Setting up Zed and Mistral

I’d been putting off setting up a better AI dev environment for a while, but with the release of Mistral Large 4 (ML4) I figured it was high time.

Picking my stack

Starting with a minimal stack, we only need a couple of pieces: an editor and an AI provider.

Which editor to choose?

There were really a couple of options here:

VSCode — The Standard
Cursor
WebStorm
Vim
Zed
Helix
Cline
Continue

Wanting an open source option disqualified Cursor and WebStorm. Wanting to learn something new disqualified VSCode. Continue was discontinued (haha, pun intended). Cline is an extension. So we were left with:

Vim
Zed
Helix

Never got too used to Vim, and Helix is super interesting, but would require a bit of time to learn. That left only Zed, which felt like a good choice, given it is now the fastest-growing standalone option (Brave AI said so).

Which AI hosting provider to choose

Many options to choose from here:

But I wanted something with EU residency, and maybe even EU-based, so I looked at EU options:

Having previously used Scaleway gave it an edge, but Mistral recently launched EU-residency options, which made things even easier. Just use Mistral EU servers with ML4.

Model

I had already chosen the model — ML4.

Decisions

Zed + ML4 on Mistral EU

Setup

Zed

Zed was easy enough to set up. The Linux script worked fine for me, but there are also Windows and Mac options available on their site. Just download, install, and you should be good to go.

I did also use zed-101-setup, but edited it further, so I’d say it is optional.

Mistral

This took a bit more time. Previously one could create API keys without having an active subscription or being on a pay-as-you-go plan, but apparently they changed that on October 7th, 2026. Now, I need an API key for Zed to connect to Mistral, so I had to choose between the two options: pay-as-you-go or a Pro tier subscription. The first option requires only a billing method, the second costs €18/month. Not wanting to spend too much time on this, I just went with the latter — the Pro subscription.

Once that’s active, you need to create an API key:

Go to chat.mistral.ai and log in
Top left, click the “Vibe” dropdown
Select “Admin”
Under API -> API Keys -> New key
Name it (I chose “zed”) and add some future expiry date

Connecting Zed with Mistral

Open Zed
Top-left dropdown -> Open Settings (or just Ctrl + ,)
Edit settings.json
Search for language_models and add the mistral field inside it, like so:

{
  "language_models": {
    "mistral": {
      "api_url": "https://api.eu.mistral.ai/v1",
      "available_models": [
        {
          "name": "mistral-large-4",
          "display_name": "Mistral Large 4 (EU)",
          "max_tokens": 1000000,
          "supports_tools": true
        }
      ]
    }
  }
}

Notice the api_url — this is how we select the region.

It’s important to note that currently not everything is supported when using regional endpoints, see the docs.

Next, take a look at the sandbox permissions. This is how I set mine up (also in settings.json, at the top level):

// Persistent sandbox permission grants
// https://zed.dev/docs/ai/agent-settings#sandbox-permissions
"sandbox_permissions": {
  "network_hosts": [
    // Python / pip
    "pypi.org",
    "files.pythonhosted.org",

    // Node / npm
    "*.npmjs.org",
    "*.npmjs.com",

    // Elixir / Hex
    "hex.pm",
    "repo.hex.pm",

    // Rust / Cargo
    "crates.io",
    "static.crates.io",

    // GitHub (HTTPS git, API, raw)
    "github.com",
    "*.githubusercontent.com",
    "codeload.github.com",

    // Ruby
    "rubygems.org",
    "index.rubygems.org"
  ],
  "write_paths": [
    "/home/<user>/zed-sandbox"
  ]
}

Alright, that should be it! After saving the settings you can open a new AI agent session (for me Ctrl + Shift + E, but Ctrl + Shift + P -> agent: focus agent also works). Then select the model “Mistral Large 4 (EU)”.

Cool. Hopefully I haven’t missed something too important.

Now that that’s done, the fun part can begin!