Connecting Codex in the ChatGPT app

Codex in alpha testing

What this means:

A separate API key is required: a regular Explyt Enterprise Platform key will not work; ask your organization's administrator to request a Codex key from Explyt support
Only OpenAI models and auto are available; Anthropic models work in Codex CLI
DLP does not work: the data you send is not checked
The connection works only inside Explyt VPC

If the task can be solved with another tool, it is better to choose it: the other connections work without these restrictions — see the list of tools.

On the sign-in screen, click Sign in another way
- Do not click Continue to sign in: this signs you in through a ChatGPT account, and requests to the gateway will be rejected
- If you are already signed in through a ChatGPT account, sign out and sign in again
Paste the API key issued to you into the OpenAI API key field and click Continue
- If you see these instructions but do not have an API key, contact your organization's administrator
- No OpenAI account is needed: the key is used only for requests to your gateway
Open the app settings (Cmd/Ctrl + ,) → the Configuration section → click Open config.toml
The file that opens already contains the app's own settings — do not delete them. Add three lines to the very beginning of the file:
```
model = "auto"
model_provider = "explyt"
model_catalog_json = "./model-catalog.json"
```
Add the block below to the very end of the same file:
```
[model_providers.explyt]
name = "explyt"
base_url = "<YOUR_URL>"
wire_api = "responses"
requires_openai_auth = true
```
- Replace <YOUR_URL> with the OpenAI-compatible URL issued to you
  - If you see these instructions but do not have an OpenAI-compatible URL, contact your organization's administrator

Next to config.toml, create the file model-catalog.json with the list of Explyt models

Download the ready-made model-catalog.json file — or copy its contents manually:

Contents of model-catalog.json

{
  "models": [
    {
      "slug": "auto",
      "id": "auto",
      "display_name": "Explyt Auto",
      "description": "Automatic model selection: high quality with maximum availability",
      "context_window": 372000,
      "max_output_tokens": 32768,
      "base_instructions": "default",
      "shell_type": "default",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 1,
      "support_verbosity": false,
      "truncation_policy": {
        "mode": "bytes",
        "limit": 10000
      },
      "experimental_supported_tools": [],
      "supported_reasoning_levels": [
        {
          "effort": "low",
          "description": "low"
        },
        {
          "effort": "medium",
          "description": "medium"
        },
        {
          "effort": "high",
          "description": "high"
        },
        {
          "effort": "xhigh",
          "description": "xhigh"
        }
      ]
    },
    {
      "slug": "OpenAI/gpt-5.6-sol",
      "id": "OpenAI/gpt-5.6-sol",
      "display_name": "OpenAI GPT-5.6 Sol",
      "description": "The highest quality among OpenAI models",
      "context_window": 372000,
      "max_output_tokens": 32768,
      "base_instructions": "default",
      "shell_type": "default",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 2,
      "support_verbosity": false,
      "truncation_policy": {
        "mode": "bytes",
        "limit": 10000
      },
      "experimental_supported_tools": [],
      "supported_reasoning_levels": [
        {
          "effort": "low",
          "description": "low"
        },
        {
          "effort": "medium",
          "description": "medium"
        },
        {
          "effort": "high",
          "description": "high"
        },
        {
          "effort": "xhigh",
          "description": "xhigh"
        }
      ]
    },
    {
      "slug": "OpenAI/gpt-5.6-terra",
      "id": "OpenAI/gpt-5.6-terra",
      "display_name": "OpenAI GPT-5.6 Terra",
      "description": "Balance of quality and speed among OpenAI models",
      "context_window": 372000,
      "max_output_tokens": 32768,
      "base_instructions": "default",
      "shell_type": "default",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 3,
      "support_verbosity": false,
      "truncation_policy": {
        "mode": "bytes",
        "limit": 10000
      },
      "experimental_supported_tools": [],
      "supported_reasoning_levels": [
        {
          "effort": "low",
          "description": "low"
        },
        {
          "effort": "medium",
          "description": "medium"
        },
        {
          "effort": "high",
          "description": "high"
        },
        {
          "effort": "xhigh",
          "description": "xhigh"
        }
      ]
    },
    {
      "slug": "OpenAI/gpt-5.6-luna",
      "id": "OpenAI/gpt-5.6-luna",
      "display_name": "OpenAI GPT-5.6 Luna",
      "description": "The fastest and cheapest among OpenAI models",
      "context_window": 372000,
      "max_output_tokens": 32768,
      "base_instructions": "default",
      "shell_type": "default",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 4,
      "support_verbosity": false,
      "truncation_policy": {
        "mode": "bytes",
        "limit": 10000
      },
      "experimental_supported_tools": [],
      "supported_reasoning_levels": [
        {
          "effort": "low",
          "description": "low"
        },
        {
          "effort": "medium",
          "description": "medium"
        },
        {
          "effort": "high",
          "description": "high"
        },
        {
          "effort": "xhigh",
          "description": "xhigh"
        }
      ]
    },
    {
      "slug": "OpenAI/gpt-5.5",
      "id": "OpenAI/gpt-5.5",
      "display_name": "OpenAI GPT-5.5",
      "description": "Previous-generation OpenAI model",
      "context_window": 272000,
      "max_output_tokens": 16384,
      "base_instructions": "default",
      "shell_type": "default",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 5,
      "support_verbosity": false,
      "truncation_policy": {
        "mode": "bytes",
        "limit": 10000
      },
      "experimental_supported_tools": [],
      "supported_reasoning_levels": [
        {
          "effort": "low",
          "description": "low"
        },
        {
          "effort": "medium",
          "description": "medium"
        },
        {
          "effort": "high",
          "description": "high"
        },
        {
          "effort": "xhigh",
          "description": "xhigh"
        }
      ]
    }
  ]
}

To add a model, copy any block and put your own slug value in it.

The priority field sets the order in the model picker: the lower the number, the higher the model in the list
To get the full list of models, contact your organization's administrator or open <YOUR_URL>/models
If the error failed to parse model_catalog_json appears after a Codex update, add the field named in the error message to every model

Fully close and reopen the app
- The settings are read only at startup
In the bottom-right corner of the input field, click the button with the model name, then the row with the model name and the › arrow in the window that appears
Choose a model from the list
- Explyt Auto – automatic model selection: high quality with maximum availability
- OpenAI GPT-5.6 Sol – the highest quality among OpenAI models
- OpenAI GPT-5.6 Terra – balance of quality and speed among OpenAI models
- OpenAI GPT-5.6 Luna – the fastest and cheapest among OpenAI models
- OpenAI GPT-5.5 – previous-generation OpenAI model
- If there are no Explyt models in the list, Codex did not read the catalog: check that the lines from step 4 are at the beginning of the file and the lines from step 5 are at the end of the file
Send any message and make sure the model responds
- If the error 404 Not Found appears, the key is wrong or revoked: check that you signed in with the API key and not through a ChatGPT account, or contact your organization's administrator to reissue the key

For errors, see Error messages.