API Subagent Error

Hello, we have a bot that sends client information via an API endpoint to an email and separate software tool. However, recently when calls were made, we get this error: Axios Error: error code: ECONNABORTED, error message: timeout of 12000ms exceeded.

Independently from this, though from the same node/subagent, we noticed the agent hangs as if it is waiting for a response from the user before making the transfer, causing inactivity as long as neither the bot and/or the client doesn’t respond. Is there a reason for this error that we are not aware of?

Thanks.

Hello @angel.rivera Here are the two separate issues explained:

1. ECONNABORTED / timeout of 12000ms exceeded
This is happening on your external API call , not within Retell itself. Your server is taking longer than 12 seconds to respond. The Code Node has a configurable timeout (default 30s, adjustable 5–60s), but the 12s timeout is set on your Axios client side. You’ll need to either increase the Axios timeout or optimize your external endpoint’s response time.

2. Agent hanging before transfer
Have you enabled Speak During Execution so the agent talks while waiting, preventing the silence/inactivity loop.

If Yes can you share the Call ID and Agent ID ?

Thank You

Yes, that was enabled, but there wasn’t a prompt or phrase saved. Here is the call ID: call_0f3847e76b0923c8e58a365c5bf

Hey @angel.rivera Escalated with team, We’ll get back to you as soon as we have an update.

Best regards

Hello @angel.rivera These are two independent issues on the same node:

A 12s literal means either (a) the timeout_ms on the Send_SmartAdvocate_Lead tool has been set to 12000, or (b) an axios client inside your SmartAdvocate endpoint is using its own ~12s timeout when calling SmartAdvocate downstream. Can you check the tool’s configured Response Timeout in the dashboard, and confirm whether the ECONNABORTED is thrown by Retell or by your own endpoint’s logs? That determines which side to fix. On the call you sent (call_0f3847e76b0923c8e58a365c5bf) the tool actually returned 200 in 1.5s, so the error is happening on other, slower runs.

The mid-call hang is a flow-config issue. Subagent nodes are conversational and wait for caller input after they speak, even after a tool completes. On this call the tool succeeded at 18:01:50 and then the flow sat silent for 10 minutes until inactivity closed it, no transfer_call or end_call was ever invoked. Fix: either convert the node to a Function node (auto-transitions after the tool result), or keep it as Subagent but attach a transfer_call/end_call tool and an Always outgoing edge so the flow actually leaves the node.

Thank You