My Conversation Flow Agent fails every time when I test it from the Begin node.
Error: Failed to get LLM response — ClientApiError: Internal Server Error
I have refreshed, logged out and back in, tried Chrome Incognito, and retried on another day. The same error continues.
The agent was created using Conductor and has not been published yet.
Can someone please advise what is causing this or check whether it is an account/agent issue?
That error is coming from Retell’s side, not your browser. The LLM call is failing before anything gets back to you, which is why incognito and a fresh login changed nothing.
Being unpublished is not the cause either. The test panel runs against the draft on the canvas, so an unpublished flow should test fine.
Three things worth ruling out before you chase support:
- Check Global Settings and confirm the Language Model actually has a model selected. Conductor built flows sometimes land on a model the account cannot use.
- Open the Dynamic Variables dropdown and set a test value for every variable your global prompt or first node references. I am not certain an empty one throws a 500, but it takes ten seconds to rule out.
- Detach the knowledge base and pull any custom functions off the first node, then retest. A malformed function schema or a knowledge base that failed to index will break the LLM call.
If none of that shifts it, build a blank conversation flow with one plain conversation node and test that. If the blank one works, the problem sits in the Conductor flow. If it fails too, it is account level and only Retell support can see it. Send them your agent_id, the conversation_flow_id, and the exact time and timezone of a failed run. Nobody here can look at accounts.
Russell
amplifyautomation.ai
Hey @info41
Thanks for sharing the details.
Could you please provide the Agent ID and a recent Call ID where you encountered this error? This will help us review the agent and call details to investigate the issue further.
Thank you!