{ "attribution": "Each entry links to the original public post and names the reporter's public handle. Quotes are at most 30 words. Stack Overflow content is part of the Stack Exchange Network, licensed CC BY-SA 4.0.", "counts": { "merged": 0, "patterns": 48, "queued_for_review": 358, "reports": 440, "withdrawn": 18 }, "data_as_of": "2026-08-27T18:10:49Z", "entries": [ { "activity": 5.813411, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "assemblyai", "azure-speech", "deepgram", "elevenlabs", "livekit", "openai-realtime", "pipecat", "silero", "twilio" ] }, "atlas_id": "VFA-2026-0001", "content_digest": "sha256:d4645a0f0620c222f17fdbc428d5e52274ae402e732d2d76a0b12913ed68d5b9", "first_reported": "2026-01-14T01:19:59Z", "first_seen": "2026-07-15T10:59:40.675Z", "hotato_coverage": { "command": "hotato autopsy ./call.wav", "detector_kind": "overlap_while_agent_talking", "measured": true, "note": "Printed as BARGE-IN: the caller took the floor and the agent kept speaking, reported with the overlap in seconds. Needs the caller and agent on separate channels." }, "id": "pat_d4645a0f0620c222f17fdbc4", "independent_sources": 3, "last_reported": "2026-08-20T17:02:40Z", "last_seen": "2026-08-20T17:02:40Z", "pattern_class": "barge-in-teardown-race", "references": [ { "date": "2026-08-20", "handle": "superMDguy", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6929" }, { "date": "2026-08-15", "handle": "MahmoudHousam", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5337" }, { "date": "2026-07-30", "handle": "dtran26", "role": "report", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2167" }, { "date": "2026-07-25", "handle": "kyle_shechtman", "role": "report", "source": "livekit_forum", "url": "https://community.livekit.io/t/node-1-5-3-pre-playout-interruption-wedges-current-speech-later-llm-tts-succeed-but-the-call-stays-silent/1728" }, { "date": "2026-07-24", "handle": "arsenieciprian", "role": "report", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2108" }, { "date": "2026-07-21", "handle": "xhdrq", "role": "report", "source": "github", "url": "https://github.com/Azure-Samples/cognitive-services-speech-sdk/issues/3056" }, { "date": "2026-07-19", "handle": "Raysharr", "role": "report", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2065" }, { "date": "2026-07-12", "handle": "thecarlmartin", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6391" }, { "date": "2026-06-01", "handle": "timofey-TK", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/4611" }, { "date": "2026-03-27", "handle": "OlliTapio", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat-flows/issues/246" }, { "date": "2026-01-14", "handle": "R_Student", "role": "report", "source": "stackexchange", "url": "https://stackoverflow.com/questions/79867418/openai-realtime-proper-way-to-truncate-a-live-streaming-conversation-on-speech-i" } ], "report_count": 11, "reports": [ { "body_sha256": "5cd240b7f75bafe04331fd6981a12f72a512c3883b7276c87c958e0fc604dd19", "content_license": "", "fetched_at": "2026-08-20T17:03:54Z", "first_seen": "2026-08-20T17:02:40Z", "handle": "superMDguy", "handle_url": "https://github.com/superMDguy", "quote": "...in #5844, I created a load testing script that has a lot of artificial interruptions. During a recent test, it surfaced another error: ```text Error in _main_task Traceback (most recent...", "source": "github", "url": "https://github.com/livekit/agents/issues/6929" }, { "body_sha256": "aa5b60aa445b367dfb182bfff9cba43dff6aa8f58963b0a930df84ea0d416cab", "content_license": "", "fetched_at": "2026-08-15T14:03:57Z", "first_seen": "2026-08-15T13:16:22Z", "handle": "MahmoudHousam", "handle_url": "https://github.com/MahmoudHousam", "quote": "...same LLMSwitcher Feedback node sets LLMConfigureOutputFrame(skip_tts=True) so markdown is not spoken We also queue InterruptionWorkerFrame + flush_pipeline(timeout=2.0) before set_node_from_config Observed Bot spoke the last examiner question after the feedback node...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5337" }, { "body_sha256": "51b46fa19ac8b92891ca0bdc24bf3d448758fb18c0396b2ad269923019b66b80", "content_license": "", "fetched_at": "2026-07-30T14:26:01Z", "first_seen": "2026-07-30T04:05:22Z", "handle": "dtran26", "handle_url": "https://github.com/dtran26", "quote": "...`callerHangupInstruction`. Because that generation still has the consultation persona, briefing context, and the agent's interrupted turn, the model can finish or repeat the briefing before complying with the hangup instruction....", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2167" }, { "body_sha256": "567777915ec2711a494d76301348b5d5bb402f2dfbb8fe1d141a74f9c56e292e", "content_license": "", "fetched_at": "2026-07-25T19:13:16Z", "first_seen": "2026-07-25T18:58:59.839Z", "handle": "kyle_shechtman", "handle_url": "https://community.livekit.io/u/kyle_shechtman", "quote": "...speech handle remained current for approximately 293 seconds and was repeatedly targeted by subsequent interruptions. During that period, eight later LLM and TTS operations completed successfully, but there were no...", "source": "livekit_forum", "url": "https://community.livekit.io/t/node-1-5-3-pre-playout-interruption-wedges-current-speech-later-llm-tts-succeed-but-the-call-stays-silent/1728" }, { "body_sha256": "152a8e13216dffc996d6da205dee5d3684fc835bb7d0335091166824e009514a", "content_license": "", "fetched_at": "2026-07-25T17:50:27Z", "first_seen": "2026-07-24T11:33:42Z", "handle": "arsenieciprian", "handle_url": "https://github.com/arsenieciprian", "quote": "## Summary When the user interrupts the agent (barge-in) in a cascaded `AgentSession` using the Google plugin (Vertex AI), the truncated assistant message is committed to the chat context **after**...", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2108" }, { "body_sha256": "04ea39c23559c56a28eef27b149f1a1a392fc735a7ee281eb7ffdd596ffb46ce", "content_license": "", "fetched_at": "2026-08-04T20:03:52Z", "first_seen": "2026-07-21T06:11:43Z", "handle": "xhdrq", "handle_url": "https://github.com/xhdrq", "quote": "...**SIGSEGV / SEGV_ACCERR** - Crashing frame: `_platform_memmove` ← `MicrosoftCognitiveServicesSpeech` (`GetModuleObject+…`, stripped) - Observed while interrupting / restarting synthesis (barge-in style): `stopSpeaking()` followed quickly by `startSpeakingSsml(...)` on the same `SPXSpeechSynthesizer` At...", "source": "github", "url": "https://github.com/Azure-Samples/cognitive-services-speech-sdk/issues/3056" }, { "body_sha256": "cc6b78bc59ee42a31d71551487bf2571f97fc40e27bbbc754cb07fbe002119f2", "content_license": "", "fetched_at": "2026-07-25T19:11:42Z", "first_seen": "2026-07-19T08:11:08Z", "handle": "Raysharr", "handle_url": "https://github.com/Raysharr", "quote": "### Describe the bug In `@livekit/agents` 1.5.3, when a pipeline reply is interrupted before its playout has started, `AgentActivity.mainTask` can hang forever on the interrupted speech handle's `generation` future. `_currentSpeech`...", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2065" }, { "body_sha256": "fb48d33249c3cc8f1a1b9c972ed96ccc67a32145fff08f7b7037ace61b15c976", "content_license": "", "fetched_at": "2026-07-26T22:06:27Z", "first_seen": "2026-07-12T06:14:32Z", "handle": "thecarlmartin", "handle_url": "https://github.com/thecarlmartin", "quote": "### Bug Description On a `livekit-plugins-xai` realtime call, an interruption can desync the plugin's local `RemoteChatContext` from xAI's server-side item list. Once that happens, **every subsequent `conversation.item.added` is dropped**, because...", "source": "github", "url": "https://github.com/livekit/agents/issues/6391" }, { "body_sha256": "4891230123214ad19df9d8d472036a522928b4cc15f8bb4baf02da5ceda394ea", "content_license": "", "fetched_at": "2026-07-25T17:50:31Z", "first_seen": "2026-06-01T13:11:32Z", "handle": "timofey-TK", "handle_url": "https://github.com/timofey-TK", "quote": "...single `_aggregation` buffer. That buffer is only cleared by `push_aggregation()` (on `LLMFullResponseEndFrame`) and by `_handle_interruptions()`. The turn-start path does **not** clear it: ```python async def _handle_llm_start(self, _: LLMFullResponseStartFrame): await self._trigger_assistant_turn_started()...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/4611" }, { "body_sha256": "c4f08a35b3de8b4acca2fd58732a046ac5dfaf6a6adf1a1f6b8b68e00e3df757", "content_license": "", "fetched_at": "2026-07-25T17:50:33Z", "first_seen": "2026-03-27T13:26:34Z", "handle": "OlliTapio", "handle_url": "https://github.com/OlliTapio", "quote": "...function_name=\"save_notes\", arguments={}, ) # Simulate: FunctionCallsStartedFrame adds orphan entry aggregator._function_calls_in_progress[call_1.tool_call_id] = None # Simulate: InterruptionFrame arrives — no FunctionCallCancelFrame emitted # aggregator._function_calls_in_progress still has {\"call_1\": None} # Turn 2: LLM...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat-flows/issues/246" }, { "body_sha256": "d2ac04d9774ae9aa41e8eb2b3353438131e1551f4b1e3388528b4fac09257ae0", "content_license": "CC BY-SA 4.0", "fetched_at": "2026-07-24T20:54:01Z", "first_seen": "2026-01-14T01:19:59Z", "handle": "R_Student", "handle_url": "https://stackoverflow.com/users/15587184/r-student", "quote": "OpenAI Realtime Proper way to truncate a live streaming conversation on speech interruption Twilio media streams", "source": "stackexchange", "url": "https://stackoverflow.com/questions/79867418/openai-realtime-proper-way-to-truncate-a-live-streaming-conversation-on-speech-i" } ], "stacks": [ "assemblyai", "azure-speech", "deepgram", "elevenlabs", "livekit", "openai-realtime", "pipecat", "silero", "twilio" ], "statement": "When the caller interrupts the agent mid-speech, the system's interrupt handling fails to cleanly tear down the in-progress turn, corrupting the conversation state.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-07-26T08:28:10Z", "evidence_date": "2026-07-02T22:26:02Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat-flows/issues/246", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/barge-in-teardown-race/", "withdrawn": false }, { "activity": 3.708961, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "azure-speech", "cartesia", "deepgram", "elevenlabs", "kokoro", "livekit", "nemo", "pipecat", "silero", "twilio", "whisper" ] }, "atlas_id": "VFA-2026-0002", "content_digest": "sha256:4be4cb5af54e4a4aaa9ff1a4d0a837211d90735f9dea2da06e96a9bbe6b4d80a", "first_reported": "2025-04-05T07:42:33Z", "first_seen": "2026-07-21T08:56:59Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_4be4cb5af54e4a4aaa9ff1a4", "independent_sources": 3, "last_reported": "2026-08-06T07:35:56Z", "last_seen": "2026-08-06T07:35:56Z", "pattern_class": "endpoint-misdetection", "references": [ { "date": "2026-08-06", "handle": "Nasrullah-Ahmed", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6726" }, { "date": "2026-08-02", "handle": "whitepepper355", "role": "report", "source": "github", "url": "https://github.com/huggingface/speech-to-speech/issues/412" }, { "date": "2026-07-14", "handle": "Propconnect-ai", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6415" }, { "date": "2026-07-12", "handle": "Pulkit0729", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5006" }, { "date": "2026-06-10", "handle": "Marc_Torrellas_Socastro", "role": "report", "source": "livekit_forum", "url": "https://community.livekit.io/t/warning-stt-end-of-speech-received-while-vad-is-still-in-a-speech-segment-flushing-vad/1399" }, { "date": "2026-06-03", "handle": "TimeHorizon2100", "role": "report", "source": "github", "url": "https://github.com/twilio/twilio-python/issues/927" }, { "date": "2026-03-03", "handle": "evara-ai", "role": "report", "source": "hn", "url": "https://news.ycombinator.com/item?id=47230233" }, { "date": "2026-02-16", "handle": "nowittyusername", "role": "report", "source": "hn", "url": "https://news.ycombinator.com/item?id=47029721" }, { "date": "2025-11-25", "handle": "IngLP", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/4087" }, { "date": "2025-04-05", "handle": "Jeeva-MV", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/1893" } ], "report_count": 10, "reports": [ { "body_sha256": "83863ae585498701fff44337b48ecce3c2e7d47050a8a0f31a88ff79ad4a1739", "content_license": "", "fetched_at": "2026-08-06T08:06:38Z", "first_seen": "2026-08-06T07:35:56Z", "handle": "Nasrullah-Ahmed", "handle_url": "https://github.com/Nasrullah-Ahmed", "quote": "### Bug Description With `turn_detection=\"stt\"` and an explicitly configured VAD, `min_endpointing_delay` does not run after the STT's `EndOfTurn` signal. It is measured from the earlier moment the speech itself ended,...", "source": "github", "url": "https://github.com/livekit/agents/issues/6726" }, { "body_sha256": "52d2cfc6ac1d12ec5fa9b5be4e9169637a42b47c01459edc107d002601a13955", "content_license": "", "fetched_at": "2026-08-02T12:10:51Z", "first_seen": "2026-08-02T11:51:55Z", "handle": "whitepepper355", "handle_url": "https://github.com/whitepepper355", "quote": "...after that is dropped as `input-after-final`, even though the user is still speaking and VAD has not detected the end of speech yet. ## Root cause `STT/base_stt_handler.py`'s `before_emit_output` locks a...", "source": "github", "url": "https://github.com/huggingface/speech-to-speech/issues/412" }, { "body_sha256": "e96eff80b1a938ce3017f761e4e03ef07585043a419a9da00556a6aac51d94ad", "content_license": "", "fetched_at": "2026-07-25T17:50:25Z", "first_seen": "2026-07-14T02:46:31Z", "handle": "Propconnect-ai", "handle_url": "https://github.com/Propconnect-ai", "quote": "...report — I may well be misreading the design intent. ### What I observed `turn-detector-v1`, English, `unlikely_threshold: 0.56`, no overrides. **The same word, spoken twice, in two sessions — same...", "source": "github", "url": "https://github.com/livekit/agents/issues/6415" }, { "body_sha256": "4891230123214ad19df9d8d472036a522928b4cc15f8bb4baf02da5ceda394ea", "content_license": "", "fetched_at": "2026-07-25T17:50:31Z", "first_seen": "2026-07-12T05:04:06Z", "handle": "Pulkit0729", "handle_url": "https://github.com/Pulkit0729", "quote": "...await self._handle_prediction_result(prediction) self._turn_complete = state == EndOfTurnState.COMPLETE # Start the STT timeout (adjusted by VAD stop_secs since that time already elapsed) timeout = max(0, self._stt_timeout - self._stop_secs) ``` The comment...