ChatGPT Desktop

ChatGPT Desktop (integrating Codex Desktop capabilities) supports connecting directly to MoArk models via custom Model Providers, with dynamic model catalog discovery.

Protocol Support

MoArk natively supports the Responses API (wire_api = "responses"), including streaming responses, tool calling, and code generation.

Command-Line Tool

To use the terminal-based CLI version, see Codex CLI.

Prerequisites

1. Install the Client: Download and install ChatGPT Desktop from the official website or the Codex website.
2. Obtain an Access Token: Go to Workbench -> Settings -> Access Tokens to create or copy your token. Keep your access token secure.
3. Choose a Default Model: Determine your default model slug. This guide uses DeepSeek-V4-Flash as an example (the platform supports case-insensitive model slugs, recognizing both deepseek-v4-flash and DeepSeek-V4-Flash).

AI-Assisted Configuration

If ChatGPT Desktop or CLI is already running, you can paste the prompt below to have it inspect the local environment and merge the configuration automatically:

Please configure the current ChatGPT Desktop installation to connect to Moark custom models.

Parameters:
- provider ID: my-gateway
- provider name: Moark
- base_url: https://moark.ai/v1
- wire_api: responses
- model_catalog_url: https://moark.ai/v1/models
- model: DeepSeek-V4-Flash
- auth field: experimental_bearer_token with my Moark access token
- feature flag: api_key_model_discovery = true under [features]

Instructions:
1. Locate the user-level config.toml based on the OS and CODEX_HOME.
2. Read the existing configuration and create a backup.
3. Merge the configuration instead of appending the entire block to the end of the file.
4. Place model and model_provider at the top level before any table declarations; update existing keys in place to prevent duplicate definitions.
5. Check for model_catalog_json, active profiles, or other overriding configurations; report conflicts with their locations rather than deleting unrelated settings.
6. Do not output the access token in chat, logs, or diffs; prompt me to enter it locally if not already provided.
7. Validate TOML syntax after modification and summarize changes.
8. Prompt me to completely exit and restart the client without automatically terminating the application.
9. Guide me through testing model discovery, basic streaming, read-only tool calling, and multi-turn conversation.

Manual Configuration

Locate Configuration File

ChatGPT Desktop stores user-level settings in config.toml. The default path on each operating system is:

Windows

Open the file in File Explorer or a text editor (create it if it does not exist):

%USERPROFILE%\.codex\config.toml

macOS / Linux

Open the file in your terminal or text editor:

~/.codex/config.toml

Configuration File Path

ChatGPT Desktop's built-in Codex mode uses ~/.codex/config.toml. If CODEX_HOME is set, edit the config.toml file inside that directory instead. Back up your existing configuration before making changes.

Edit config.toml

Merge the following block into config.toml, replacing placeholders with your values:

# Top-level settings: must be placed before the first [table] header
model = "DeepSeek-V4-Flash"
model_provider = "my-gateway"

[features]
api_key_model_discovery = true

[model_providers.my-gateway]
name = "Moark"
base_url = "https://moark.ai/v1"
wire_api = "responses"
model_catalog_url = "https://moark.ai/v1/models"
experimental_bearer_token = "<your-access-token>"

Configuration Parameters

Field	Description
model	Default model slug (e.g., DeepSeek-V4-Flash)
model_provider	Active provider ID. Must match the section name [model_providers.<id>] (e.g., my-gateway)
[features]	Set api_key_model_discovery = true to automatically populate the desktop model selector from the server catalog
name	Provider display name (e.g., Moark)
base_url	API base endpoint, configured as https://moark.ai/v1 (or https://moark.ai/api/v1 if routing via /api/v1)
wire_api	Communication protocol format; must be set to "responses"
model_catalog_url	Full remote catalog URL including /models, such as https://moark.ai/v1/models
experimental_bearer_token	Your MoArk access token

Configuration merge rules:

• Top-Level Placement: model and model_provider are top-level keys and must be placed before the first [...] table declaration. Update their values if they already exist.
• Avoid Duplicate Tables: If [features] already exists, add api_key_model_discovery = true under it rather than declaring [features] twice.
• Provider ID Consistency: The top-level model_provider value must exactly match the key in [model_providers.<id>].
• Full Catalog URL: model_catalog_url does not automatically append /models; specify the complete URL.
• Check Overrides: If model_catalog_json is set, it overrides the remote catalog. Remove or comment it out to use remote discovery. Also verify that active profiles do not override your provider or model settings.

Restart and Verification

Save the file and completely exit and restart the client (closing the window does not terminate the background process; exit from the system tray or task manager).

Test the connection in order:

1. Model Catalog: Open the model selector in the lower-left corner and verify that the remote models appear, then select your target model.

   

2. Streaming Response: Send a greeting or simple query to verify streaming output.

3. Tool Calling: In a test project directory, send a read-only request (e.g., "List the files in the current directory and explain their purpose, do not modify files") to confirm tool execution.

4. Multi-Turn Follow-Up: Ask a follow-up question based on the previous response to confirm context continuation.

Troubleshooting

Models Do Not Appear in the Selector

• Check TOML Syntax: Ensure there are no duplicate table headers and that top-level keys are not placed inside a table.
• Verify Catalog URL: Confirm that model_catalog_url includes /models and is reachable.
• Clear Local Model Cache: Codex caches catalog metadata locally. After completely exiting the client, delete the cache file and restart:

Windows (PowerShell)

Remove-Item "$env:USERPROFILE\.codex\models_cache.json" -ErrorAction SilentlyContinue

macOS / Linux

rm -f ~/.codex/models_cache.json

If CODEX_HOME is set, delete models_cache.json in that directory.

Request Errors or Tool Failures

• Verify that your access token is valid, has sufficient quota, and contains no leading or trailing whitespace.
• Ensure wire_api is set to "responses".
• If contacting support, provide the client version, model slug, and error message. Do not share your access token.