Same. It's quite remarkable. Two other things Claude and I have done. I have a spare workstation running proxmox in my garage. Claude can now create, snapshot, delete, extend, start, and restart vms. It's been great. I also have skills that allow it to spin up spot instances when I need to test things that rely on hardware I don't have like certain Nvidia gpus or 100gb networking. What a time to be alive.
True but it was a whole bunch of fragile scripts/functions with lots of ifs. I'll give you a real world example of what's different. Agent spins up a VM to run a build on windows. During the build it realizes it's out of disk space. Agent resizes disk on demand. Build finishes and it launches the app. Runs some tests and realizes it's cpu and memory constrained. Adjust the cpu and memory and runs new tests. tears it down and sends the perf logs back to me. Or it looks at system resources and decides what other vms it can pause while it runs the tests and then turns them back on after.
There are so many unknowns and the ability for the agents to dynamically adjust and make decisions allows for some really powerful autonomous workflows.
Man. HN is brutal. Just one example. And yes, when you're looping agents and builds to identify performance optimizations and build features sometimes you don't know what those bounds are going to be and it's much faster to let the agents work it out, update their learnings, and keep building.
To be fair, often you have good ideas, but not enough time. I have had a lot of fun having Qwen3.8 27b running at home, having it explore and implement ideas that was way to ambitious for the time I had available to tinker with Home Assistant and Proxmox. It is an enabler and can also be a good tutor, when used as such. I have learned a lot.
"be able" implying capability rather than time constraints and opportunity cost which is what most people on here refer to when they mention something cool they're doing with LLMs that they weren't doing before.
Not to take your credit away but there is something called automation, scripting, etc.
All those actions within my homelab are automated via Ansible and bash scripts (I only run Linux), I am only notified if something goes wrong like a backup failed.
I get it if you never worked in IT, that is fine, but you need zero AI to get the above fully automated.
Does Claude keep versioning of the changes?? I host my own repos via Forgejo, Ansible and the automation process consume the config files from there, one source of truth. If my whole setup burns with fire, I can bring everything back online from the offsite Ansible playbook located at Codeberg.
Why not both? My entire home network (dns, zfs nas, subnets, kube, proxmox hosts, ipa domain, its overkill homeprod) is represented by a git repo. Docs, plans, scripts to bootstrap nodes to domain join, kube config, proxmox tooling, ipa as code, nas as code, cert management, my own ec2 as a service, etc. Making changes is done via a mix of one off tooling, ansible and terraform.
I use Claude or codex to manage the entire thing, interchangeably. They keep detailed daily journals. They keep a central TODO.md up to date. They refresh stale docs or stale plans when new plans invalidate old decisions.
I have so many layers of backups and fail safes that I work entirely in yolo mode. It’s really insane what I’ve been able to deploy myself on nights and weekends. Stuff I’ve wanted to tackle for years but never got around to it.
You're approaching this from the wrong angle it's not the technology in the router average people are buying. They care about ease of use, reliability, and features. In that order. Dropbox won because it provided rsync in an easy to understand product. If ddwrt allows you to build features into routers that have good UX for average people then you win. What would those features be? That's the work you should be doing
Off the top of my head. A mobile app for setting up and remotely monitoring and controlling my home network. Pihole/pivpn built in with wireguard and a good UX for creating tunnels. Host isolation. A local llm that I can ask questions and get fixed (why can't I print to the printer from my laptop on ssid1. Fix it). Kid friendly things for parents. Easy kid schedules and device groups to limit screen time and reports on what sites they visited.
Ubiquitis amplifi app had a decent mobile app with some of these things.
I would pay if I could get an openwrt router that was already set up with features like
- ad-blocking
- kid network vs grown-up network vs guest network
- blocking sites (on kid network), with easy way to temporarily disable or disable on a schedule
- safe and regular updates from openwrt, that I can trust will work even if your company goes bust (ie. your layer on top needs to play well with upstream)
Always ask for your imaging data. I have never been denied it. Then take that and run it through your own analysis. Years ago I had some lab work done looking for a specific genetic marker that could explain something. While that was running I took my 23andme genome and cross referenced it with snpedia and had my answer (didn't have the marker) before my doctor did. She was blown away. This was 10 years ago. I'm guessing things have improved quite a bit.
I have been happy with Google's Antigravity harness and Jules so looking forward to playing with this. Thanks for sharing. Simultaneously I am looking to also revisit local offline models.
While I feel like I have a decent understanding of the model landscape I'm feeling a bit lost at which agentic harness to leverage for local models. Hermes, Cline, Aider, Qwen Code, Goose, Pi, OpenCode, something else? I live in the terminal so Desktop UX is a bonus but not a must have.
Can I modify the antigravity settings/program to point to a local model? Where should I spend my energy?
I'm stuck on Windows, so oh-my-pi has been really nice. The others I've tried such as kilo do alright but tool calling can mess up a bit.
Only complaint is that connecting the agent harness to my local model took more work getting configured right than I'd like, but that's been true of most harnesses I've tried as well. Most assume you're using a cloud model and local model configuration is a bit of an afterthought.
I tried many of these pi extensions but they have tons of paper cuts. Don't remember which one had which issue but here are some I ran into:
* commands run by me (! prefix) are also sandbox blocked
* agent has no way to _request_ unsandboxed execution (e.g. if `kubectl whatever` is rejected by the sandbox, the model should have the chance to request permission)
* does not understand shell composition patterns (e.g. if `git status` is allowed and `git log` is allowed, then `git status && git log` should be allowed automatically)
* sandbox only supported on mac or linux. not both
All of that can be fixed by yourself. That's certainly the spirit of pi. But if you want strong defaults and batteries included (like omp promises) then that's just annoying.
- instruct model to write a markdown file with a phased plan to implement whatever feature or change I want
- start a new context, instruct model to implement one phase of the file
- review changes manually, then start a new context and have it do the next phase
- repeat as needed
I've never seen omp touch a file outside of the directory I start it up in, and the few times where I've been unhappy with a change git has been there to revert.
This could easily be a case of survivor bias but I've not had an issue with letting it go yolo yet.
Sometimes agents create huge half-minified one-off python/bash scripts to do some data processing. I'd prefer to neither review nor yolo these. Sandbox restricts reads/writes to designated directories, so at least there's no `rm -rf /` in there.
Since local models are largest constrained by context window you want to have a tiny system prompt. I know Hax: https://github.com/OleksandrChekhovskyi/hax was designed with local models in mind, but I haven't used it.
What's going on with Goose? Seems like Block donated it to some consortium; I can't tell if that's a good signal or a bad one. With so many "contenders", if Goose is going into maintenance mode it'd be helpful to know.
reply