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5006" }, { "body_sha256": "29133c36b1e069f573ecbef62174d8e59db9e8a2b083067f6f8fce7420d579bd", "content_license": "", "fetched_at": "2026-07-25T18:25:16Z", "first_seen": "2026-06-10T14:19:55.251Z", "handle": "Marc_Torrellas_Socastro", "handle_url": "https://community.livekit.io/u/Marc_Torrellas_Socastro", "quote": "I’m getting message: “stt end of speech received while vad is still in a speech segment, flushing vad | extras: vad_speech_start_time=1781089362.6959133, flushed=True” as a warning but it seems like a...", "source": "livekit_forum", "url": "https://community.livekit.io/t/warning-stt-end-of-speech-received-while-vad-is-still-in-a-speech-segment-flushing-vad/1399" }, { "body_sha256": "fededd97052cf30030a30d48c47f516e06bb755a8f371c8e2ac0f0e6d5543d70", "content_license": "", "fetched_at": "2026-07-24T21:00:36Z", "first_seen": "2026-06-03T06:05:20Z", "handle": "TimeHorizon2100", "handle_url": "https://github.com/TimeHorizon2100", "quote": "...was filtering silence. - Now forwarding full audio stream including silence to allow model-side VAD/end-of-turn. - Reduced and simplified prompt for faster speech output. - Fixed language typo in prompt...", "source": "github", "url": "https://github.com/twilio/twilio-python/issues/927" }, { "body_sha256": "7b3c299ade3943aa6cd9152eca4c99383643172fb116042b2eec4d0071240b34", "content_license": "", "fetched_at": "2026-07-25T17:50:19Z", "first_seen": "2026-03-03T09:38:40Z", "handle": "evara-ai", "handle_url": "https://news.ycombinator.com/user?id=evara-ai", "quote": "...jumping in mid-sentence or waiting too long. Switching from fixed silence thresholds to semantic end-of-turn detection was night and day. One dimension I'd add: geography matters even more when your...", "source": "hn", "url": "https://news.ycombinator.com/item?id=47230233" }, { "body_sha256": "2c0a1d37a6c52fa76f65be965deb0122284c63c90812885c6a24d7088a770762", "content_license": "", "fetched_at": "2026-07-24T20:53:10Z", "first_seen": "2026-02-16T01:15:56Z", "handle": "nowittyusername", "handle_url": "https://news.ycombinator.com/user?id=nowittyusername", "quote": "...him PTT by default as the problems of the model accurately guessing the end of utterance are just too great. I think it can be solved in the future but,...", "source": "hn", "url": "https://news.ycombinator.com/item?id=47029721" }, { "body_sha256": "37e093a57a47ffa50c1c66a89603120a96c747cb11727ec1739d7e20ab42830e", "content_license": "", "fetched_at": "2026-07-30T14:25:59Z", "first_seen": "2025-11-25T18:37:34Z", "handle": "IngLP", "handle_url": "https://github.com/IngLP", "quote": "...talking, I tried in a room with absolute silence too. ### Expected Behavior Silero VAD commits the turn after detecting user silence. Agent respond to the user message already timely...", "source": "github", "url": "https://github.com/livekit/agents/issues/4087" }, { "body_sha256": "1bbeb9bd6df51b2d9f1440c17db0675949dea732629fb6c478b3d93af9282ef8", "content_license": "", "fetched_at": "2026-07-26T01:11:30Z", "first_seen": "2025-04-05T07:42:33Z", "handle": "Jeeva-MV", "handle_url": "https://github.com/Jeeva-MV", "quote": "...any suggestions on how to tackle this will be helpful. Attaching the STT and VAD configs along with a sample log file for reference. Azure-configs (Tried with multi-lingual detection as...", "source": "github", "url": "https://github.com/livekit/agents/issues/1893" } ], "stacks": [ "azure-speech", "cartesia", "deepgram", "elevenlabs", "kokoro", "livekit", "nemo", "pipecat", "silero", "twilio", "whisper" ], "statement": "End-of-turn detection fires at the wrong moment, so the agent either answers before the caller has finished or waits after they have.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-07-26T08:28:10Z", "evidence_date": "2026-07-26T00:09:39Z", "evidence_quote": "issue closed (not_planned)", "evidence_url": "https://github.com/livekit/agents/issues/1893", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/endpoint-misdetection/", "withdrawn": false }, { "activity": 1.953437, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit", "silero" ] }, "atlas_id": "VFA-2026-0011", "content_digest": "sha256:8534cd0ace8a24cfc4955e32a71e86a1a37c128e0ef75a7a970debfdb78c4e1f", "first_reported": "2026-02-10T10:32:39Z", "first_seen": "2026-02-10T10:32:39Z", "hotato_coverage": { "command": "hotato trust --stereo ./call.wav", "detector_kind": "echo_correlated_activity", "measured": true, "note": "Printed as a crosstalk line carrying the coherence and the lag: the caller channel looks like a lag-shifted copy of the agent's own audio, so the barge-in may be leaked output rather than a caller." }, "id": "pat_8534cd0ace8a24cfc4955e32", "independent_sources": 2, "last_reported": "2026-08-23T20:25:37Z", "last_seen": "2026-08-23T20:25:37Z", "pattern_class": "self-echo-vad-false-trigger", "references": [ { "date": "2026-08-23", "handle": "1967xbill", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6956" }, { "date": "2026-02-10", "handle": "PranayKumarJain", "role": "report", "source": "hn", "url": "https://news.ycombinator.com/item?id=46957802" } ], "report_count": 2, "reports": [ { "body_sha256": "33fe3914c037b2f33368ed36113b5ce0d8cc20c6aa5acb1ae583bbeb7d263e38", "content_license": "", "fetched_at": "2026-08-24T14:42:21Z", "first_seen": "2026-08-23T20:25:37Z", "handle": "1967xbill", "handle_url": "https://github.com/1967xbill", "quote": "...`livekit-plugins-silero==1.6.10` * `livekit-plugins-openai==1.6.10` * WebRTC browser voice session * Silero VAD used for turn detection/interruption ### Checkpoints tested Previous checkpoint: ```text 597d30b3ec076608d059477bb14cfeffdf951bf5cae370d38f65d33bbfe82004 ``` Current bundled checkpoint: ```text 1a153a22f4509e292a94e67d6f9b85e8deb25b4988682b7e174c65279d8788e3 ``` The...", "source": "github", "url": "https://github.com/livekit/agents/issues/6956" }, { "body_sha256": "7b3c299ade3943aa6cd9152eca4c99383643172fb116042b2eec4d0071240b34", "content_license": "", "fetched_at": "2026-07-25T17:50:19Z", "first_seen": "2026-02-10T10:32:39Z", "handle": "PranayKumarJain", "handle_url": "https://news.ycombinator.com/user?id=PranayKumarJain", "quote": "...duplex audio: if you don’t subtract your own TTS audio (or at least gate VAD while TTS is playing), you’ll get false user-starts. Even a crude ‘TTS playing → raise...", "source": "hn", "url": "https://news.ycombinator.com/item?id=46957802" } ], "stacks": [ "livekit", "silero" ], "statement": "The agent's own synthesized speech bleeds into the audio input and voice-activity detection mistakes it for the caller starting to talk, triggering a false interruption.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/self-echo-vad-false-trigger/", "withdrawn": false }, { "activity": 1.847778, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "cartesia", "elevenlabs", "home-assistant", "pipecat", "piper" ] }, "atlas_id": "VFA-2026-0009", "content_digest": "sha256:6956631b1ac909621924b4b6346559ddc532da177d58424ea36954566f771367", "first_reported": "2026-07-14T07:17:44Z", "first_seen": "2026-07-14T07:17:44Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_6956631b1ac909621924b4b6", "independent_sources": 2, "last_reported": "2026-08-10T08:17:10Z", "last_seen": "2026-08-10T08:17:10Z", "pattern_class": "silent-tts-failover-deadlock", "references": [ { "date": "2026-08-10", "handle": "Pulkit0729", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5271" }, { "date": "2026-07-28", "handle": "timofey-TK", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5138" }, { "date": "2026-07-23", "handle": "ninicp", "role": "report", "source": "home_assistant_forum", "url": "https://community.home-assistant.io/t/voice-pe-responds-with-led-ring-but-produces-no-audio-announce-tts-and-play-media-all-stay-silent/1018477" }, { "date": "2026-07-14", "handle": "mannyb223", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5026" } ], "report_count": 4, "reports": [ { "body_sha256": "fa36ab51bc315689578ead2fedd71a3ab6f93d8f0857acfd7b31c1240b9d43fd", "content_license": "", "fetched_at": "2026-08-10T13:49:31Z", "first_seen": "2026-08-10T08:17:10Z", "handle": "Pulkit0729", "handle_url": "https://github.com/Pulkit0729", "quote": "...no BotStartedSpeakingFrame within 3.0s of pausing frame processing (e.g. a TTS context completed with no audio) — force-resuming ERROR | pipecat.processors.frame_processor:push_error_frame:732 - ElevenLabsTTSService#1 error: ElevenLabsTTSService#1 no BotStartedSpeakingFrame within 3.0s of...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5271" }, { "body_sha256": "dfc36bd7998b0f0be917cba8f1972981622e0cca2ee5e255cbe2854d4081b3b9", "content_license": "", "fetched_at": "2026-07-28T11:03:01Z", "first_seen": "2026-07-28T10:08:20Z", "handle": "timofey-TK", "handle_url": "https://github.com/timofey-TK", "quote": "...the turn succeeded, and no `ErrorFrame` is pushed. The bot produces an empty turn (dead air) and nothing is logged, so error accounting, failover wrappers (e.g. `ServiceSwitcherStrategy`), and observability all...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5138" }, { "body_sha256": "aa93f24df4bf0fe5ce9b886379bd7de2df67d6d9b20cbaea3813e3b2198d08b4", "content_license": "", "fetched_at": "2026-07-25T18:22:55Z", "first_seen": "2026-07-23T12:11:57.858Z", "handle": "ninicp", "handle_url": "https://community.home-assistant.io/u/ninicp", "quote": "...follows. assist_satellite.announce Service call succeeds. Satellite changes responding → idle . LED ring activates. No audio. Tested with preannounce: true as well. Direct tts.speak Tested directly against the Voice PE...", "source": "home_assistant_forum", "url": "https://community.home-assistant.io/t/voice-pe-responds-with-led-ring-but-produces-no-audio-announce-tts-and-play-media-all-stay-silent/1018477" }, { "body_sha256": "4891230123214ad19df9d8d472036a522928b4cc15f8bb4baf02da5ceda394ea", "content_license": "", "fetched_at": "2026-07-25T17:50:31Z", "first_seen": "2026-07-14T07:17:44Z", "handle": "mannyb223", "handle_url": "https://github.com/mannyb223", "quote": "...during the incident): - The greeting is \"spoken\" (transcript logged, usage chars counted) but no audio exists. - With `pause_frame_processing`, the TTS service pauses its process task after the utterance...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5026" } ], "stacks": [ "cartesia", "elevenlabs", "home-assistant", "pipecat", "piper" ], "statement": "The agent's speech synthesis silently produces no audio even though it logs the turn as spoken, and the safeguard meant to detect and recover from this never triggers, leaving the caller in permanent silence.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-07-26T08:28:10Z", "evidence_date": "2026-07-24T21:15:54Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/5026", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/silent-tts-failover-deadlock/", "withdrawn": false }, { "activity": 1.723048, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "deepgram", "pipecat", "telnyx" ] }, "atlas_id": "VFA-2026-0032", "content_digest": "sha256:fd38e18e6f9fff32bad8064187e62963babff1d7ffd0cc69b3539b3b0aba0b09", "first_reported": "2026-08-12T13:31:47Z", "first_seen": "2026-08-12T13:31:47Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_fd38e18e6f9fff32bad80641", "independent_sources": 1, "last_reported": "2026-08-25T08:01:38Z", "last_seen": "2026-08-25T08:01:38Z", "pattern_class": "interruption-fallback-overstates-delivered-text", "references": [ { "date": "2026-08-25", "handle": "mannyb223", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5426" }, { "date": "2026-08-12", "handle": "timofey-TK", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5303" } ], "report_count": 2, "reports": [ { "body_sha256": "3de1b74ecb223a702f16e2987287612c243b3087a45b76dd20db8da1bd15b5c2", "content_license": "", "fetched_at": "2026-08-25T08:07:13Z", "first_seen": "2026-08-25T08:01:38Z", "handle": "mannyb223", "handle_url": "https://github.com/mannyb223", "quote": "...by the LLM token stream and end-of-turn finalisation, fully decoupled from playback. - On interruption, the aggregation is committed rather than discarded. On a turn ending with `EndFrame` (e.g. a...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5426" }, { "body_sha256": "a9beea0f91000cc41551fe95cc3be3a01ebf387c1f35289d0f7d2b67c2835a61", "content_license": "", "fetched_at": "2026-08-12T14:04:59Z", "first_seen": "2026-08-12T13:31:47Z", "handle": "timofey-TK", "handle_url": "https://github.com/timofey-TK", "quote": "...completes normally: without it, the assistant context would lose text that was spoken. On interruption, however, the same fallback can add text that was not delivered to the listener. For...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5303" } ], "stacks": [ "deepgram", "pipecat", "telnyx" ], "statement": "When the caller interrupts mid-response and the system lacks word-level delivery timestamps, it falls back to recording the full intended text as spoken, so the conversation history includes words the caller never actually heard.