Skip to content
LUNAROUTEDocs

Zed

Zed reaches LunaRoute through its OpenAI-compatible provider (language_models.openai_compatible). lunaroute setup zed merges one entry — lunaroute, with the gateway URL and your catalog’s models — into ~/.config/zed/settings.json.

The provider id is load-bearing: Zed derives the API-key variable from it as UPPER_SNAKE(id) + _API_KEY, so lunaroute means it reads LUNAROUTE_API_KEY from the environment at launch and sends it as Authorization: Bearer. That is why no key goes in the file.

1. Merge the provider

Section titled “1. Merge the provider”
Terminal window
npx @lunaroute/cli setup zed

Merge, do not replace. Your settings.json also holds your theme, keymap and any other language models; the CLI keeps all of it, backs the file up, and writes only language_models.openai_compatible.lunaroute:

{
  "language_models": {
    "openai_compatible": {
      "lunaroute": {
        "api_url": "https://gw.lunaroute.com/v1",
        "available_models": [
          {
            "name": "<model-id>",
            "display_name": "<label>",
            "max_tokens": 200000,
            "max_output_tokens": 8192,
            "capabilities": {
              "tools": true,
              "images": false,
              "parallel_tool_calls": false,
              "prompt_cache_key": false,
              "chat_completions": true,
              "interleaved_reasoning": false,
              "max_tokens_parameter": false
            }
          }
        ]
      }
    }
  }
}

Zed requires max_tokens (the context window) on every model, so setup writes only catalog entries that carry a context window — and refuses outright, rather than write a provider that offers nothing to pick, when none do. Models are written from the live catalog at setup time, so re-run the command when you want a newly enabled model to appear.

On Windows the file is %APPDATA%\Zed\settings.json; the CLI resolves the right path for your channel including the Flatpak location on Linux.

2. Supply the key

Section titled “2. Supply the key”

Two ways, and the first one wins when both are set:

  1. Environment — the export LUNAROUTE_API_KEY=lr_… line setup printed, in your shell profile. Zed imports your login shell’s variables at startup, so a desktop-launched Zed picks it up too.
  2. Zed’s keychain — Settings → AI → General → LLM Providers → Configure → find lunaroute → Configure → paste the key. It is stored in the OS keychain; the environment variable takes precedence over it.

To inject the key for one launch instead, lunaroute setup zed once and then:

Terminal window
npx @lunaroute/cli run zed

run zed only adds LUNAROUTE_API_KEY to Zed’s environment — the provider block still comes from settings.json, and it says so if it finds none. Quit a running Zed first: the variable reaches a new process, not the one already open.

3. Pick a model

Section titled “3. Pick a model”

Restart Zed and choose a LunaRoute model in the agent panel. Zed initializes LLM providers in the local app, so remote (SSH) projects use this same settings.json.

4. Undo

Section titled “4. Undo”

Delete the lunaroute entry from language_models.openai_compatible in your settings.json (setup kept your previous file at settings.json.bak), restart Zed, and remove the key from wherever you supplied it — the LUNAROUTE_API_KEY line in your shell profile, or the keychain entry under Settings → AI. Zed has no --remove reverse; deleting the block is the whole undo.

Verify

Section titled “Verify”
Terminal window
curl https://gw.lunaroute.com/v1/models \
  -H "Authorization: Bearer $LUNAROUTE_API_KEY"
Section titled “Related”
  • CLI — install, run, setup, and the Windows/WSL notes.
  • Harness setup — the other harnesses, and the MCP server.
  • VS Code — the other IDE.
LunaRoute documentation