We use the nexos.ai AI Gateway instead of connecting every application directly to individual AI providers. Other organizations may use different commercial or self-hosted AI gateways.
An AI gateway allows an organization to centrally manage model routing, costs, fallback models, security policies, logging, and provider access. For example, a simple classification request can be routed to a lower-cost model, while a more complex request can be handled by a more capable model.
It would be very useful if the Intelligence extension supported a provider-independent gateway configuration.
Suggested functionality
A generic provider interface would avoid creating a separate EspoCRM integration for every AI gateway. It would also help organizations reduce vendor lock-in and apply their own cost, privacy, availability, and governance policies.
Is there currently a provider adapter or extension point that could be used to implement this without modifying the Intelligence extension itself? If not, please consider adding a generic OpenAI-compatible or custom HTTP provider.
An AI gateway allows an organization to centrally manage model routing, costs, fallback models, security policies, logging, and provider access. For example, a simple classification request can be routed to a lower-cost model, while a more complex request can be handled by a more capable model.
It would be very useful if the Intelligence extension supported a provider-independent gateway configuration.
Suggested functionality
- Add a “Custom AI Provider” or “OpenAI-Compatible Provider” option.
- Allow administrators to configure:
- API base URL;
- API key;
- custom HTTP headers;
- model identifier or model alias;
- optional organization, project, or tenant identifiers.
- Allow model identifiers to be entered manually instead of restricting them to a predefined provider model list.
- Support gateway-side model aliases, such as email-classification, data-extraction, or email-composer, while the gateway decides which actual model should process the request.
- Allow different endpoints or model aliases to be selected for different Intelligence features.
- Provide a connection-testing function in the administration interface.
- Preserve EspoCRM usage logging and quota functionality when requests are sent through a gateway.
- Where supported by the gateway response, record token usage, cost, latency, and the actual model selected by the gateway.
A generic provider interface would avoid creating a separate EspoCRM integration for every AI gateway. It would also help organizations reduce vendor lock-in and apply their own cost, privacy, availability, and governance policies.
Is there currently a provider adapter or extension point that could be used to implement this without modifying the Intelligence extension itself? If not, please consider adding a generic OpenAI-compatible or custom HTTP provider.
Comment