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-08-27T08:31:05Z", "evidence_date": "2026-08-26T17:25:54Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/5426", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/interruption-fallback-overstates-delivered-text/", "withdrawn": false }, { "activity": 1.515717, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit", "silero" ] }, "atlas_id": "VFA-2026-0044", "content_digest": "sha256:d392796b8104bded7e711f6724ecc8a5865a42c8b36f3078c3beb0e9fd9f9bef", "first_reported": "2026-08-27T10:15:30Z", "first_seen": "2026-08-27T10:15:30Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_d392796b8104bded7e711f67", "independent_sources": 1, "last_reported": "2026-08-27T10:15:30Z", "last_seen": "2026-08-27T10:15:30Z", "pattern_class": "vad-retrigger-cancels-committed-turn-delivery", "references": [ { "date": "2026-08-27", "handle": "vignesh-koruverse", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/7010" } ], "report_count": 1, "reports": [ { "body_sha256": "4b07cb3bc61d823f570d0ae853061fcec1ef6ff30369cc5eb279e533767a8e9d", "content_license": "", "fetched_at": "2026-08-27T11:07:31Z", "first_seen": "2026-08-27T10:15:30Z", "handle": "vignesh-koruverse", "handle_url": "https://github.com/vignesh-koruverse", "quote": "...end-of-turn bounce, which sleeps the endpointing delay before delivering the turn. With `turn_detection=\"manual\"`, a VAD `START_OF_SPEECH` inside that window cancels the pending task unconditionally (`audio_recognition.py`, SOS handler). Audio tail around...", "source": "github", "url": "https://github.com/livekit/agents/issues/7010" } ], "stacks": [ "livekit", "silero" ], "statement": "During the short delay before an already-finished user turn is delivered, a fresh VAD speech-start event unconditionally cancels that delivery, so the caller's completed utterance is silently dropped and never reaches the agent.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/vad-retrigger-cancels-committed-turn-delivery/", "withdrawn": false }, { "activity": 1.515717, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "gemini-live", "pipecat" ] }, "atlas_id": "VFA-2026-0045", "content_digest": "sha256:4f9e8ed7828f967210edf1fcb0075ac3d2dd1a4f2123b690bc32aa61a24a0de8", "first_reported": "2026-08-27T10:13:25Z", "first_seen": "2026-08-27T10:13:25Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_4f9e8ed7828f967210edf1fc", "independent_sources": 1, "last_reported": "2026-08-27T10:13:25Z", "last_seen": "2026-08-27T10:13:25Z", "pattern_class": "mid-call-reconnect-drops-buffered-content", "references": [ { "date": "2026-08-27", "handle": "FestoneX", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5465" } ], "report_count": 1, "reports": [ { "body_sha256": "8a4efe0ab3b75a1c3fdfa494985d55094dc3040a6f4ecc9d5f5cd61f29fcfc70", "content_license": "", "fetched_at": "2026-08-27T11:08:05Z", "first_seen": "2026-08-27T10:13:25Z", "handle": "FestoneX", "handle_url": "https://github.com/FestoneX", "quote": "...line diagnosis instead of an afternoon. Env: pipecat main, `gemini-3.1-flash-live-preview`, SmallWebRTC transport, server side VAD, python 3.13 on macOS. Happy to send a PR. The bit I'd want your call...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5465" } ], "stacks": [ "gemini-live", "pipecat" ], "statement": "When the live connection to the speech/language service needs to reconnect mid-call, buffered audio, text, and tool results that were in flight are silently discarded instead of resent or recovered.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/mid-call-reconnect-drops-buffered-content/", "withdrawn": false }, { "activity": 1.288788, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "cartesia", "livekit", "whisper" ] }, "atlas_id": "VFA-2026-0043", "content_digest": "sha256:d68d94bbabf4b4768036b5a45d82101a1fba43176abb4782f9d3a873e4a2d4a2", "first_reported": "2026-08-26T20:30:41Z", "first_seen": "2026-08-26T20:30:41Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_d68d94bbabf4b4768036b5a4", "independent_sources": 1, "last_reported": "2026-08-26T20:30:41Z", "last_seen": "2026-08-26T20:30:41Z", "pattern_class": "amd-dual-stream-desync-stall", "references": [ { "date": "2026-08-26", "handle": "jonahlipsky", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6996" } ], "report_count": 1, "reports": [ { "body_sha256": "c1d0cedd1742116136d91402ae6afa5ccc03b279b3d0fa895fd0d329ca5e311d", "content_license": "", "fetched_at": "2026-08-26T21:05:38Z", "first_seen": "2026-08-26T20:30:41Z", "handle": "jonahlipsky", "handle_url": "https://github.com/jonahlipsky", "quote": "## Summary The AMD classifier consumes two **independent** input streams — a transcript from AMD's own STT and speech-start/end boundaries from the session's VAD. When those streams desync, a live...", "source": "github", "url": "https://github.com/livekit/agents/issues/6996" } ], "stacks": [ "cartesia", "livekit", "whisper" ], "statement": "A call-answered classifier depends on two independently arriving signal streams, and when those streams fall out of sync a caller who answers and greets quickly is never classified in time, so the call stalls until the detection timeout.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/amd-dual-stream-desync-stall/", "withdrawn": false }, { "activity": 1.245731, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "deepgram", "livekit", "openai-realtime", "pipecat", "silero" ] }, "atlas_id": "VFA-2026-0028", "content_digest": "sha256:bb19437bf013452afc5c56f008b2df328e33b81d8f290352ea1dc23993eb59b8", "first_reported": "2026-08-07T17:57:59Z", "first_seen": "2026-08-07T17:57:59Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_bb19437bf013452afc5c56f0", "independent_sources": 1, "last_reported": "2026-08-20T00:20:18Z", "last_seen": "2026-08-20T00:20:18Z", "pattern_class": "audio-gate-cooldown-blocks-endpoint-finalization", "references": [ { "date": "2026-08-20", "handle": "saran-007", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5380" }, { "date": "2026-08-07", "handle": "CampbellMBXJ", "role": "report", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2242" } ], "report_count": 2, "reports": [ { "body_sha256": "e6d6fcc3ee4b30dd7202ba29dce82fe923e0919bf8156a3b7c37883a91fe7ac0", "content_license": "", "fetched_at": "2026-08-20T01:03:10Z", "first_seen": "2026-08-20T00:20:18Z", "handle": "saran-007", "handle_url": "https://github.com/saran-007", "quote": "...context frame, the service never forwards a single byte of caller audio, and with server-VAD turn detection the session deadlocks: - no audio reaches xAI, so server VAD never emits...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5380" }, { "body_sha256": "6aef0a7fdf83dbbd573624aa27202e5bb77e2472e740c51c57c2f3f178eda6e1", "content_license": "", "fetched_at": "2026-08-07T18:11:29Z", "first_seen": "2026-08-07T17:57:59Z", "handle": "CampbellMBXJ", "handle_url": "https://github.com/CampbellMBXJ", "quote": "...the Deepgram plugin sends to the websocket. Once the 1 second cooldown elapses with no audio above its RMS threshold, the plugin stops sending audio entirely (KeepAlive still goes out,...", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2242" } ], "stacks": [ "deepgram", "livekit", "openai-realtime", "pipecat", "silero" ], "statement": "A local audio-level gate stops forwarding the caller's audio to the speech-to-text service after a brief silence cooldown, so the remote signal that would mark the turn finished never arrives and the call stalls.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-08-20T08:32:12Z", "evidence_date": "2026-08-20T02:20:00Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/5380", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/audio-gate-cooldown-blocks-endpoint-finalization/", "withdrawn": false }, { "activity": 1.161692, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "deepgram", "elevenlabs", "livekit", "pipecat", "whisper" ] }, "atlas_id": "VFA-2026-0013", "content_digest": "sha256:786ae449b5decd6ecd7f7c5a388af7b5e4ef1616815d7ad3f33ac75734ae854e", "first_reported": "2026-05-14T15:27:21.632Z", "first_seen": "2026-05-14T15:27:21.632Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_786ae449b5decd6ecd7f7c5a", "independent_sources": 2, "last_reported": "2026-08-07T08:33:32Z", "last_seen": "2026-08-07T08:33:32Z", "pattern_class": "vad-fires-but-stt-yields-no-transcript", "references": [ { "date": "2026-08-07", "handle": "Pulkit0729", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5240" }, { "date": "2026-05-14", "handle": "Ganesh_Krishnan", "role": "report", "source": "livekit_forum", "url": "https://community.livekit.io/t/participant-was-audible-in-room-demo-recording-but-livekit-agent-heard-very-low-audio-and-stt-vad-pipeline-missed-most-of-the-response/1124" } ], "report_count": 2, "reports": [ { "body_sha256": "6ae599f6890eda624872bc36f42d1f13ca7aa8b80970e8320c7614b18c240ec1", "content_license": "", "fetched_at": "2026-08-07T09:10:25Z", "first_seen": "2026-08-07T08:33:32Z", "handle": "Pulkit0729", "handle_url": "https://github.com/Pulkit0729", "quote": "...a cascade pipeline, what is the intended way to handle a user turn where VAD detects speech but STT returns no transcript at all? Right now the bot just goes...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5240" }, { "body_sha256": "5995abed33177b26c2d79f7bb4aefdd7e89a419f893eb3241ce0218054e589ff", "content_license": "", "fetched_at": "2026-07-25T18:25:28Z", "first_seen": "2026-05-14T15:27:21.632Z", "handle": "Ganesh_Krishnan", "handle_url": "https://community.livekit.io/u/Ganesh_Krishnan", "quote": "...usable STT transcripts. Example pattern from my logs: User started speaking STT HEALTH CHECK: VAD detected speech 7s ago but no STT transcripts received for christopher! Nudging. Later, some partial...", "source": "livekit_forum", "url": "https://community.livekit.io/t/participant-was-audible-in-room-demo-recording-but-livekit-agent-heard-very-low-audio-and-stt-vad-pipeline-missed-most-of-the-response/1124" } ], "stacks": [ "deepgram", "elevenlabs", "livekit", "pipecat", "whisper" ], "statement": "Voice-activity detection registers that the caller is speaking, but the speech-to-text stage fails to produce any transcript even though the same audio is clearly audible in a separate recording of the call.", "status": "fixed", "status_history": [ { "changed_at": "2026-07-26T08:28:10Z", "evidence_date": "2026-05-16T20:57:25.130Z", "evidence_quote": "topic marked solved", "evidence_url": "https://community.livekit.io/t/participant-was-audible-in-room-demo-recording-but-livekit-agent-heard-very-low-audio-and-stt-vad-pipeline-missed-most-of-the-response/1124", "from": "open", "to": "fixed" }, { "changed_at": "2026-08-08T08:41:40Z", "evidence_date": "2026-05-16T20:57:25.130Z", "evidence_quote": "topic marked solved", "evidence_url": "https://community.livekit.io/t/participant-was-audible-in-room-demo-recording-but-livekit-agent-heard-very-low-audio-and-stt-vad-pipeline-missed-most-of-the-response/1124", "from": "fixed", "to": "acknowledged" }, { "changed_at": "2026-08-10T09:46:36Z", "evidence_date": "2026-08-10T10:37:27Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/5240", "from": "acknowledged", "to": "fixed" } ], "url": "https://hotato.dev/atlas/vad-fires-but-stt-yields-no-transcript/", "withdrawn": false }, { "activity": 1.148698, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "deepgram", "pipecat", "telnyx" ] }, "atlas_id": "VFA-2026-0042", "content_digest": "sha256:819608553061546ff06d9a22e8d9f05e60911f51ed6153e404341337bee74181", "first_reported": "2026-08-25T08:01:35Z", "first_seen": "2026-08-25T08:01:35Z", "hotato_coverage": { "command": "hotato autopsy ./call.wav", "detector_kind": "overlap_while_agent_talking", "measured": true, "note": "Printed as BARGE-IN: the caller took the floor and the agent kept speaking, reported with the overlap in seconds. Needs the caller and agent on separate channels." }, "id": "pat_819608553061546ff06d9a22", "independent_sources": 1, "last_reported": "2026-08-25T08:01:35Z", "last_seen": "2026-08-25T08:01:35Z", "pattern_class": "muted-greeting-overlap-strands-turn", "references": [ { "date": "2026-08-25", "handle": "mannyb223", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5425" } ], "report_count": 1, "reports": [ { "body_sha256": "3de1b74ecb223a702f16e2987287612c243b3087a45b76dd20db8da1bd15b5c2", "content_license": "", "fetched_at": "2026-08-25T08:07:13Z", "first_seen": "2026-08-25T08:01:35Z", "handle": "mannyb223", "handle_url": "https://github.com/mannyb223", "quote": "User-input muting + Deepgram Flux STT: overlapping speech during muted greeting strands the turn — caller unheard for the rest of the call", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5425" } ], "stacks": [ "deepgram", "pipecat", "telnyx" ], "statement": "When the caller speaks while their microphone is deliberately muted during the agent's opening greeting, the overlap corrupts turn-tracking so the caller is never heard again for the rest of the call.