Configure Codex

Configure a coding API service for MOI to invoke. This tool form does not sign in to the Codex desktop application.

Prepare the service connection

Obtain a matching Responses-compatible Base URL, API Key, and supported model from the model-service administrator. A ChatGPT or Codex login or subscription is not the API key for this form. Use the supplied endpoint and model range rather than guessing.

You need permission to create or edit tool configuration in the current MOI workspace. If the entry is missing or access is denied, contact the workspace administrator rather than using another person’s account.

Credentials for a direct OpenAI API connection

  1. Open OpenAI API Keys and select Create new secret key.

  2. Copy the generated key immediately; its complete value will not be displayed again.

  3. Use https://api.openai.com/v1 as the Base URL. Do not append /responses, query parameters, or credentials.

For another Responses-compatible service, use its administrator-provided base URL, API key, and model range.

Configure MOI

  1. Open Resource Center → Tools → Codex, select the coding capability, and create an instance.

  2. Enter its name, Base URL, and API Key.

  3. Open the model dropdown, wait for the service’s available models to load, and select one.

  4. Select a reasoning level supported by that model, save, and test the connection.

  5. Select this Codex instance in the agent’s tool configuration.

Field

Requirement

Value

Instance name

Required

A recognizable name such as “Development Codex”

Base URL

Required

The service’s Responses-compatible base address, including any required gateway path; not a browser chat URL

API Key

Required

Credentials matching that service

Model

Required

A model loaded after entering the endpoint and key

Reasoning level

Required

An offered level supported by the server-side model

Validate agent use

After saving, reopen the configuration to confirm it persists and run Test connection. Then bind the capability to the target agent, selecting the intended connection wherever an instance selector is provided. Inspect an actual tool call and compare its result with the target platform. A configured badge or a plausible answer alone is insufficient.

For a small test, ask: “Generate a Python function that sums a list and two tests. Do not modify a repository or deploy anything.” Confirm that the Codex tool was invoked and inspect the returned code. Validate the working directory and operation authorization separately before repository work.

Troubleshoot

Symptom

Check

Empty model dropdown

Base URL, key, network, and the service’s model-list endpoint

Model listed but execution fails

Responses execution compatibility and model capability; a working model list is insufficient

Reasoning level rejected

Use a level explicitly supported by the selected model
Last updated on