How to make OnlyOffice directly access the AI URL instead of AI-proxy when calling the AI model?

[zzjjelle]

I have deployed OnlyOffice in an internal network using Docker on Linux. The API for the AI model uses the Qwen3 model deployed on the internal network. When I use the AI feature, I found that OnlyOffice accesses ai-proxy, which prevents the AI model from being called properly. How can I change it to directly access the AI model’s URL instead of ai-proxy?

[Constantine]

Hello @zzjjelle

@zzjjelle

When I use the AI feature, I found that OnlyOffice accesses ai-proxy, which prevents the AI model from being called properly.

Can you elaborate what exactly being called improperly?

In general, this service is used to keep intermediate user data (for instance, API keys) inside the app, i.e. without making them easily interceptable on the device or from a browser request (if Document Server is used).