# Codex CLI and Codex Desktop

> Add Vikey as a model provider in ~/.codex/config.toml with wire_api responses.

Codex uses the OpenAI Responses format (`/v1/responses`). Vikey supports it for every text model, including Claude and DeepSeek.

Tested with Codex CLI 0.167 on 13 November 2026.

## Recommended config

1. Keep the API key in an environment variable. Add this line to `~/.zshrc` or `~/.bashrc` so new terminals have it too:

    ```bash
    export VIKEY_API_KEY=YOUR_API_KEY
    ```

2. Add the Vikey provider to `~/.codex/config.toml`:

    ```toml title="~/.codex/config.toml"
    model = "vikey/agent-gpt"
    model_provider = "vikey"

    [model_providers.vikey]
    name = "Vikey"
    base_url = "https://api.vikey.ai/v1"
    env_key = "VIKEY_API_KEY"
    wire_api = "responses"
    ```

3. Run `codex`. Change the model for one session with `codex -m openai/gpt-6.1-sol`.

### Why each line is there

| Line | Purpose |
| --- | --- |
| `model` | Default model. A fixed model such as `openai/gpt-6.1-sol` or [Vikey Auto](/en/models/vikey-auto/) such as `vikey/agent-gpt`. |
| `model_provider` | Selects the `vikey` provider defined below it. `openai`, `ollama` and `lmstudio` are taken by Codex, so do not use them as the provider id. |
| `base_url` | With `/v1`. Codex adds `/responses` itself. |
| `env_key` | Name of the environment variable that holds the key. The key is not written in the file. |
| `wire_api = "responses"` | Must be `"responses"`. Current Codex rejects `"chat"` with `wire_api = "chat" is no longer supported`. |

Provider settings are only read from `~/.codex/config.toml`. Codex ignores provider settings in a project's `.codex/config.toml`.

## Codex Desktop (macOS)

Codex Desktop uses the same `~/.codex/config.toml`. The difference: an app opened from the Dock does not see variables exported in a terminal. Register the key at the OS level, then quit Codex Desktop completely and open it again:

```bash
launchctl setenv VIKEY_API_KEY YOUR_API_KEY
```

`launchctl setenv` is cleared when the Mac restarts. Run it again after a restart, or open Codex Desktop from a terminal that has `VIKEY_API_KEY`.

## Several setups (profiles)

Since Codex 0.134, profiles are separate files. An example budget profile in `~/.codex/budget.config.toml`:

```toml title="~/.codex/budget.config.toml"
model = "vikey/agent-mocin"
model_provider = "vikey"
```

Run it with `codex --profile budget`. The `vikey` provider still comes from `~/.codex/config.toml`.

## Test the connection

```bash
curl -sS https://api.vikey.ai/v1/responses \
  -H "Authorization: Bearer $VIKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "openai/gpt-6-luna", "input": "ping"}'
```

A response with `"output"` means the key and base URL are right. Then try `codex exec "reply with the word ok"`.

## Common problems

| What you see | Cause | Fix |
| --- | --- | --- |
| `401` or `Missing environment variable VIKEY_API_KEY` | The variable is not exported in this terminal, or Codex Desktop cannot see it | `export VIKEY_API_KEY=...`, or `launchctl setenv` for Codex Desktop |
| `404` on `/v1/v1/responses` | `/v1` written twice | `base_url` is just `https://api.vikey.ai/v1` |
| `wire_api = "chat" is no longer supported` | An old config | Change it to `wire_api = "responses"` |
| `Model '...' not found` | A typo in the model id | Get the id from `GET /v1/models` or the [pricing page](https://vikey.ai/en/pricing) |

Source: https://docs.vikey.ai/en/agents/codex/