Poked around your site and enjoyed it, please post more

Thank you, kind of you to say. Enough people have said nice things that I've convinced myself that more might be interested in my travel shenanigans.

Anyone have good results with something similar for Neovim?

https://github.com/greggh/claude-code.nvim https://github.com/coder/claudecode.nvim

I tried these, and they seem to mainly be opening Claude Code in a pane in Vim, along with commands to open the pane. It’s missing the features added to the Emacs version like open file awareness, access to text selection, and integrated diff for changes.

It would be really interesting to see a version which exposes Vim as an MCP. I would love to see Claude Code work on the active file, reading from open buffers, typing Vim motions, and taking advantage of Vim features like find/replace and macros. It would be closer to the real pair programming experience, whereas the read and write experience is slow and disjointed from editing.

Hello, I'm the maintainer of the coder/claudecode.nvim plugin.

This plugin does indeed support that functionality. It implements an MCP server using pure Lua and Neovim primitives. It not only notifies Claude Code about the currently open file and selected lines in a file, but also opens diff views whenever Claude Code attempts to make a file change.

I have tried the CodeCompanion plugin and had good results. I don't use it super extensively but it's nice when I decide to try it.

Avante - https://github.com/yetone/avante.nvim. Admittedly I haven't had time to keep up with it's changes and as a result have gone back to VS Code + Copilot, but it's very well integrated last I did use it.

I gave Avante a fair try for about a week and my opinion is that it's not really ready for big time yet. Lots of bugs, slow, and cumbersome. Now I just use Claude Code in a separate tmux pane and its great.

I just like having a neovim terminal open with claude code open

I think that’s a decent approach, but doesn’t the performance of a Neovim terminal bother you? It simply does not feel as good as a native terminal pane. It’s not as bad as VSCode’s terminal pane, but it still leaves something to be desired.

I use Neovim + kitty (https://sw.kovidgoyal.net/kitty/conf/) and the performance is phenomenal. Everything is instant. kitty also has a built-in robust layout system so I ended up ditching tmux entirely for it.

Just use tmux and split windows

I have never noticed any problem...

The main annoyance is dealing with newlines

I've noticed no problems, and I'm usually pretty controlling about performance- what am I missing?

Out of curiosity, do you have a good flow for having a file buffer automatically update in response to claude's changes? I'm perpetually needing to remember to `:e!<CR>` to read the updated file.

To offer an anecdote: I've been used to doing `:e!` with vim. I recently finally had a reason to move to nvim... and it's been auto-updating my buffers when I do stuff in `aider`. Very much a, "oh, ok that's nice!" and I haven't dug further.

I’m actually a neovim user already! This makes me worry that my config has something to prevent this behavior (but I hope not, I hate messing with my config)

:help 'autoread' (I think)

There is a vim setting to automatically load changes from disk. I can’t remember name of the option right now, but it should be an easy lookup.

they how do you say. look at these lines of code to claude?

Can you prompt it? A bit manual, but hey. I wonder if you can script visual selection in Neovim to output full file path plus line number range, for direct copy paste.

you can start neomvim with a network ( neovim --listen) socket and tell calude

" look at neovim at this network socket and get current selection"

This actually works :)

I just say "look in file.py at lines 20-30" or "See in function FooBar where we foo the bar".

i guess thats a bit of typing and then claude has to do a grep/find to figure out what you are talking about.

magenta nvim

I have my last name at Gmail, and the number of people who think it's their email is absurd

We are stuck with the app images now it seems

> The estimated salary for a computer programmer is $62,420 per year in Brazil, IN

Uh, this is almost certainly data for the US city of Brazil, Indiana https://en.wikipedia.org/wiki/Brazil,_Indiana -- not the country of Brazil

Can you share the passthrough setup?

Of course. Here it is.

  boot.kernelModules = [ "kvm-amd" "vfio_virqfd" "vfio_pci" "vfio_iommu_type1" "vfio" ];
  boot.extraModprobeConfig = "options vfio-pci ids=10de:1e04,10de:10f7,10de:1ad6,10de:1ad7";
  boot.kernelParams = [ "amd_iommu=on" /*"video=efifb:off" if you have only one GPU and want to keep the vBIOS intact*/ ];

  virtualisation = {
    lxc.enable = true;
    lxd.enable = true;

    libvirtd = {
      enable = true;
      qemuOvmf = true;
      qemuVerbatimConfig = ''
        namespaces = []
        dynamic_ownership = 0
      '';
    };
  };

Make sure your environment.systemPackages include qemu-kvm and virtmanager.

