Instead of only provisioning antrophic-models also provision /openapi/ with Responses API, /litellm for other models, and /gemini for gemini models.
Also IIRC the litellm endpoint is based on orchestration so it could require something like this, to avoid breaking in the presence of prompt templates: https://github.com/jerome-benoit/sap-ai-provider/blob/8ed89f83a7fecc06bf66feb1f3ccdd8148729c34/src/convert-to-sap-messages.ts#L367-L383
There is other ways to avoid this but likely not via hai.
Instead of only provisioning antrophic-models also provision
/openapi/with Responses API,/litellmfor other models, and/geminifor gemini models.Also IIRC the litellm endpoint is based on orchestration so it could require something like this, to avoid breaking in the presence of prompt templates: https://github.com/jerome-benoit/sap-ai-provider/blob/8ed89f83a7fecc06bf66feb1f3ccdd8148729c34/src/convert-to-sap-messages.ts#L367-L383
There is other ways to avoid this but likely not via hai.