Made a new thread, so you can post your burning question to Wendell there!
Agenda
After the Live Stream We’ll do some Q&A
Level1Linux / Linux Weekly
I’m going to try a lighter-weight Linux Weekly format covering Linux news, interesting drivers/kernel developments, and weird hardware/software topics worth talking about before they become full videos.
There is now a forum category for ideas, suggestions, and comments for the show.
Support / housekeeping
I want to thank everyone supporting this work.
We’ve also set up sponsorship through GitHub if that’s the easiest way to support these projects.
Intel B60 / SR-IOV
One possible Linux Weekly topic is the work we’ve been doing on Intel B60, especially SR-IOV finally working on B60, as a follow-up to the B50 work.
I just moved my daily Desktop from Win 11 back to Linux Mint after the windows 12 rumors. I’m going back to 100% Linux and would love to have some regular L1Linux content to listen to along the way!
Super excited to try out Turnstone/kappa. Especially since it’ll fit on my sad little 4070.
Let’s go linux streams! And you’ve set the bar high with the first one
I might try making a gui for sinfo, seems like a very cool project. It’s a bit scary that it talks to the i2c bus directly tho, although I guess if you want to abstract away from specific versions of the kernel, and random out of tree modules, this is unavoidable.
Topic suggestion - How new linux users can contribute/add to the community.
I’m a longtime vmware/windows sysadmin convert, scripts? Sure… but not a developer by any means. Short of encouraging people to join the year of the linux desktop, or contributing monetarily to projects.. is there anything else new folks can do to help?
How does Turnstone compare with say Agent Zero or Nanoclaw?
Have you read the Stripe blog post about their Minions platform? How they describe Minions is how I think agentic should work - user prompts are orchestrated to a sandboxed container and environment where the model can run whatever commands necessary to complete the task without any real risk then the user reviews the outputs in a PR. For sysadmin tasks I envision giving the container and context access to either virtualized or a specific lab hardware/environment/tenant that is always restored to known state not much different than we would give junior engineers to test their work.
Currently I’m running pi agent in docker calling to ollama over the network with dependencies and guardrails baked into the container and limited file system access with bind mounts but this doesn’t currently scale past one user and active task. Both because I don’t have it plugged into an orchestrator or input besides cli and because I only have one R9700
Haven’t looked at agent zero or the claws really. Turnstone is just a clean slate implementation ive been building to solve the issue of multi-node orchestration of intelligence. I am focusing on having controls around execution. The infrastructure as code side of ops is a subset of the problems turnstone is capable of solving. So you could have nodes/agents inside of containers that are modifying your IaC code, reviewing changes, and committing them so your normal deployment pipeline picks them up and you can have nodes/agents running directly on the baremetal troubleshooting your k8s stack, verifying deployments, etc.
EDIT: Should add there is currently a client sdk so you can easily hook it into CI/CD pipelines, alerting systems, escalation paths. I will be implementing webhook notifications so it can be tied into services like pagerduty as well.
Very excited about the topic. I’m personally interested in creating a project management tool or tools to help schedule my tasks in Obsidian or DB. Using Turnstone could help get a more deterministic result from local LLM agents and defined tool paths for it to run.
Especially since I can run local LLMs on my gaming PC, but not on my home server. I could have a queue of tasks on the home server that wait for the gaming PC to be free and run the model.
