Hi everyone, me and my colleague have been trying out custom APIs to select models for our agents on Retool. For this case we used Alibaba Cloud API and Infereence APIs for Qwen models, but only one model worked with out issue (Qwen 3 Instruct I believe), others all yielded problematic responses.
Is Retool planning to enable the APIs for Qwen models in the future? Has anyone had this issue or similar that may share with us their insights on this matter? Thanks.
Agents ai
Hi! @Enming_Zhang Have you tried adding Qwen as a custom AI provider? We were able to test it out on our end by adding Qwen as the custom provider, and it worked! Connect a custom AI provider | Retool Docs
Let me know how it goes! 
Hi Chi, yes, that’s exactly what we did, but when the Qwen-based agent uses tools there were errors like this, we were wondering about the root cause of this and would MCP solve it?
Hey @Enming_Zhang, thanks for the update! The tool call failures with Qwen are likely related to the fact that custom model providers aren't officially supported for Retool Agents. Retool Agents are built and tested against providers that follow standard API contracts (like OpenAI and Anthropic). Custom providers like Qwen may have implementation differences that cause intermittent tool call failures, the engineering team has seen the same pattern with other custom providers like Groq and DeepSeek.
For reliable tool call behavior, we would highly recommend switching to one of the officially supported model providers. Let me know if that's helpful at all! Thanks for understanding 