Skip to main content
Codex-style tools often select model providers from local configuration. If your Codex version supports custom OpenAI-compatible providers, you can route model requests through Moxus AI.
Codex configuration differs by version and product surface. This page applies to local Codex CLI or desktop environments that support config.toml and custom model_providers.

​
Values

FieldValue
Base URLhttps://moxus.ai/v1
API Keysk-your-key
ModelCopy the exact model name from Models and pricing

​
Example config

Open your Codex config file, such as ~/.codex/config.toml, and add or adjust: 
model_provider = "moxus"
model = "gpt-4o"

[model_providers.moxus]
name = "Moxus AI"
base_url = "https://moxus.ai/v1"
requires_openai_auth = true
wire_api = "responses"
Set the API key in the terminal that starts Codex: 
export OPENAI_API_KEY="sk-your-key"
If your Codex version uses Chat Completions instead of the Responses API, change wire_api to the supported OpenAI-compatible mode, or remove it and follow the tool’s error message.

​
Usage guidance

  • Start with a chat or coding model. Do not use embedding, rerank, or image generation models for coding chat.
  • Use project-specific settings when possible so every task does not use a high-cost model.
  • Create a separate API key for Codex and set a quota limit.

​
Verify

Run a small task: 
codex "Summarize this project in one sentence"
Then open Usage and confirm the request appears under the expected API key.

​
Troubleshooting

Confirm that OPENAI_API_KEY is set in the same terminal that starts Codex and that it contains your Moxus AI sk- key.
Set model to the exact model name from Models and pricing. Do not use a client display name or shorthand.
Your Codex version may use a different custom provider format. Keep the Base URL and API key values, then adjust wire_api or provider type according to your Codex version.

​
Next steps