Custom OpenAI-Compatible Provider

Use AI » Custom for a service that implements an OpenAI-compatible API at a custom base URL.

Examples include a private gateway, a hosted compatible provider, or a local service that exposes the required OpenAI-compatible endpoints.

Connect the provider

1. Open Administration > Integrations > AI » Custom.
2. Enable the integration.
3. Enter the provider's /v1 Base URL.
4. Enter an API key when the service requires bearer authentication.
5. Click Test & Fetch Models without saving the form.
6. Select a model and test again to verify generation.
7. Save.
8. Select Custom under Administration > AI Settings > General when it should be the default provider.

Base URL rules

Use an HTTP or HTTPS URL without:

• Embedded username or password
• Query parameters
• A fragment
• A model-specific path

Example:

https://ai-gateway.example.com/v1

The API key is optional only when the service does not require authentication.

Compatibility

“OpenAI-compatible” does not guarantee support for every OpenAI feature. Ebla AI keeps its own provider adapter authoritative and does not assume native internet search.

Verify the exact service and model for:

• Model listing
• Text generation
• Tool calling
• Structured output
• PDF or image input
• Context and output limits

Unknown and private models remain selectable even when external model information is unavailable.

Troubleshooting

• Confirm the URL reaches the provider from the EspoCRM server.
• Confirm the URL ends at the provider's compatible API root, commonly /v1.
• Run Test & Fetch Models before saving.
• If models load but generation fails, select a model and run the test again.
• Review AI Log. Credentials are redacted from logs and must never be placed in the URL.