Support custom model base_url / BYOK for the agent's inference endpoint?

[nets9836]

Is there any plan to let the SDK point the agent's underlying model at a custom OpenAI-compatible (or Anthropic-compatible) base_url + API key instead of only Cursor-hosted models?

[HerShin5]

Same ask here. For SDKs/agents the cleanest shape is usually just model + API key + base_url, and then the caller can decide whether that points to Cursor-hosted, OpenAI, Anthropic-compatible, or a gateway. That also makes BYOK/testing much easier because you can swap providers without changing agent code.

For example, I’ve been testing OpenAI-compatible gateways with https://kiroforge.cloud/v1; as long as the client respects base_url, the agent layer doesn’t need to care whether the backend is direct OpenAI or routed/fallback. So +1 for exposing this at the agent inference config level rather than only at provider-specific wrappers.