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/muted-greeting-overlap-strands-turn/", "withdrawn": false }, { "activity": 0.965921, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "cartesia", "elevenlabs", "livekit", "pipecat" ] }, "atlas_id": "VFA-2026-0031", "content_digest": "sha256:f3db4c5fde41bdc3745057a63fcf38b57cfe043c6d26b3fa5d57d64351c5d265", "first_reported": "2026-08-11T17:56:36Z", "first_seen": "2026-08-11T17:56:36Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_f3db4c5fde41bdc3745057a6", "independent_sources": 1, "last_reported": "2026-08-12T19:18:06Z", "last_seen": "2026-08-12T19:18:06Z", "pattern_class": "stream-error-recovery-false-success", "references": [ { "date": "2026-08-12", "handle": "SwAt1563", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5305" }, { "date": "2026-08-11", "handle": "RobinChartrand-PennyMac", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6793" } ], "report_count": 2, "reports": [ { "body_sha256": "31f98fef2c87d6b460150fabf7b4f1ce42a5be8d96b3bdcc442762c2582aad22", "content_license": "", "fetched_at": "2026-08-12T20:08:37Z", "first_seen": "2026-08-12T19:18:06Z", "handle": "SwAt1563", "handle_url": "https://github.com/SwAt1563", "quote": "...first attempt. Three separate things went wrong, and together they produced 66 seconds of dead air that only ended because the user spoke first. **1. Text that was never spoken...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5305" }, { "body_sha256": "a8a541413a2b5458b6848966b76c7f9150d108a6dd5c631f4212c084f12ddd42", "content_license": "", "fetched_at": "2026-08-11T18:11:49Z", "first_seen": "2026-08-11T17:56:36Z", "handle": "RobinChartrand-PennyMac", "handle_url": "https://github.com/RobinChartrand-PennyMac", "quote": "...response for the remainder of the call. The end user hears the agent cut off mid-sentence, followed by indefinite silence. In the production log this manifests as `Turn restarted successfully...", "source": "github", "url": "https://github.com/livekit/agents/issues/6793" } ], "stacks": [ "cartesia", "elevenlabs", "livekit", "pipecat" ], "statement": "When the model's response stream errors out mid-turn, the system's retry logic logs a successful restart but never actually resumes speaking, cutting the agent off mid-sentence and leaving the caller silent for the rest of the call.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/stream-error-recovery-false-success/", "withdrawn": false }, { "activity": 0.965921, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "gemini-live", "livekit", "openai-realtime" ] }, "atlas_id": "VFA-2026-0026", "content_digest": "sha256:b2e4ba45acdd09ac156d3edd64c1856eef9e30871a18ab93e62a540b7268561d", "first_reported": "2026-08-05T13:40:34Z", "first_seen": "2026-08-05T13:40:34Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_b2e4ba45acdd09ac156d3edd", "independent_sources": 1, "last_reported": "2026-08-12T08:51:41Z", "last_seen": "2026-08-12T08:51:41Z", "pattern_class": "generation-timeout-not-cancelled-on-early-abort", "references": [ { "date": "2026-08-12", "handle": "areebkhan-tech", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6817" }, { "date": "2026-08-05", "handle": "luce98", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6708" } ], "report_count": 2, "reports": [ { "body_sha256": "becbf4d83faed1da10e21f7b88ac0be7bd7e7a4a6da6254b8f5f22092e1eaf96", "content_license": "", "fetched_at": "2026-08-12T09:03:11Z", "first_seen": "2026-08-12T08:51:41Z", "handle": "areebkhan-tech", "handle_url": "https://github.com/areebkhan-tech", "quote": "...turn detection, two independent writers can start a response: (1) the server itself after VAD end-of-speech, and (2) the client via RealtimeSession.generate_reply(). Two distinct problems: generate_reply() (realtime_model.py, ~L1604) sends response.create...", "source": "github", "url": "https://github.com/livekit/agents/issues/6817" }, { "body_sha256": "a40e4951cb89cfbe0fa1cd2ddc2e6ca0330b4f3cad19ec4e17cb08e1077705bf", "content_license": "", "fetched_at": "2026-08-05T14:05:37Z", "first_seen": "2026-08-05T13:40:34Z", "handle": "luce98", "handle_url": "https://github.com/luce98", "quote": "...a flat 5-second hang before `\"generate_reply timed out waiting for generation_created event.\"` fires — dead air stacked on top of a failure the server reported instantly. **Reproduction** - Native-audio Gemini...", "source": "github", "url": "https://github.com/livekit/agents/issues/6708" } ], "stacks": [ "gemini-live", "livekit", "openai-realtime" ], "statement": "When speech generation fails almost instantly, the agent still waits out its entire fixed timeout before reacting, leaving the caller in silence for seconds longer than the actual failure took.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-08-14T16:11:00Z", "evidence_date": "2026-08-14T02:59:12Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/livekit/agents/issues/6817", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/generation-timeout-not-cancelled-on-early-abort/", "withdrawn": false }, { "activity": 0.953979, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit" ] }, "atlas_id": "VFA-2026-0003", "content_digest": "sha256:2a36d2f9762944e3073b088be25bc1c06ecfdf13375926bddb81b21f307a0ef1", "first_reported": "2026-03-11T06:31:00Z", "first_seen": "2026-07-24T17:53:14.628Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_2a36d2f9762944e3073b088b", "independent_sources": 2, "last_reported": "2026-07-24T17:53:14.628Z", "last_seen": "2026-07-24T17:53:14.628Z", "pattern_class": "vad-endpoint-misdetection", "references": [ { "date": "2026-07-24", "handle": "kyle_shechtman", "role": "report", "source": "livekit_forum", "url": "https://community.livekit.io/t/bundled-inference-vad-false-negative-depends-on-32-ms-window-alignment-and-retained-state/1724" }, { "date": "2026-03-11", "handle": "sampdirector", "role": "report", "source": "github", "url": "https://github.com/snakers4/silero-vad/issues/765" } ], "report_count": 2, "reports": [ { "body_sha256": "0f2dd3dabf348cde0e895cc8a41cee429131b816ca0ff7b8cf7c75d6c44133f9", "content_license": "", "fetched_at": "2026-07-25T17:51:33Z", "first_seen": "2026-07-24T17:53:14.628Z", "handle": "kyle_shechtman", "handle_url": "https://community.livekit.io/u/kyle_shechtman", "quote": "Summary I found that the bundled inference.VAD sometimes fails to detect a clearly audible caller utterance in a telephony recording. The audio was captured correctly, but LiveKit generated no VAD...", "source": "livekit_forum", "url": "https://community.livekit.io/t/bundled-inference-vad-false-negative-depends-on-32-ms-window-alignment-and-retained-state/1724" }, { "body_sha256": "c2e3ce78528cdf45a660738c0cf9aa809c2f5a9b6e6d7a0dda299d66507c145e", "content_license": "", "fetched_at": "2026-07-25T17:51:05Z", "first_seen": "2026-03-11T06:31:00Z", "handle": "sampdirector", "handle_url": "https://github.com/sampdirector", "quote": "## ❓ Questions and Help I am encountering consistent issues where Silero VAD fails to detect speech or exhibits significantly lower sensitivity when processing female voices compared to male voices,...", "source": "github", "url": "https://github.com/snakers4/silero-vad/issues/765" } ], "stacks": [ "livekit" ], "statement": "The system fails to detect the caller's speech because voice-activity detection misjudges frame windowing and retained internal state, so the caller appears silent when they are talking.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/vad-endpoint-misdetection/", "withdrawn": false }, { "activity": 0.905126, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "pipecat" ] }, "atlas_id": "VFA-2026-0024", "content_digest": "sha256:3eb23e05b185cec32e61cf0eb7a6d652c205fad8f8a4721ed1b4e0e6ce7f4551", "first_reported": "2026-02-06T08:50:52Z", "first_seen": "2026-02-06T08:50:52Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_3eb23e05b185cec32e61cf0e", "independent_sources": 1, "last_reported": "2026-08-09T20:54:20Z", "last_seen": "2026-08-09T20:54:20Z", "pattern_class": "tool-call-cancel-race-stuck-in-progress", "references": [ { "date": "2026-08-09", "handle": "scttbnsn", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5263" }, { "date": "2026-02-06", "handle": "OlliTapio", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/3661" } ], "report_count": 2, "reports": [ { "body_sha256": "741fb84adf146351936da1d4c46f451f922055cb5d04c6b801e8a08634a3d081", "content_license": "", "fetched_at": "2026-08-09T21:07:06Z", "first_seen": "2026-08-09T20:54:20Z", "handle": "scttbnsn", "handle_url": "https://github.com/scttbnsn", "quote": "...and calls the real `reset()`, so nothing about the drop is simulated. ```python \"\"\"Repro: interrupting a Flows tts_say/function action leaves the ongoing-actions count stuck.\"\"\" import asyncio from unittest.mock import AsyncMock,...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5263" }, { "body_sha256": "8eb817a4d4602a290f33aec7abbcf25429dce03bea532391b4ed8282fce979ca", "content_license": "", "fetched_at": "2026-08-01T18:10:29Z", "first_seen": "2026-02-06T08:50:52Z", "handle": "OlliTapio", "handle_url": "https://github.com/OlliTapio", "quote": "...version 3.12.12 ### Operating System MacOS 26.2 ### Issue description When we manage to interrupt the function call before the FunctionCallInProgressFrame has been sent, the function call gets stuck IN_PROGRESS...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/3661" } ], "stacks": [ "pipecat" ], "statement": "When the caller interrupts a tool call before the system has finished marking it as started, the call gets permanently stuck flagged as in-progress, stalling the conversation.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-08-17T16:40:03Z", "evidence_date": "2026-04-08T13:49:05Z", "evidence_quote": "issue closed (not_planned)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/3661", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/tool-call-cancel-race-stuck-in-progress/", "withdrawn": false }, { "activity": 0.871269, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "gemini-live", "livekit", "twilio" ] }, "atlas_id": "VFA-2026-0018", "content_digest": "sha256:04bad3017bd38e6cf7c0950eb33bbebea9890fb193b5a034a90d8bd9ed13d54f", "first_reported": "2026-07-27T18:09:32Z", "first_seen": "2026-07-27T18:09:32Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_04bad3017bd38e6cf7c0950e", "independent_sources": 1, "last_reported": "2026-08-07T18:05:08Z", "last_seen": "2026-08-07T18:05:08Z", "pattern_class": "tool-call-mid-utterance-audio-truncation", "references": [ { "date": "2026-08-07", "handle": "shrikanthcodes", "role": "report", "source": "github", "url": "https://github.com/livekit/livekit/issues/4737" }, { "date": "2026-07-27", "handle": "blake-newman", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6569" } ], "report_count": 2, "reports": [ { "body_sha256": "13c5751f05cd9d4f749a0bd1e9e2feb9867faf0d0943e1c3289a39bed62c86af", "content_license": "", "fetched_at": "2026-08-07T18:11:35Z", "first_seen": "2026-08-07T18:05:08Z", "handle": "shrikanthcodes", "handle_url": "https://github.com/shrikanthcodes", "quote": "...in a different release) - Environment: LiveKit Cloud, outbound via Twilio SIP trunk - Voicemail detection is our own LLM-based tool, not LiveKit's AMD (shouldn't be relevant to root cause...", "source": "github", "url": "https://github.com/livekit/livekit/issues/4737" }, { "body_sha256": "bb3a8940412a5e2543012b3e03f363c60af9597f3eac8da06e1125427650775c", "content_license": "", "fetched_at": "2026-07-27T22:10:25Z", "first_seen": "2026-07-27T18:09:32Z", "handle": "blake-newman", "handle_url": "https://github.com/blake-newman", "quote": "...2.12.1 - Python 3.11, `AgentSession` with `RealtimeModel` (Gemini Live over Vertex, native audio), server-side turn detection (`turn_handling=TurnHandlingOptions(turn_detection=\"realtime_llm\")` equivalent — automatic activity detection enabled) ## Summary Gemini Live native-audio models routinely...", "source": "github", "url": "https://github.com/livekit/agents/issues/6569" } ], "stacks": [ "gemini-live", "livekit", "twilio" ], "statement": "When the agent calls a tool while still speaking, the system marks buffered audio as fully played once generation finishes, cutting the agent's speech off mid-word and sometimes stalling the call.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/tool-call-mid-utterance-audio-truncation/", "withdrawn": false }, { "activity": 0.814952, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "elevenlabs", "gemini-live", "pipecat" ] }, "atlas_id": "VFA-2026-0025", "content_digest": "sha256:8d03997ee78bfd3e815179998b98ad17c75a2078f6bde1f43c8566924726f7a0", "first_reported": "2026-08-03T00:27:48Z", "first_seen": "2026-08-03T01:04:23Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_8d03997ee78bfd3e81517999", "independent_sources": 1, "last_reported": "2026-08-03T01:04:23Z", "last_seen": "2026-08-03T01:04:23Z", "pattern_class": "stale-partial-transcript-phantom-interruption", "references": [ { "date": "2026-08-03", "handle": "vahidkowsari", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5197" }, { "date": "2026-08-03", "handle": "vahidkowsari", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5196" } ], "report_count": 2, "reports": [ { "body_sha256": "62753dd55ee73b40ff0b3cf5141e02bb9e4612dab4b03d317ed9212a83823a04", "content_license": "", "fetched_at": "2026-08-03T14:26:18Z", "first_seen": "2026-08-03T01:04:23Z", "handle": "vahidkowsari", "handle_url": "https://github.com/vahidkowsari", "quote": "...after the final — right as the bot starts answering — and triggers an interruption. In our production-like testing every bot reply was cut off within ~1s. ## Suggested fix...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5197" }, { "body_sha256": "62753dd55ee73b40ff0b3cf5141e02bb9e4612dab4b03d317ed9212a83823a04", "content_license": "", "fetched_at": "2026-08-03T14:26:18Z", "first_seen": "2026-08-03T00:27:48Z", "handle": "vahidkowsari", "handle_url": "https://github.com/vahidkowsari", "quote": "...pipeline the duplicate final registers as new user speech and acts as a phantom barge-in, cancelling the in-flight LLM turn — the bot randomly goes silent after a user turn....", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5196" } ], "stacks": [ "elevenlabs", "gemini-live", "pipecat" ], "statement": "A stale partial speech-to-text result gets re-sent after the final transcript and is mistaken for new caller speech, cutting the agent off right as it starts answering.