We’re running an outbound Conversation Flow agent (agent_id: agent_b7ef8160c126dbb63014868e59) that calls drivers. We’ve had a real-world report of calls dropping partway through, and pulling the call data shows a pattern that looks like a silent disconnect being misreported as user_hangup, not an actual hangup on the far end.
Two example calls, same recipient, same day:
Call 1 (call_d5269550c7c066bd8f4589e9288): duration 8 seconds, disconnection_reason: user_hangup. The transcript shows the agent’s own greeting cut off mid-word (“I’m reaching out about an order that was” — nothing after that). There is no user turn anywhere in the transcript, no audio, no words, from the far end at all.
Call 2 (call_eb8190506410fae47726bcfe787), same driver, roughly an hour later: duration 32 seconds, disconnection_reason: user_hangup. This time the agent completes its full opening line and asks a question. The call then ends with, again, zero captured user speech anywhere in the transcript.
Neither call shows any tool call, error, or webhook failure on our side, both terminated before our flow’s logic ever branched. in_voicemail is false on both, and both have recording_url/public_log_url populated if that helps with investigation.
This looks consistent with a few other threads we found searching the forum describing the same shape of problem, calls logged as user_hangup where the actual audio shows a silent transport-level failure rather than a real hangup:
- “Agent hangup in the middle of the conversation”
- “Calls Dropped at the LiveKit Level”
Could someone take a look at these two calls specifically and confirm whether this is the same known issue, and if there’s a fix or workaround available? Happy to provide additional call IDs from the same agent if that’s useful, we can reproduce this pattern somewhat regularly on outbound calls.
Thanks.
