Using Mini Me in a secure and comfortable development environment

Yesterday, I wrote an article: Reproducing the devcontainer DX, but securely with SBX, which explained how to reproduce the devcontainer developer experience, but in a fully secure way, using sbx and VS Code's ability to connect to a VM over SSH. And sbx comes with a feature that makes this integration simpler. I also explained how to streamline the process with the sbx env create command and a simple yaml configuration file (sbxenv.yaml).

But I don't always use VS Code. Actually, when I need to stay focused on a specific topic (just code, for example, or figuring out the right settings for a particular model I use with Mini Me, ...), I prefer to use Zed Editor.

And, as it happens, Zed can also connect to a VM over SSH, so it works with sbx too 🎉.

New kits for Mini Me

So I created and published a zed-ssh-kit kit of type "workload". (An sbx workload kit is the kit that defines the base environment of a sandbox and the main command to run in it)

Then, to be able to use it with my favorite agent 🤓, I created a mini-me-agent-kit kit of type "mixin". (A mixin kit is an add-on that brings tools, configuration or capabilities to a sandbox)

And since I've decided to get into Nushell these days, I took the opportunity to create a nushell-kit kit.

How do I use all of this?

Well, it's simple. A new command arrived with sbx version v0.39.0: sbx env. It's the command that lets you define and manage a sandbox declaratively, from a sbxenv.yaml file. (More info here: https://docs.docker.com/ai/sandboxes/configuration/environment-files/)

So I created the following sbxenv.yaml file in my working directory:

schemaVersion: "1"

name: zed-mellum

agent: zed-ssh-kit

kits:
  - docker.io/k33g/zed-ssh-kit:0.0.3
  - docker.io/k33g/mini-me-agent-kit:0.0.3
  - docker.io/k33g/nushell-kit:0.0.3
  - docker.io/k33g/rust-toolchain-kit:0.0.3

workspace: ./

I also took the opportunity to add the Rust toolchain (it's about time I got into it).

Note: what matters here is name: zed-mellum, it's the name of the sbx sandbox (you'll understand why in a moment).

I added my Mini Me configuration file to my working directory:

.
├── agent.yaml
├── README.md
└── sbxenv.yaml

And now, all I have to do is run the following command to create my secure working environment with the runtimes, tools, ... I picked:

sbx env create --force

--force to avoid being asked questions

And once the sandbox is created, to launch Zed and connect to the sandbox, just use the following command:

zed --new ssh://zed-mellum.sbx$PWD

Where zed-mellum is the name of the sandbox (the one we defined in the sbxenv.yaml file)

And there you go! I have a Zed Editor running in a Linux environment with my tools:

Nushell is available:

I can run Mini Me in tui mode:

And if you add this to Zed's global settings (the host ones):

  "agent_servers": {
     "🤓 mini-me": {
       "type": "custom",
       "command": "mm",
       "args": ["-acp", "-config", "agent.yaml"],
     },
  }

You'll be able to use Mini Me in acp mode to chat "more comfortably":

Once you've closed the editor, all you have to do is destroy the sbx sandbox with the following command:

sbx env rm --force

That's all for today! On my side, I'm going to try to teach myself Rust with Mini Me and a local agent (Mellum2-12B-A2.5B-Instruct). I'll keep you posted, it will be the topic of an upcoming blog post.