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-08-04T08:49:04Z", "evidence_date": "2026-08-03T20:14:42Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/5196", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/stale-partial-transcript-phantom-interruption/", "withdrawn": false }, { "activity": 0.766431, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "deepgram", "livekit" ] }, "atlas_id": "VFA-2026-0039", "content_digest": "sha256:6102237778de0947bf5a938689936bb0840e5e8ec58f161315e513967315a318", "first_reported": "2026-08-18T14:03:29Z", "first_seen": "2026-08-18T14:03:29Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_6102237778de0947bf5a9386", "independent_sources": 1, "last_reported": "2026-08-18T14:03:29Z", "last_seen": "2026-08-18T14:03:29Z", "pattern_class": "teardown-cancels-trailing-transcript-commit", "references": [ { "date": "2026-08-18", "handle": "jonahlipsky", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6889" } ], "report_count": 1, "reports": [ { "body_sha256": "ee15a43b35b4b8b2a0945621498965bcf26ad7a126d4b6ef7d4c78172a8e5b4c", "content_license": "", "fetched_at": "2026-08-18T14:06:40Z", "first_seen": "2026-08-18T14:03:29Z", "handle": "jonahlipsky", "handle_url": "https://github.com/jonahlipsky", "quote": "...\"the mailbox is full and cannot accept new messages at this time\" and then hung up (SIP BYE) immediately after finishing that sentence. VAD (`user_speaking`) on our side shows ~13...", "source": "github", "url": "https://github.com/livekit/agents/issues/6889" } ], "stacks": [ "deepgram", "livekit" ], "statement": "When a call ends, the system cancels the still-in-progress task finalizing the caller's last speech-to-text result, silently discarding whatever the caller said right before the call closed.", "status": "fixed", "status_history": [ { "changed_at": "2026-08-20T08:32:12Z", "evidence_date": "2026-08-19T09:35:12Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/livekit/agents/issues/6889", "from": "open", "to": "fixed" } ], "url": "https://hotato.dev/atlas/teardown-cancels-trailing-transcript-commit/", "withdrawn": false }, { "activity": 0.71689, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "pipecat" ] }, "atlas_id": "VFA-2026-0036", "content_digest": "sha256:1599e738e491cfce43127c9ebdae30d98dd330a875284c29c2fef9219c868f3d", "first_reported": "2026-08-16T07:55:15Z", "first_seen": "2026-08-16T07:55:15Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_1599e738e491cfce43127c9e", "independent_sources": 1, "last_reported": "2026-08-16T07:55:15Z", "last_seen": "2026-08-16T07:55:15Z", "pattern_class": "orphaned-interrupted-tail-flushed-post-goodbye", "references": [ { "date": "2026-08-16", "handle": "mannyb223", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5341" } ], "report_count": 1, "reports": [ { "body_sha256": "9616869641936ea0674d29b18fb89c0a1ababa3cc0a815bc4c22a4369bc9603c", "content_license": "", "fetched_at": "2026-08-17T16:24:39Z", "first_seen": "2026-08-16T07:55:15Z", "handle": "mannyb223", "handle_url": "https://github.com/mannyb223", "quote": "...— it waits in the buffer for an explicit flush. If a response was interrupted/cancelled, no `LLMFullResponseEndFrame` ever arrives, and that buffered tail sits there for the rest of the...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5341" } ], "stacks": [ "pipecat" ], "statement": "When a response is interrupted before it can be properly closed off, the unspoken leftover text stays buffered and later gets synthesized anyway, so the caller hears a stray fragment after the conversation should have ended.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/orphaned-interrupted-tail-flushed-post-goodbye/", "withdrawn": false }, { "activity": 0.677008, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "pipecat" ] }, "atlas_id": "VFA-2026-0035", "content_digest": "sha256:212e4c4c1f151299614a0f1a71813f40758974c99e73177b31bafc1acb7fc133", "first_reported": "2026-08-14T18:07:57Z", "first_seen": "2026-08-14T18:07:57Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_212e4c4c1f151299614a0f1a", "independent_sources": 1, "last_reported": "2026-08-14T18:07:57Z", "last_seen": "2026-08-14T18:07:57Z", "pattern_class": "voicemail-detected-end-call-frame-dropped", "references": [ { "date": "2026-08-14", "handle": "vykintazo", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5330" } ], "report_count": 1, "reports": [ { "body_sha256": "f5f32be2440a8d42bdb05694301c8d95f53fca91869f7b2045f79b46077b76d3", "content_license": "", "fetched_at": "2026-08-14T19:03:41Z", "first_seen": "2026-08-14T18:07:57Z", "handle": "vykintazo", "handle_url": "https://github.com/vykintazo", "quote": "...System Ubuntu 24.04.4 LTS on WSL2 ### Issue description The on_voicemail_detected example in the Voicemail Detection docs does not end the call. The frame it pushes is silently discarded inside...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5330" } ], "stacks": [ "pipecat" ], "statement": "After correctly detecting a voicemail greeting, the agent's command to hang up the call is silently discarded by internal message handling, so the call never actually ends.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/voicemail-detected-end-call-frame-dropped/", "withdrawn": false }, { "activity": 0.677008, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "pipecat" ] }, "atlas_id": "VFA-2026-0034", "content_digest": "sha256:6a864297563c0103509744409a1d0a91695ce6968978b49f2ae97dbccc19aad7", "first_reported": "2026-08-14T08:59:30Z", "first_seen": "2026-08-14T08:59:30Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_6a864297563c010350974440", "independent_sources": 1, "last_reported": "2026-08-14T08:59:30Z", "last_seen": "2026-08-14T08:59:30Z", "pattern_class": "post-interruption-context-timing-bleed", "references": [ { "date": "2026-08-14", "handle": "timofey-TK", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5325" } ], "report_count": 1, "reports": [ { "body_sha256": "459d3f56cebca7a5ec7532613ec5d4ad4c7de2643ea355cf589e1c32e4f4fb03", "content_license": "", "fetched_at": "2026-08-14T14:12:15Z", "first_seen": "2026-08-14T08:59:30Z", "handle": "timofey-TK", "handle_url": "https://github.com/timofey-TK", "quote": "...explicit synthetic sequence because timestamp transport is asynchronous and old-context events can arrive after interruption. ### Expected Behavior Generation offsets should be scoped by `contextId`. `flushCompleted(A)` must only advance A....", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5325" } ], "stacks": [ "pipecat" ], "statement": "After the caller interrupts the agent, word-timing data from the interrupted turn keeps being applied because generation timing isn't scoped per turn, corrupting the record of which words the caller actually heard.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/post-interruption-context-timing-bleed/", "withdrawn": false }, { "activity": 0.603418, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "gemini-live", "livekit", "openai-realtime" ] }, "atlas_id": "VFA-2026-0029", "content_digest": "sha256:1208db7183c17680facc24a0f4cdc1ac2bdf39872421eb60060e6988e4006cde", "first_reported": "2026-08-09T18:48:25Z", "first_seen": "2026-08-09T18:48:25Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_1208db7183c17680facc24a0", "independent_sources": 1, "last_reported": "2026-08-09T18:48:25Z", "last_seen": "2026-08-09T18:48:25Z", "pattern_class": "unanswered-tool-call-blocks-mic-input", "references": [ { "date": "2026-08-09", "handle": "AdityaSharma2168", "role": "report", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2249" } ], "report_count": 1, "reports": [ { "body_sha256": "6eb8777fa6db0a02a55af030372a899e367cba834fbd2f98091f1c1ad9be38d4", "content_license": "", "fetched_at": "2026-08-09T19:05:29Z", "first_seen": "2026-08-09T18:48:25Z", "handle": "AdityaSharma2168", "handle_url": "https://github.com/AdityaSharma2168", "quote": "...gates **microphone input** on the set of unanswered tool call ids, so a single dropped call silently discards every subsequent audio frame. The session does not error or close —...", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2249" } ], "stacks": [ "gemini-live", "livekit", "openai-realtime" ], "statement": "When a tool call is dropped without a response, the system permanently gates future microphone input on that unanswered call, silently discarding everything the caller says afterward.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/unanswered-tool-call-blocks-mic-input/", "withdrawn": false }, { "activity": 0.58169, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit", "ultravox" ] }, "atlas_id": "VFA-2025-0001", "content_digest": "sha256:d636db67e9409ac2a305794a46dc902bbec8407be69c7f247f693f878da6b4c5", "first_reported": "2025-03-30T23:59:45Z", "first_seen": "2025-03-30T23:59:45Z", "hotato_coverage": { "command": "hotato autopsy ./call.wav", "detector_kind": "long_response_gap", "measured": true, "note": "The silence between the caller finishing and the agent responding. Printed as DEAD AIR at 5 seconds or more, and as LATENCY SPIKE between 2 and 5 seconds." }, "id": "pat_d636db67e9409ac2a305794a", "independent_sources": 2, "last_reported": "2026-04-27T16:18:49.899Z", "last_seen": "2026-04-27T16:18:49.899Z", "pattern_class": "endpointing-latency-short-reply", "references": [ { "date": "2026-04-27", "handle": "Sanchit_Singh", "role": "report", "source": "livekit_forum", "url": "https://community.livekit.io/t/high-turn-detection-latency-issue/958" }, { "date": "2025-03-30", "handle": "rojithaDev", "role": "report", "source": "github", "url": "https://github.com/fixie-ai/ultravox/issues/290" } ], "report_count": 2, "reports": [ { "body_sha256": "ba1655a942c9474b7f7760b83ea10f3d590055365fc1c9f3701347af43deec21", "content_license": "", "fetched_at": "2026-07-25T18:25:37Z", "first_seen": "2026-04-27T16:18:49.899Z", "handle": "Sanchit_Singh", "handle_url": "https://community.livekit.io/u/Sanchit_Singh", "quote": "i’m seeing high eou delay and high turn detection inference i’m using multilingual turn detection model which claims 50-160 ms inference time but in my case it’s around 700-1 sec...", "source": "livekit_forum", "url": "https://community.livekit.io/t/high-turn-detection-latency-issue/958" }, { "body_sha256": "c7399707deeb5c2fc7f1e5d32d248ca242b9351a34756642d17a77424e53e94d", "content_license": "", "fetched_at": "2026-07-25T17:50:37Z", "first_seen": "2025-03-30T23:59:45Z", "handle": "rojithaDev", "handle_url": "https://github.com/rojithaDev", "quote": "Slow responses with short one word replies", "source": "github", "url": "https://github.com/fixie-ai/ultravox/issues/290" } ], "stacks": [ "livekit", "ultravox" ], "statement": "After switching to neural voice-activity detection, the system takes an unusually long time to respond when the caller gives a short one-word reply.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/endpointing-latency-short-reply/", "withdrawn": false }, { "activity": 0.560978, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit" ] }, "atlas_id": "VFA-2026-0027", "content_digest": "sha256:f19bf211c08a9edc36d28ef5b7a4e511bfc8c267948d6bfbf893ac43054547b1", "first_reported": "2026-08-05T04:54:18Z", "first_seen": "2026-08-05T04:54:18Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_f19bf211c08a9edc36d28ef5", "independent_sources": 1, "last_reported": "2026-08-05T04:54:18Z", "last_seen": "2026-08-05T04:54:18Z", "pattern_class": "hold-state-audio-leak-exposes-consult-tools", "references": [ { "date": "2026-08-05", "handle": "dtran26", "role": "report", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2221" } ], "report_count": 1, "reports": [ { "body_sha256": "fa9bf5ea0989be097ee549aa5ea2b2d9c4b3779906237980f5af8762e1f4fac3", "content_license": "", "fetched_at": "2026-08-05T05:02:22Z", "first_seen": "2026-08-05T04:54:18Z", "handle": "dtran26", "handle_url": "https://github.com/dtran26", "quote": "...behavior: once the caller is placed on hold, their audio should not reach STT, turn detection, the LLM, or tools until caller input is re-enabled. Consult instructions and transfer tools...", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2221" } ], "stacks": [ "livekit" ], "statement": "When a caller is placed on hold during a transfer, their audio keeps reaching speech recognition, turn detection, and the model instead of being blocked, exposing tools meant only for the behind-the-scenes consult call.", "status": "fixed", "status_history": [ { "changed_at": "2026-08-07T08:41:54Z", "evidence_date": "2026-08-06T23:48:11Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/livekit/agents-js/issues/2221", "from": "open", "to": "fixed" } ], "url": "https://hotato.dev/atlas/hold-state-audio-leak-exposes-consult-tools/", "withdrawn": false }, { "activity": 0.520081, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "elevenlabs" ] }, "atlas_id": "VFA-2026-0022", "content_digest": "sha256:409b0678dc4fec4185ed09ae1a116ef4e7a277af07e937a1e2bfca00465bf87b", "first_reported": "2026-07-31T15:51:12Z", "first_seen": "2026-07-31T15:51:12Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_409b0678dc4fec4185ed09ae", "independent_sources": 1, "last_reported": "2026-07-31T15:51:12Z", "last_seen": "2026-07-31T15:51:12Z", "pattern_class": "interruption-audio-event-id-collision-drop", "references": [ { "date": "2026-07-31", "handle": "saintpete", "role": "report", "source": "github", "url": "https://github.com/elevenlabs/elevenlabs-python/issues/829" } ], "report_count": 1, "reports": [ { "body_sha256": "d04b959933802c403cf0a8df874a9f3be517ea7a6f6aa0f44fa65929891a5010", "content_license": "", "fetched_at": "2026-07-31T16:08:16Z", "first_seen": "2026-07-31T15:51:12Z", "handle": "saintpete", "handle_url": "https://github.com/saintpete", "quote": "...never called. Client-side instrumentation shows why: 1. `send_multimodal_message` lands → the server fires an `interruption` event ~60 ms later. 