The Retell Community is where innovators, call center leaders, and AI enthusiasts come together to share ideas, learn best practices, and explore the future of voice automation. Join to connect with peers, get early access to product updates, and collaborate on building human-like AI voice experiences that transform customer service.

Retell AI

When trying to use Claude I get this error everytime. It happens with other models too. Only working with gemini 3.0 flash. Agent id: agent_2b6bbe1a81cabbe1c348d95a39

Try using gpt 4.1/5 as they are more fast and recommend

For my use case I have achieved better results with that model. Is there a way to know why it happens? How can I be sure my customers are not going to have problems because of the model connection? If I have a subscription with azure/amazon/gcp can I just connect that endpoint to serve as llm?

You can send agent and some call ids where you faced that problem

call id call_c23e4bdb5b8dc58477e748ead8d agent id agent_2b6bbe1a81cabbe1c348d95a39

another call id call_02e851563defe67e814b167c074

Ok I report this to team to look and what they say about this

Hi <@1470807379540054272> 
“Failed to get LLM response” often results from function-calling or formatting issues. The error I see is "400 Invalid schema for function 'jira_create_issue': [None, 'null'] is not valid under any of the given schemas." 
Can you confirm this or give more details

I do not have more info, thats what the logs say

Even though it says something about the tool, the error is way before calling it

Hi <@1405365110880338062> could you give me an update? I understand the error could be related to the mcp, but it happens before calling the tool. The error appears in the first message I send to the agent. I only say hi

I check with the team and get to you <@1470807379540054272>

Thanks! I finally solved it. Apparently OpenAI and Anthropic does not like null as a valid schema for tools so I had to change the schema of the mcp by myself. Just so if something similar happen to any of your users, thanks!