Account: pierre.gte@gmail.com (Free trial workspace)
Agent ID: agent_211674004d94795971f3babaf5
Conversation Flow ID: conversation_flow_7469a29c0a04
We have a “Transfer To Team” flow: a global condition-triggered Conversation node (transfer_request) speaks an acknowledgment, then has an unconditional “Always” edge into a Branch node (equation-based), which leads into a Function node (Transfer Call 1, wait_for_result=true, speak_during_execution=true) that calls our own backend to dial out via Telnyx.
Across multiple live test calls, we see total silence between the acknowledgment line finishing and the Function tool actually being invoked – confirmed via GET /v2/get-call’s tool_calls[].start_time_sec, which shows the function starting many seconds (sometimes over a minute) after the node that should lead into it via an unconditional edge.
Two concrete examples:
- call_id: call_a77f32091bf6a80093d0551f667
- 8.34s-11.55s agent: “Sure, I’ll connect you with James – just a moment.”
-
- Total silence until 23.14s, when Retell’s own silence-reengagement fires: “Hi, are you still there?”
-
- User replies “Yes, you can connect me with James, please” at 26.63s
-
- tool_calls shows transfer_call_1 only starts at 29.119s (latency 8962ms)
-
- ~18s of dead air, and the function is provably not invoked until after the caller re-confirms verbally, despite the edge being unconditional (“Always”).
-
- call_id: call_466fc025b0b6fa8f33de20a5335
-
- 10.75s-14.24s agent: “Sure, I’ll connect you with James. Just a moment.”
-
- 26.25s-30.61s agent (unprompted): “Sorry to keep you waiting – I’m still working on connecting you with James.”
-
- Total silence until 113.25s, when the USER says “Hello? Are you still there?”
-
- tool_calls shows transfer_call_1 only starts at 115.134s (latency 15069ms, hit our 15s ring timeout, no answer)
-
- ~85s of dead air, with no caller input required to explain the delay (caller only spoke once, near the end).
- We’ve already tried:
-
- Rewording the acknowledgment node’s instruction (no effect)
-
- Setting speak_during_execution=true on the Function node with a filler line (no effect – per the timestamps above the Function node hasn’t even been entered yet during the silence)
- Our agent is on Rigid Mode (Transition Flexibility). I found an older resolved thread here on the forum (“Rigid Mode Latency Issue - 5000ms vs Normal 1300ms in Flex Mode”) noting Rigid Mode carries extra latency vs Flex Mode, and that Flex Mode is recommended for flows under 20 nodes – ours has 16. That may be a related contributing factor, though our delays (12-85s) are far larger than the 5000ms reported there. I also found a similar unresolved report (“Node not advancing even though transition conditions were met”) describing the same category of stuck-node behavior, so this doesn’t seem to be isolated to our account.
- Could someone from Retell take a look at these two call_ids and let us know what’s causing the gap between the node’s speech finishing and the next tool actually being invoked? Happy to share the full conversation flow JSON or call transcripts if useful.