2. The reply's `audio` events then arrive carrying an `event_id` **equal...", "source": "github", "url": "https://github.com/elevenlabs/elevenlabs-python/issues/829" } ], "stacks": [ "elevenlabs" ], "statement": "An interruption event fired at nearly the same moment as the agent's reply causes the reply's audio to share the interruption's event id, so a staleness filter wrongly discards it and the caller hears nothing.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/interruption-audio-event-id-collision-drop/", "withdrawn": false }, { "activity": 0.50639, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "pipecat" ] }, "atlas_id": "VFA-2026-0021", "content_digest": "sha256:3e6ca7980486864b1c4d60695c003eaf09882bfa9d66496530fa99f4f2f6dff0", "first_reported": "2026-07-29T12:06:58Z", "first_seen": "2026-07-29T12:06:58Z", "hotato_coverage": { "command": "hotato autopsy ./call.wav", "detector_kind": "long_response_gap", "measured": true, "note": "The silence between the caller finishing and the agent responding. Printed as DEAD AIR at 5 seconds or more, and as LATENCY SPIKE between 2 and 5 seconds." }, "id": "pat_3e6ca7980486864b1c4d6069", "independent_sources": 1, "last_reported": "2026-07-29T12:06:58Z", "last_seen": "2026-07-29T12:06:58Z", "pattern_class": "turn-completion-filter-dead-air-timer-orphan", "references": [ { "date": "2026-07-29", "handle": "SwAt1563", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5151" } ], "report_count": 1, "reports": [ { "body_sha256": "912d7b200970396f679696a4649a116982c67efbd1e1512e54eb61d5ed04f7cc", "content_license": "", "fetched_at": "2026-07-29T12:10:43Z", "first_seen": "2026-07-29T12:06:58Z", "handle": "SwAt1563", "handle_url": "https://github.com/SwAt1563", "quote": "...can never fire again. Net result: one low-probability generation puts the session into **unrecoverable dead air** — no timer of any kind is running, and only the human giving up...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5151" } ], "stacks": [ "pipecat" ], "statement": "A turn-completion check accepts a degenerate, textless model output as a valid finish, ending the turn with no reply while also disabling the timer meant to recover from silence, leaving the caller in permanent dead air.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/turn-completion-filter-dead-air-timer-orphan/", "withdrawn": false }, { "activity": 0.5, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "cartesia", "pipecat" ] }, "atlas_id": "VFA-2026-0020", "content_digest": "sha256:8397c709654193afcd3c391cd3fbc09b703f11896af68015416c006aededdd88", "first_reported": "2026-07-28T19:00:19Z", "first_seen": "2026-07-28T19:00:19Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_8397c709654193afcd3c391c", "independent_sources": 1, "last_reported": "2026-07-28T19:00:19Z", "last_seen": "2026-07-28T19:00:19Z", "pattern_class": "idle-reprompt-turn-latch-response-drop", "references": [ { "date": "2026-07-28", "handle": "SwAt1563", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5145" } ], "report_count": 1, "reports": [ { "body_sha256": "0605102922e9d71659dc2b62a1c18e8aec97615e34617671e6ec9654394a656f", "content_license": "", "fetched_at": "2026-07-28T19:10:47Z", "first_seen": "2026-07-28T19:00:19Z", "handle": "SwAt1563", "handle_url": "https://github.com/SwAt1563", "quote": "...from any later inference is dropped. The latch only resets on `UserStartedSpeakingFrame`, `VADUserStartedSpeakingFrame`, or `InterruptionFrame` — i.e. **only when the user speaks**. But re-prompt patterns (most notably the documented user-idle...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5145" } ], "stacks": [ "cartesia", "pipecat" ], "statement": "After the system re-prompts a silent caller, a turn-completion gate that only resets when the caller speaks treats the caller's reply as already answered and silently discards the agent's response.", "status": "fixed", "status_history": [ { "changed_at": "2026-08-01T08:37:35Z", "evidence_date": "2026-07-31T20:40:08Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/5145", "from": "open", "to": "fixed" } ], "url": "https://hotato.dev/atlas/idle-reprompt-turn-latch-response-drop/", "withdrawn": false }, { "activity": 0.488021, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit" ] }, "atlas_id": "VFA-2026-0016", "content_digest": "sha256:e056fac7e6cf3eb1621eff88027176e4a8f38be862961b8bf306717887b52646", "first_reported": "2026-07-26T02:09:49Z", "first_seen": "2026-07-26T02:09:49Z", "hotato_coverage": { "command": "hotato autopsy ./call.wav", "detector_kind": "overlap_while_agent_talking", "measured": true, "note": "Printed as BARGE-IN: the caller took the floor and the agent kept speaking, reported with the overlap in seconds. Needs the caller and agent on separate channels." }, "id": "pat_e056fac7e6cf3eb1621eff88", "independent_sources": 1, "last_reported": "2026-07-26T02:09:49Z", "last_seen": "2026-07-26T02:09:49Z", "pattern_class": "async-interruption-response-misattribution", "references": [ { "date": "2026-07-26", "handle": "toubatbrian", "role": "report", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2119" } ], "report_count": 1, "reports": [ { "body_sha256": "a5091d4d64d4c61e8231c689cf1b9333c6df99db777bc4afcba275d5e13e0df9", "content_license": "", "fetched_at": "2026-07-26T03:05:24Z", "first_seen": "2026-07-26T02:09:49Z", "handle": "toubatbrian", "handle_url": "https://github.com/toubatbrian", "quote": "...is accepted and attributed to that later overlap — which can emit a false `isInterruption: true` and cut off the agent for a user utterance the model never scored that...", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2119" } ], "stacks": [ "livekit" ], "statement": "A delayed decision about whether an earlier overlap counted as an interruption gets wrongly applied to a later, unrelated moment, cutting the agent off for speech that was never actually judged an interruption.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/async-interruption-response-misattribution/", "withdrawn": false }, { "activity": 0.488021, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit" ] }, "atlas_id": "VFA-2026-0017", "content_digest": "sha256:636a4f247f3cbc84d12ab9a9dbf2e7b9d272e21753aaaab2614823763370f7cc", "first_reported": "2026-07-26T02:08:47Z", "first_seen": "2026-07-26T02:08:47Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_636a4f247f3cbc84d12ab9a9", "independent_sources": 1, "last_reported": "2026-07-26T02:08:47Z", "last_seen": "2026-07-26T02:08:47Z", "pattern_class": "segmented-turn-interruption-disarm", "references": [ { "date": "2026-07-26", "handle": "toubatbrian", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6548" } ], "report_count": 1, "reports": [ { "body_sha256": "70a2ab9a2b60732fe1353a1776db01cee8b29160a042509e4a1959580dedfec1", "content_license": "", "fetched_at": "2026-07-26T03:05:22Z", "first_seen": "2026-07-26T02:08:47Z", "handle": "toubatbrian", "handle_url": "https://github.com/toubatbrian", "quote": "### Summary A second agent speech segment inside a single agent turn disarms an adaptive-interruption overlap that the user is still speaking through. From that moment the user's audio stops...", "source": "github", "url": "https://github.com/livekit/agents/issues/6548" } ], "stacks": [ "livekit" ], "statement": "When the agent's turn is split into multiple speech segments, such as around a tool call, a later segment resets interruption tracking so the caller's still-ongoing interruption is silently ignored.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/segmented-turn-interruption-disarm/", "withdrawn": false }, { "activity": 0.482395, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit", "openai-realtime" ] }, "atlas_id": "VFA-2026-0004", "content_digest": "sha256:a15a17a1fb34eb6f5b99cf593991f9165660fe785613cdcc6dfbc1502b35f2e2", "first_reported": "2026-07-25T15:24:52Z", "first_seen": "2026-07-24T23:15:54Z", "hotato_coverage": { "command": "hotato autopsy ./call.wav", "detector_kind": "agent_start_during_caller", "measured": true, "note": "Printed as TALK-OVER: the agent began talking while the caller still held the floor, reported with the overlap in seconds. Needs the caller and agent on separate channels." }, "id": "pat_a15a17a1fb34eb6f5b99cf59", "independent_sources": 1, "last_reported": "2026-07-25T15:24:52Z", "last_seen": "2026-07-25T15:24:52Z", "pattern_class": "agent-interrupts-caller", "references": [ { "date": "2026-07-25", "handle": "socevicius", "role": "report", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2115" } ], "report_count": 1, "reports": [ { "body_sha256": "ab3e4acdc5db8206c316f78425f336b4ca20d080650aeb738fde378e7c3ea49b", "content_license": "", "fetched_at": "2026-07-25T16:04:52Z", "first_seen": "2026-07-25T15:24:52Z", "handle": "socevicius", "handle_url": "https://github.com/socevicius", "quote": "### Summary With an OpenAI Realtime `RealtimeModel`, syncing the chat context after an **interrupted** agent turn that has **skipped** (never-played) outputs throws `Unsupported item type: agent_config_update`. It is caught and...", "source": "github", "url": "https://github.com/livekit/agents-js/issues/2115" } ], "stacks": [ "livekit", "openai-realtime" ], "statement": "The agent starts speaking while the caller still holds the floor, cutting the caller off mid-utterance.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/agent-interrupts-caller/", "withdrawn": false }, { "activity": 0.47699, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "cartesia", "elevenlabs", "pipecat" ] }, "atlas_id": "VFA-2026-0015", "content_digest": "sha256:13dc36af11a51811c065b3a048ecd27a60185c523a61a148dcb0f1f2d2b1d98e", "first_reported": "2026-07-24T22:56:47Z", "first_seen": "2026-07-24T22:56:47Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_13dc36af11a51811c065b3a0", "independent_sources": 1, "last_reported": "2026-07-24T22:56:47Z", "last_seen": "2026-07-24T22:56:47Z", "pattern_class": "tts-word-drop-context-race", "references": [ { "date": "2026-07-24", "handle": "Sarita-Joshi", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5119" } ], "report_count": 1, "reports": [ { "body_sha256": "4891230123214ad19df9d8d472036a522928b4cc15f8bb4baf02da5ceda394ea", "content_license": "", "fetched_at": "2026-07-25T17:50:31Z", "first_seen": "2026-07-24T22:56:47Z", "handle": "Sarita-Joshi", "handle_url": "https://github.com/Sarita-Joshi", "quote": "...word of a sentence is intermittently dropped from the assistant `LLMContext` on completely normal, non-interrupted turns**. The audio is spoken in full — only the stored transcript is short. Because...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5119" } ], "stacks": [ "cartesia", "elevenlabs", "pipecat" ], "statement": "A queue-ordering race in assembling the agent's own spoken output into conversation history drops the turn's final word from the stored record, even on a normal turn with no interruption, though the caller still hears the full sentence.", "status": "fixed", "status_history": [ { "changed_at": "2026-07-26T08:28:10Z", "evidence_date": "2026-07-25T00:36:44Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/5119", "from": "open", "to": "fixed" } ], "url": "https://hotato.dev/atlas/tts-word-drop-context-race/", "withdrawn": false }, { "activity": 0.47699, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "pipecat" ] }, "atlas_id": "VFA-2026-0006", "content_digest": "sha256:361aece86cc7a9daf0c4ab020c6ae08c175afc9761ad16357d5a7260b27e7073", "first_reported": "2026-07-24T09:12:14Z", "first_seen": "2026-07-24T22:10:52Z", "hotato_coverage": { "command": "hotato autopsy ./call.wav", "detector_kind": "long_response_gap", "measured": true, "note": "The silence between the caller finishing and the agent responding. Printed as DEAD AIR at 5 seconds or more, and as LATENCY SPIKE between 2 and 5 seconds." }, "id": "pat_361aece86cc7a9daf0c4ab02", "independent_sources": 1, "last_reported": "2026-07-24T09:12:14Z", "last_seen": "2026-07-24T22:10:52Z", "pattern_class": "dead-air-after-caller", "references": [ { "date": "2026-07-24", "handle": "timofey-TK", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5110" } ], "report_count": 1, "reports": [ { "body_sha256": "4891230123214ad19df9d8d472036a522928b4cc15f8bb4baf02da5ceda394ea", "content_license": "", "fetched_at": "2026-07-25T17:50:31Z", "first_seen": "2026-07-24T09:12:14Z", "handle": "timofey-TK", "handle_url": "https://github.com/timofey-TK", "quote": "...Nothing reaches the transport — on a phone call this is 20–40 s of dead air until the callee says \"Hello?\". Observed on pipecat 1.6.0 (telephony, `auto_mode=True`, default buffer settings...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/5110" } ], "stacks": [ "pipecat" ], "statement": "The caller finishes and the line goes silent for seconds before the agent responds, with no audible acknowledgement in between.