Environment

• OS: Linux (Arch)
• t3code version: 0.0.8
• Codex CLI version: 0.112.0
• Custom provider: Portkey (model_provider = "portkey" in ~/.codex/config.toml)

Codex CLI config (working standalone)

  model_provider = "portkey"

  [model_providers.portkey]
  base_url = "https://api.portkey.ai/v1"
  env_key = "PORTKEY_API_KEY"
  wire_api = "responses"

Running codex directly from the terminal works without issues.

Problem

t3code does not work when the Codex CLI is configured with a custom model_provider in ~/.codex/config.toml (e.g. Portkey, Azure OpenAI
proxy). Two issues prevent this:

Auth check rejects custom provider authentication

t3code runs codex login status and parses the output for strings like "not logged in":

  // apps/server/dist/index.mjs ~L15531
  const authProbe = yield* runCodexCommand(["login", "status"])

  // apps/server/dist/index.mjs ~L15424
  function parseAuthStatusFromOutput(result) {
      const lowerOutput = `${result.stdout}\n${result.stderr}`.toLowerCase();
      // ...
      if (lowerOutput.includes("not logged in") || lowerOutput.includes("login required")
          || lowerOutput.includes("authentication required")
          || lowerOutput.includes("run `codex login`")
          || lowerOutput.includes("run codex login")) return {
          status: "error",
          authStatus: "unauthenticated",
          message: "Codex CLI is not authenticated. Run `codex login` and try again."
      };

When using a custom model provider, authentication is handled via environment variables (e.g. PORTKEY_API_KEY), not via codex login.
The CLI works perfectly fine — but codex login status still reports "not logged in" because there's no ~/.codex/auth.json. t3code
treats this as a hard error.

Provider is hardcoded to "codex"

The session object unconditionally sets the provider:

  // apps/server/dist/index.mjs ~L9580
  const session = {
      provider: "codex",
      // ...
  };

There is no path to configure or detect alternative providers.

Portkey needs specific headers, which the model_provider of the codex cli handles. In this case: {"status":"failure","message":"Either x-portkey-config or x-portkey-provider header is required"}

Expected behavior

t3code should respect the Codex CLI's configured model provider and not require codex login when a custom provider with env-based auth
is configured.

Suggested fix

• Skip or downgrade the codex login status check when the CLI has a custom model_provider configured (e.g. by checking codex config
  output or parsing ~/.codex/config.toml)
• Read the provider from the Codex CLI config instead of hardcoding "codex"

Related

• macOS: Electron app cannot access AZURE_API_KEY from ~/.zshrc when using env_key in Codex CLI config #317 — environment variables not inherited on macOS (same area, different root cause)