Generic OpenAI-compatible tools
Editors and agents such as Cline, Aider, and Continue can connect through their custom OpenAI-compatible provider option. The labels differ, but the values are consistent.
Valeurs à saisir
| Field in the tool | NexoAI value |
|---|---|
| Provider | OpenAI-compatible, custom OpenAI, or equivalent |
| API base | https://api.nexoai.ma/v1 |
| API key | Your dashboard-issued sk-… virtual key |
| Model | A current NexoAI model ID, such as gpt-5.4 |
If a tool explicitly says its base field should exclude the API version, enter https://api.nexoai.ma. Otherwise, custom OpenAI provider fields commonly expect the versioned /v1 base.
Cline
In Cline's provider settings:
- Choose its OpenAI-compatible provider option.
- Enter
https://api.nexoai.ma/v1as the base URL. - Paste a named NexoAI key into the API key field.
- Enter a catalog ID such as
claude-opus-4-8in the model field. - Save, start a small task, and check key usage in the dashboard.
Use a dedicated key such as cline-local so you can give the editor a specific budget and revoke it independently.
Aider
Aider versions and model backends can use different names for the custom OpenAI base. Configure the values through Aider's documented config or environment variables, mapping its OpenAI API base to https://api.nexoai.ma/v1, its OpenAI API key to the NexoAI key, and its model to a NexoAI catalog ID.
Run Aider's configuration diagnostic before a large session. If it displays a vendor host rather than api.nexoai.ma, stop and correct the provider.
Continue
Create a custom model entry using Continue's OpenAI provider adapter. Set the API base to https://api.nexoai.ma/v1, reference the NexoAI key from Continue's secret storage, and use the exact model ID. Avoid hard-coding the key in a repository-level Continue config.
Other clients
The same mapping works when a client provides all three of these controls:
- a custom OpenAI-compatible base URL;
- a user-supplied API key;
- a free-form model ID.
If any one is fixed to a vendor value, the client may not be suitable for NexoAI. A product that only offers “Sign in with OpenAI” cannot be redirected by setting a shell variable.
Test the transport first
Use the curl chat-completions request with the same key and model. If curl succeeds but the tool fails, the remaining issue is its local provider configuration.
Keep editor configs out of Git
Many tools support project-level settings. Store the base and model there if useful, but keep the actual key in user-level secret storage or an environment variable.