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/dead-air-after-caller/", "withdrawn": false }, { "activity": 0.46196, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "cartesia", "deepgram", "livekit" ] }, "atlas_id": "VFA-2026-0007", "content_digest": "sha256:bd14f264e1296384f93642aee678b462ba39237df1fdf86f7b7c446b2eaa2016", "first_reported": "2026-07-21T23:53:03Z", "first_seen": "2026-07-22T20:32:21Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_bd14f264e1296384f93642ae", "independent_sources": 1, "last_reported": "2026-07-21T23:53:03Z", "last_seen": "2026-07-22T20:32:21Z", "pattern_class": "late-stt-final-race-after-turn-flush", "references": [ { "date": "2026-07-21", "handle": "Propconnect-ai", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6504" } ], "report_count": 1, "reports": [ { "body_sha256": "cfd41fb8fe6770e3fa80c3dcba1c70372dd6c5fbfbca936be4703562b3bb22df", "content_license": "", "fetched_at": "2026-07-28T12:08:11Z", "first_seen": "2026-07-21T23:53:03Z", "handle": "Propconnect-ai", "handle_url": "https://github.com/Propconnect-ai", "quote": "...splitting one utterance into two turns ## Summary When a turn is committed by VAD and the final STT transcript arrives shortly afterwards, the late transcript triggers a **second** turn...", "source": "github", "url": "https://github.com/livekit/agents/issues/6504" } ], "stacks": [ "cartesia", "deepgram", "livekit" ], "statement": "A speech-to-text result that finalizes just after the system has already ended and flushed the caller's turn is still processed, racing against the turn boundary and risking corrupted turn state.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/late-stt-final-race-after-turn-flush/", "withdrawn": false }, { "activity": 0.418256, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "pipecat" ] }, "atlas_id": "VFA-2026-0008", "content_digest": "sha256:84c2e63007c368f3aeecc01aca88fe615f33df03e8894ba645d646e9de86bf11", "first_reported": "2026-07-10T00:22:41Z", "first_seen": "2026-07-19T01:36:37Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_84c2e63007c368f3aeecc01a", "independent_sources": 1, "last_reported": "2026-07-10T00:22:41Z", "last_seen": "2026-07-19T01:36:37Z", "pattern_class": "tool-fire-after-cancel", "references": [ { "date": "2026-07-10", "handle": "mannyb223", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/4997" } ], "report_count": 1, "reports": [ { "body_sha256": "4891230123214ad19df9d8d472036a522928b4cc15f8bb4baf02da5ceda394ea", "content_license": "", "fetched_at": "2026-07-25T17:50:31Z", "first_seen": "2026-07-10T00:22:41Z", "handle": "mannyb223", "handle_url": "https://github.com/mannyb223", "quote": "## Description A function call requested by a generation that was **interrupted/discarded** still executes. When that function is terminal — an `end_call` / hangup tool — the bot hangs up...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/4997" } ], "stacks": [ "pipecat" ], "statement": "A tool call the agent decided on before being interrupted still executes afterward, so an action like hanging up can fire even though the caller is still on the line.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-07-26T08:28:10Z", "evidence_date": "2026-07-23T22:14:47Z", "evidence_quote": "issue closed (not_planned)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/4997", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/tool-fire-after-cancel/", "withdrawn": false }, { "activity": 0.40261, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "asterisk", "home-assistant" ] }, "atlas_id": "VFA-2026-0038", "content_digest": "sha256:405e43bb45bb0c5cb3ba840856d26674652f5fe9a39676f4ffaedb8d8a10a199", "first_reported": "2026-07-05T21:48:11Z", "first_seen": "2026-07-05T21:48:11Z", "hotato_coverage": { "command": "hotato autopsy ./call.wav", "detector_kind": "long_response_gap", "measured": true, "note": "The silence between the caller finishing and the agent responding. Printed as DEAD AIR at 5 seconds or more, and as LATENCY SPIKE between 2 and 5 seconds." }, "id": "pat_405e43bb45bb0c5cb3ba8408", "independent_sources": 1, "last_reported": "2026-07-05T21:48:11Z", "last_seen": "2026-07-05T21:48:11Z", "pattern_class": "post-hangup-pipeline-wedge-blocks-future-calls", "references": [ { "date": "2026-07-05", "handle": "26tajeen", "role": "report", "source": "github", "url": "https://github.com/home-assistant/core/issues/175718" } ], "report_count": 1, "reports": [ { "body_sha256": "6af2d599a9d31c806fc1ac10abda4610a7df5f5b831306f1d8aeafc944793e01", "content_license": "", "fetched_at": "2026-08-04T04:01:52Z", "first_seen": "2026-07-05T21:48:11Z", "handle": "26tajeen", "handle_url": "https://github.com/26tajeen", "quote": "...at the SIP level but produce **no listening tone and no pipeline activity** — dead air. The caller hears nothing at all. In the wedged state the logs still show...", "source": "github", "url": "https://github.com/home-assistant/core/issues/175718" } ], "stacks": [ "asterisk", "home-assistant" ], "statement": "After a call is torn down mid-pipeline, the system's call-handling state fails to reset, so later calls are answered but produce complete dead air with no agent activity until the service is restarted.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/post-hangup-pipeline-wedge-blocks-future-calls/", "withdrawn": false }, { "activity": 0.396899, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "pipecat" ] }, "atlas_id": "VFA-2026-0030", "content_digest": "sha256:e24de83e4273dbb7100316f5e50f4ac91bd419f615fdfb967f09999c9d7a6e63", "first_reported": "2026-07-03T12:37:53Z", "first_seen": "2026-07-03T12:37:53Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_e24de83e4273dbb7100316f5", "independent_sources": 1, "last_reported": "2026-07-03T12:37:53Z", "last_seen": "2026-07-03T12:37:53Z", "pattern_class": "tool-call-not-cancelled-on-pipeline-teardown", "references": [ { "date": "2026-07-03", "handle": "anotine10", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/4948" } ], "report_count": 1, "reports": [ { "body_sha256": "d9b076dd05b1df23520c2387acadaabbe260113a7da36820e34c6bb86a7e6f84", "content_license": "", "fetched_at": "2026-08-11T12:08:30Z", "first_seen": "2026-07-03T12:37:53Z", "handle": "anotine10", "handle_url": "https://github.com/anotine10", "quote": "...``` In-flight `_run_function_call` tasks are only cancelled via `_cancel_function_call` / `_cancel_function_calls_by_tool_call_id`, which fire on **interruption** or the async-tool-cancellation tool — never on a pipeline `CancelFrame`. `pipecat/pipeline/worker.py` - Teardown `finally` cancels...", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/4948" } ], "stacks": [ "pipecat" ], "statement": "When the call itself is torn down or cancelled, tool calls still in flight are never cancelled and keep executing in the background after the caller has left the conversation.", "status": "fixed", "status_history": [ { "changed_at": "2026-08-26T08:21:35Z", "evidence_date": "2026-08-11T11:15:58Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/4948", "from": "open", "to": "fixed" } ], "url": "https://hotato.dev/atlas/tool-call-not-cancelled-on-pipeline-teardown/", "withdrawn": false }, { "activity": 0.394148, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "pipecat" ] }, "atlas_id": "VFA-2026-0023", "content_digest": "sha256:322c4bcc269bf13643ba5470b3d46c484698f5dc388bc104f0371f996dc55766", "first_reported": "2026-07-02T18:30:21Z", "first_seen": "2026-07-02T18:30:21Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_322c4bcc269bf13643ba5470", "independent_sources": 1, "last_reported": "2026-07-02T18:30:21Z", "last_seen": "2026-07-02T18:30:21Z", "pattern_class": "tool-call-timeout-zombie-handler", "references": [ { "date": "2026-07-02", "handle": "Sarita-Joshi", "role": "report", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/4936" } ], "report_count": 1, "reports": [ { "body_sha256": "d240c6d29dcd44a92cc7cd48ef6b4cf6b4326931ef400758e7ccd981de37b717", "content_license": "", "fetched_at": "2026-07-31T16:09:20Z", "first_seen": "2026-07-02T18:30:21Z", "handle": "Sarita-Joshi", "handle_url": "https://github.com/Sarita-Joshi", "quote": "Function call timeout doesn't cancel the handler — it keeps running as a zombie, and its late result is silently dropped (double-side-effect risk)", "source": "github", "url": "https://github.com/pipecat-ai/pipecat/issues/4936" } ], "stacks": [ "pipecat" ], "statement": "When a tool call times out, the system stops waiting but never cancels the underlying handler, which keeps executing in the background and can trigger the same real-world action twice.", "status": "fixed", "status_history": [ { "changed_at": "2026-08-26T08:21:35Z", "evidence_date": "2026-08-13T21:17:10Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/pipecat-ai/pipecat/issues/4936", "from": "open", "to": "fixed" } ], "url": "https://hotato.dev/atlas/tool-call-timeout-zombie-handler/", "withdrawn": false }, { "activity": 0.369851, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit", "twilio" ] }, "atlas_id": "VFA-2026-0019", "content_digest": "sha256:6e3cef201759ffed346906ea6e1fb875ba1fb819be358f8612ba2ec379eeed56", "first_reported": "2026-06-22T19:34:32Z", "first_seen": "2026-06-22T19:34:32Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_6e3cef201759ffed346906ea", "independent_sources": 1, "last_reported": "2026-06-22T19:34:32Z", "last_seen": "2026-06-22T19:34:32Z", "pattern_class": "amd-premature-timeout-before-answer", "references": [ { "date": "2026-06-22", "handle": "mirandamon", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6187" } ], "report_count": 1, "reports": [ { "body_sha256": "e404d12836c0a7dfc3719da85fe17d6192b13e268ebc5708ef848da7b6962eb3", "content_license": "", "fetched_at": "2026-07-28T13:08:10Z", "first_seen": "2026-06-22T19:34:32Z", "handle": "mirandamon", "handle_url": "https://github.com/mirandamon", "quote": "### Summary On outbound SIP calls with carrier early media, AMD emits `uncertain` / `detection_timeout` with `speech_duration=0` **before the callee answers**, so the agent falls through and treats a voicemail...", "source": "github", "url": "https://github.com/livekit/agents/issues/6187" } ], "stacks": [ "livekit", "twilio" ], "statement": "On outbound calls, the answering-machine-detection timer starts before the callee has actually answered, so ring or early media triggers a false timeout that the system wrongly treats as a voicemail greeting.", "status": "fixed", "status_history": [ { "changed_at": "2026-08-08T08:41:40Z", "evidence_date": "2026-08-05T13:57:44Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/livekit/agents/issues/6187", "from": "open", "to": "fixed" } ], "url": "https://hotato.dev/atlas/amd-premature-timeout-before-answer/", "withdrawn": false }, { "activity": 0.357494, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "deepgram", "livekit", "openai-agents", "openai-realtime", "silero" ] }, "atlas_id": "VFA-2025-0002", "content_digest": "sha256:15da2b2e712d5062cd9ba7350569c72c4d5dd96b8d2d38449aaa6dcadcd38bcc", "first_reported": "2025-07-17T15:09:47Z", "first_seen": "2025-07-17T15:09:47Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_15da2b2e712d5062cd9ba735", "independent_sources": 1, "last_reported": "2026-02-06T05:57:22Z", "last_seen": "2026-02-06T05:57:22Z", "pattern_class": "tool-call-audio-output-silence", "references": [ { "date": "2026-02-06", "handle": "Jaya-sys", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/4729" }, { "date": "2025-07-17", "handle": "sibblegp", "role": "report", "source": "github", "url": "https://github.com/openai/openai-agents-python/issues/1168" } ], "report_count": 2, "reports": [ { "body_sha256": "ff375d9702ac4e03d0bbe1668a15163628c3544fc7f728e334d714bb574b853a", "content_license": "", "fetched_at": "2026-08-10T13:48:55Z", "first_seen": "2026-02-06T05:57:22Z", "handle": "Jaya-sys", "handle_url": "https://github.com/Jaya-sys", "quote": "...livekit.agents: expected to receive only one AgentTask from the tool executions\"\"\"\" After this error: No audio response is returned No fallback or partial response is produced The call/session becomes silent...", "source": "github", "url": "https://github.com/livekit/agents/issues/4729" }, { "body_sha256": "6b3215c5a3151e14d20d1d682a03cc4f21f9302b1ded8794151e57c3e31640b0", "content_license": "", "fetched_at": "2026-07-26T14:05:22Z", "first_seen": "2025-07-17T15:09:47Z", "handle": "sibblegp", "handle_url": "https://github.com/sibblegp", "quote": "...were those times again?\" then it responds. See this trace: [Imgur](https://imgur.com/a/BCjukuj) Note there is no audio out after the tool calls. ### Debug information - Agents SDK version: v0.2.1 -...", "source": "github", "url": "https://github.com/openai/openai-agents-python/issues/1168" } ], "stacks": [ "deepgram", "livekit", "openai-agents", "openai-realtime", "silero" ], "statement": "After the agent completes one or more tool calls within a turn, its spoken response never gets synthesized, leaving the caller in silence.