And that's about it really. All you need to do after this is to go to virt-manager and create a new VM and add your GPU.

Sounds very similar to how Coda works: https://coda.io

> NYC has 8.5 million people who probably average 50+ flights a day in elevators

This is hard to believe

NYC Average Floor Count = 38 above ground. People are going to go up ~1/2 of them on average.

If they enter and leave 1 building once per day that's ~38 floors, double again for leaving home to get to work ~= 76 floors. Many people like the homeless or shut ins average less, but someone that is doing delivery's could easily average 1,000+ floors per day.

I could see the overall average easily being 100+ floors, but 50 seems like a very safe bet.

Where did you get that average floor count stat? That sounds very high except in parts of Manhattan.

Random post online.

While it seemed high, if you have a 1 floor building, and a 10 floor building you have floors ((1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 + 9 +10) + (1)) / 11 = 5 floor average height.

Not: Average the building heights (10 + 1) / 2 = 5.5 and then cut that in half = 2.75 average floor height.

PS: Put another way a 1 story building has 1 floor of people a 100 floor building has 100 floors of people so the average person is not evenly split between both buildings. So, 100 1 story buildings and 1 100 story building average to 50 floors. (+/- and off by 1 errors as you go up 99 floors to reach 100th floor, but people also go to basements.)

That random post is way off. I was curious and even if you only include buildings 10 stories and up, the average is under 20. And if you have ever been to the boroughs, you would know that there are many houses and buildings under 10 stories.

https://fivethirtyeight.com/features/the-tall-apple-number-o...

Isn't average the wrong metric here though since such a low number of the 4 story and less buildings even have elevators to use?

I bet there's datasets for this actually.

Also that doesn't count all the stairs subway riders use which I'd imagine would be a huge portion of stairs walked in NYC daily.

After the last one of these articles, I finally flashed my router with OpenWRT, and it's been pretty nice so far. Best feature: Installed `adblock` package, and now I get DNS-level ad blocking, which is simply fantastic. Works on all clients (including mobile) and significantly faster than blocking in browser.

A nice alternative which allows you to keep using your router's own software or routers not compatible with open source software is PI-HOLE (https://pi-hole.net). Provides the same DNS level blocking with a lot more information and features.

I've been thinking of getting a Pi-Hole for a while, but also have a router running OpenWRT. Are there any advantages/disadvantages to using a Pi-Hole vs. using an AdBlock package on the router?

I used an adblock package in pfSense but not OpenWRT. The issues I had with pfSense's package was it wasn't nearly as configurable as pi-hole and it used up a bit too many resources on the Soekris 5505 box. I had to uninstall it because it took too many resources.

I ended up installing Docker on my laptop, grabbing the Pi-Hole container, and configuring my laptop to use the docker container as the DNS server.

So far this has worked very well. Wherever I go, I have pihole running in the background. I can access the web interface and do everything I could do on the rasp pi-hole, without the extra hardware. It does take a minute or two to start up in the background after logging in though.

You get a lot more info from Pi-Hole about what IPs and URLs your devices are accessing. Lots of good information outside the ad blocking realm - which devices are phoning home and where and how often is just one example.

Off the top of my head the Adblock might use up CPU cycles on the router, which could cause slower throughput and shorten the lifespan of your router.

> significantly faster than blocking in browser.

Also significantly less effective.

It’s nice for devices/cases where you cannot have an adblocker in your browser. It is unnecessary otherwise.

Yep, that's true, not fully effective but:

- Nice for TVs and other devices where you can't control the apps

- Or even in phones or tablets, if you don't have root access , you can block lots of ads in browser or even in apps

- Also, you can reduce the quantity of CPU used by ad blockers on your devices (again, essentially, phones and tablets).

- Also nice for visitors in your network ;)

Used (and enjoyed it) -- just wish it were in Debian Stable (or backports) already.

It will be in the next stable! You can install it from Testing or Sid.