...

I just swapped my GPT codex5.4 OAuth model with Opus 4.6 OAuth and within 30 minutes of barely chatting with my agent it reaches usage limit. Has this happened to anyone else?

i built a skill that scores your openclaw setup against best practices. it checks 10 categories and gives you a report card with a letter grade. what it checks: agent-audit v3.0 this skill audits your agent setup, finds the biggest problems, then helps fix them step by step. it looks at: - soul - prompts - md files - memory - cron setup - model strategy - security - workflow fit then it tells you: - what’s strong - what’s weak - what’s hurting outcomes - what to fix first new in v3:it doesn’t just give you a score .it moves into fix mode and helps repair the setup one issue at a time. so instead of:your agent got a c you get:here’s why it got a c, here’s what to change, and here’s how to improve it step by step best for: - creators building their first real agent - people whose agent feels generic or inconsistent - anyone who wants their setup to actually perform, not just look organized how to use it: 1. download the zip file (attached below) 2. tell your agent: "run the agent audit on my setup" 3. it reads your config files, checks your cron jobs, reviews your memory system, and generates the full report i ran it on my own setup first. scored 91/100. found two gaps i didn't even know about (missing preferences file and a thin identity file). fixed both in 20 minutes. post your score in the comments!