OpenAI
Connect an OpenAI API key and choose from the GPT models currently available in Mobius.
An OpenAI connection lets managed assistants run GPT models with your organization's API key. Mobius prefers that key when both organization and platform credentials are available.
The connection is only a model credential store. It does not add OpenAI actions or source events.
Connect OpenAI
In Build > Integrations > OpenAI, select Connect OpenAI. Enter the
OpenAI API key, then select Save credentials. For an API connection,
send api_key to POST /v1/integrations/providers/openai/connect. The provider
ID is openai.
The assistant model picker currently includes these OpenAI choices:
| Model | ID | Guidance shown in Mobius |
|---|---|---|
| GPT-6 Astra | gpt-6-astra | Highest capability |
| GPT-6.1 Sol | gpt-6.1-sol | Strong reasoning and coding |
| GPT-6 Luna | gpt-6-luna | Fast and efficient |
GPT-6 Luna is the Mobius default for an assistant with no saved model. GPT-6 Sol and GPT-5.6 models can remain on existing assignments but are not offered for new choices. Your plan can narrow the models your organization may run.
Check the source before a real turn
The model catalog labels an available provider byok or platform. With
byok, calls use the connected OpenAI account's quota. With platform, the
turn must also pass Mobius plan and credit checks at execution time. This is why
a populated model menu alone is not proof that a test turn will complete.
Run a short test in a new conversation. If it fails, inspect the turn error. A funding or plan error calls for account or billing work; a credential error calls for reconnecting the OpenAI account.
An assistant's saved model applies to conversations without an override. If a person picks a different model while creating a conversation, Mobius stores that choice on the conversation. Changing the assistant later does not rewrite that pinned choice.
See Integrations for the connection model and Assistant instructions and settings for model defaults and overrides.