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-08-02T08:40:27Z", "evidence_date": "2026-08-02T07:04:22Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/openai/openai-agents-python/issues/1168", "from": "open", "to": "fixed" }, { "changed_at": "2026-08-10T09:46:36Z", "evidence_date": "2026-08-02T07:04:22Z", "evidence_quote": "issue closed (completed)", "evidence_url": "https://github.com/openai/openai-agents-python/issues/1168", "from": "fixed", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/tool-call-audio-output-silence/", "withdrawn": false }, { "activity": 0.344855, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit", "twilio" ] }, "atlas_id": "VFA-2026-0012", "content_digest": "sha256:975de5c29c2b6c63233c33e9ecd6e68d69f631fa7dd68f38ed5ce4c6efeb0643", "first_reported": "2026-06-09T20:00:13.948Z", "first_seen": "2026-06-09T20:00:13.948Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_975de5c29c2b6c63233c33e9", "independent_sources": 1, "last_reported": "2026-06-09T20:00:13.948Z", "last_seen": "2026-06-09T20:00:13.948Z", "pattern_class": "agent-dispatch-latency-before-connect", "references": [ { "date": "2026-06-09", "handle": "Zaheer_Abbas", "role": "report", "source": "livekit_forum", "url": "https://community.livekit.io/t/inbound-sip-call-agent-job-dispatched-8s-after-the-invite-reached-livekit-cloud-caller-had-already-hung-up/1393" } ], "report_count": 1, "reports": [ { "body_sha256": "65d16ea3044802b0afa77d88134a48395cba322f3ea33237ae9e662df467a0ea", "content_license": "", "fetched_at": "2026-07-25T18:25:17Z", "first_seen": "2026-06-09T20:00:13.948Z", "handle": "Zaheer_Abbas", "handle_url": "https://community.livekit.io/u/Zaheer_Abbas", "quote": "...was dispatched ~8s after the INVITE — and ~2.5s after the caller had already hung up. Why we think the delay is dispatch-side: this was 1 of 208 inbound calls...", "source": "livekit_forum", "url": "https://community.livekit.io/t/inbound-sip-call-agent-job-dispatched-8s-after-the-invite-reached-livekit-cloud-caller-had-already-hung-up/1393" } ], "stacks": [ "livekit", "twilio" ], "statement": "The agent process is dispatched to an inbound call so many seconds after it arrives that the caller has already hung up before the agent joins.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/agent-dispatch-latency-before-connect/", "withdrawn": false }, { "activity": 0.344855, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "cartesia", "deepgram", "elevenlabs", "livekit" ] }, "atlas_id": "VFA-2026-0041", "content_digest": "sha256:5da48c2f7125f8333bccd4a6f7e4b4a9c11da0ccd06451ce3358e02cbf5a9351", "first_reported": "2026-06-09T06:36:47Z", "first_seen": "2026-06-09T06:36:47Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_5da48c2f7125f8333bccd4a6", "independent_sources": 1, "last_reported": "2026-06-09T06:36:47Z", "last_seen": "2026-06-09T06:36:47Z", "pattern_class": "vad-background-noise-mistaken-for-caller-speech", "references": [ { "date": "2026-06-09", "handle": "abhishekdeepmindz", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/6030" } ], "report_count": 1, "reports": [ { "body_sha256": "8481c4670e2ff326ad0834aa179b4b59a81c2548f2cc4577f50d31133b9970b3", "content_license": "", "fetched_at": "2026-08-17T16:24:05Z", "first_seen": "2026-06-09T06:36:47Z", "handle": "abhishekdeepmindz", "handle_url": "https://github.com/abhishekdeepmindz", "quote": "...background inaudible speech. How does this work internally? Following is my session parameters. `{ \"allow_interruptions\": true, \"discard_audio_if_uninterruptible\": true, \"min_interruption_duration\": 0.3, \"min_interruption_words\": 0, \"min_endpointing_delay\": 0.4, \"max_endpointing_delay\": 2, \"max_tool_steps\": 3, \"user_away_timeout\": 10,...", "source": "github", "url": "https://github.com/livekit/agents/issues/6030" } ], "stacks": [ "cartesia", "deepgram", "elevenlabs", "livekit" ], "statement": "Voice-activity detection mistakes background noise or nearby conversations for the caller talking, so silence-based inactivity timeouts never trigger even though the caller isn't actually speaking.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/vad-background-noise-mistaken-for-caller-speech/", "withdrawn": false }, { "activity": 0.343166, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit" ] }, "atlas_id": "VFA-2026-0040", "content_digest": "sha256:cd1e558d5e28f0f0db8182494b90e9e5fe104a4065ec2e73fdb398532912471a", "first_reported": "2026-06-08T13:04:00Z", "first_seen": "2026-06-08T13:04:00Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_cd1e558d5e28f0f0db818249", "independent_sources": 1, "last_reported": "2026-06-08T13:04:00Z", "last_seen": "2026-06-08T13:04:00Z", "pattern_class": "forced-close-drain-hang-orphans-connection", "references": [ { "date": "2026-06-08", "handle": "chenghao-mou", "role": "report", "source": "github", "url": "https://github.com/livekit/agents-js/issues/1731" } ], "report_count": 1, "reports": [ { "body_sha256": "3434e4daeb8a7cf4d6ac5e90be0a16324e89d96c54e2ff583305b286712432f6", "content_license": "", "fetched_at": "2026-08-21T10:05:13Z", "first_seen": "2026-06-08T13:04:00Z", "handle": "chenghao-mou", "handle_url": "https://github.com/chenghao-mou", "quote": "...left. ## Root cause `closeImplInner` (`agents/src/voice/agent_session.ts`) on a forced close does, in order: 1. `interrupt({ force: true })` 2. `await this.activity.drain()` ← hangs here, **unbounded** 3. `await this.activity.close()` ← never...", "source": "github", "url": "https://github.com/livekit/agents-js/issues/1731" } ], "stacks": [ "livekit" ], "statement": "When a call is forcibly ended, the shutdown sequence that interrupts the agent and drains pending audio/state can hang indefinitely, so the session's speech-recognition connection is never released and the call never fully tears down.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/forced-close-drain-hang-orphans-connection/", "withdrawn": false }, { "activity": 0.313253, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "asterisk" ] }, "atlas_id": "VFA-2026-0037", "content_digest": "sha256:9ab7abaf398d1913ec649d1dd754b0014cda6a80497f5f7d0e1240aa6d98d8db", "first_reported": "2026-05-18T09:40:39.757Z", "first_seen": "2026-05-18T09:40:39.757Z", "hotato_coverage": { "command": "hotato autopsy ./call.wav", "detector_kind": "overlap_while_agent_talking", "measured": true, "note": "Printed as BARGE-IN: the caller took the floor and the agent kept speaking, reported with the overlap in seconds. Needs the caller and agent on separate channels." }, "id": "pat_9ab7abaf398d1913ec649d1d", "independent_sources": 1, "last_reported": "2026-05-18T09:40:39.757Z", "last_seen": "2026-05-18T09:40:39.757Z", "pattern_class": "half-duplex-barge-in-ignored", "references": [ { "date": "2026-05-18", "handle": "shaan754", "role": "report", "source": "asterisk_forum", "url": "https://community.asterisk.org/t/enabling-barge-in-with-audiosocket-connection-between-asterisk-and-python-server/112943" } ], "report_count": 1, "reports": [ { "body_sha256": "f12647c28351e5169f2674c50c5f44c172d3e4f6854abc3c7db92ffec28b1d13", "content_license": "", "fetched_at": "2026-07-25T18:22:26Z", "first_seen": "2026-05-18T09:40:39.757Z", "handle": "shaan754", "handle_url": "https://community.asterisk.org/u/shaan754", "quote": "...is speaking. If a user interrupts mid-response — for example to clarify something or barge in — their speech is ignored completely. Attempted Solution To support full-duplex conversations and barge-in,...", "source": "asterisk_forum", "url": "https://community.asterisk.org/t/enabling-barge-in-with-audiosocket-connection-between-asterisk-and-python-server/112943" } ], "stacks": [ "asterisk" ], "statement": "While the agent is speaking, the system never listens for the caller's speech, so any attempt to interrupt or clarify goes completely unheard until the agent finishes.", "status": "stale", "status_history": [ { "changed_at": "2026-08-22T08:20:30Z", "evidence_date": "2026-08-22T08:40:36Z", "evidence_quote": "the canonical link returned HTTP 404", "evidence_url": "https://community.asterisk.org/t/enabling-barge-in-with-audiosocket-connection-between-asterisk-and-python-server/112943", "from": "open", "to": "stale" } ], "url": "https://hotato.dev/atlas/half-duplex-barge-in-ignored/", "withdrawn": false }, { "activity": 0.305122, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "elevenlabs", "livekit", "openai-agents", "twilio" ] }, "atlas_id": "VFA-2026-0014", "content_digest": "sha256:b3d0433fabd645d7dce58b6ddd107e10733eac0a47c10fa17b4a6cdc546164b2", "first_reported": "2026-05-11T20:37:37.601Z", "first_seen": "2026-05-11T20:37:37.601Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_b3d0433fabd645d7dce58b6d", "independent_sources": 1, "last_reported": "2026-05-11T20:37:37.601Z", "last_seen": "2026-05-11T20:37:37.601Z", "pattern_class": "silence-timeout-teardown-during-processing-gap", "references": [ { "date": "2026-05-11", "handle": "Jorge_Octavio_Nicolas_Diaz", "role": "report", "source": "livekit_forum", "url": "https://community.livekit.io/t/issue-sip-bridge-sends-bye-during-llm-tts-processing-gap-14s-silence-twilio-elastic-sip-trunk-reports-caller-hung-up/1080" } ], "report_count": 1, "reports": [ { "body_sha256": "ce5a494003d5cc8c0bfd009564f1763f7d7c2e179bc5c32189e55ff890acc642", "content_license": "", "fetched_at": "2026-07-25T18:25:50Z", "first_seen": "2026-05-11T20:37:37.601Z", "handle": "Jorge_Octavio_Nicolas_Diaz", "handle_url": "https://community.livekit.io/u/Jorge_Octavio_Nicolas_Diaz", "quote": "...Twilio. Twilio detects RTP inactivity and the call drops. Twilio Voice Insights confirms “Who Hung Up: caller” — meaning LiveKit’s SIP bridge is sending the BYE, not Twilio. Already ruled...", "source": "livekit_forum", "url": "https://community.livekit.io/t/issue-sip-bridge-sends-bye-during-llm-tts-processing-gap-14s-silence-twilio-elastic-sip-trunk-reports-caller-hung-up/1080" } ], "stacks": [ "elevenlabs", "livekit", "openai-agents", "twilio" ], "statement": "When the agent's response takes too long to generate, the telephony layer reads the resulting silence as call inactivity and tears down the call itself, then misreports that the caller hung up.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/silence-timeout-teardown-during-processing-gap/", "withdrawn": false }, { "activity": 0.304019, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "livekit" ] }, "atlas_id": "VFA-2026-0005", "content_digest": "sha256:618665ed104f8d900e2c6c32f5747bd36c73c1a8646eae7926241020f02005ea", "first_reported": "2026-05-10T14:52:59.089Z", "first_seen": "2026-07-24T23:42:24Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_618665ed104f8d900e2c6c32", "independent_sources": 1, "last_reported": "2026-05-10T14:52:59.089Z", "last_seen": "2026-07-24T23:42:24Z", "pattern_class": "vad-false-trigger", "references": [ { "date": "2026-05-10", "handle": "James_Lau", "role": "report", "source": "livekit_forum", "url": "https://community.livekit.io/t/gpt-realtime-2-livekit-vad-does-not-work-well/1072" } ], "report_count": 1, "reports": [ { "body_sha256": "0046654a7015ab904d16af2d360c233825798e01cd58e0b96fc0b379db292863", "content_license": "", "fetched_at": "2026-07-25T18:25:34Z", "first_seen": "2026-05-10T14:52:59.089Z", "handle": "James_Lau", "handle_url": "https://community.livekit.io/u/James_Lau", "quote": "Hi folks, I have started using realtime-2, but the VAD is horrible. Any sort of backchanelling like “uh ha”, “ok” would trigger the agent to stop talking. My current settings...", "source": "livekit_forum", "url": "https://community.livekit.io/t/gpt-realtime-2-livekit-vad-does-not-work-well/1072" } ], "stacks": [ "livekit" ], "statement": "Voice-activity detection treats non-speech or a backchannel as a floor claim, so the agent yields to something that was not a turn.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/vad-false-trigger/", "withdrawn": false }, { "activity": 0.234249, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "assemblyai", "cartesia", "deepgram", "elevenlabs", "livekit", "silero" ] }, "atlas_id": "VFA-2026-0033", "content_digest": "sha256:8e91276db1803978fd7730a05ba082134daa13505dc71e9de2faab82064e985d", "first_reported": "2026-01-28T04:53:03Z", "first_seen": "2026-01-28T04:53:03Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_8e91276db1803978fd7730a0", "independent_sources": 1, "last_reported": "2026-01-28T04:53:03Z", "last_seen": "2026-01-28T04:53:03Z", "pattern_class": "agent-replies-to-own-output", "references": [ { "date": "2026-01-28", "handle": "glOriginMind", "role": "report", "source": "github", "url": "https://github.com/livekit/agents/issues/4634" } ], "report_count": 1, "reports": [ { "body_sha256": "10ed4fe862d004dbdbedde046432090338ba6c48ab1af731b252541e1a6cd930", "content_license": "", "fetched_at": "2026-07-30T01:08:26Z", "first_seen": "2026-01-28T04:53:03Z", "handle": "glOriginMind", "handle_url": "https://github.com/glOriginMind", "quote": "...Models Used Assembly AI, Gemini-2.5-flast-lite; Cartesia ### Package Versions ```bash livekit-agents[google, openai, deepgram, cartesia, turn-detector, silero, azure, elevenlabs, assemblyai]~=1.3.5 livekit-plugins-noise-cancellation~=0.2 ``` ### Session/Room/Call IDs _No response_ ### Proposed Solution ```python...", "source": "github", "url": "https://github.com/livekit/agents/issues/4634" } ], "stacks": [ "assemblyai", "cartesia", "deepgram", "elevenlabs", "livekit", "silero" ], "statement": "The system mistakes the agent's own generated response for new caller input, causing the agent to reply to itself instead of to anything the caller said.", "status": "acknowledged", "status_history": [ { "changed_at": "2026-08-14T16:11:00Z", "evidence_date": "2026-07-30T00:22:00Z", "evidence_quote": "issue closed (not_planned)", "evidence_url": "https://github.com/livekit/agents/issues/4634", "from": "open", "to": "acknowledged" } ], "url": "https://hotato.dev/atlas/agent-replies-to-own-output/", "withdrawn": false }, { "activity": 0.077547, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "twilio" ] }, "atlas_id": "VFA-2026-0010", "content_digest": "sha256:aa7dc93826a1f02c589096377f832678bde9dc3d1c8dc6b3f84ab3d7fa1df200", "first_reported": "2017-05-30T17:13:45Z", "first_seen": "2026-04-23T11:39:55Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_aa7dc93826a1f02c58909637", "independent_sources": 1, "last_reported": "2017-05-30T17:13:45Z", "last_seen": "2026-04-23T11:39:55Z", "pattern_class": "amd-tone-detection-premature-teardown", "references": [ { "date": "2017-05-30", "handle": "user3502355", "role": "report", "source": "stackexchange", "url": "https://stackoverflow.com/questions/44268185/twilio-answering-machine-detection-not-leaving-message" } ], "report_count": 1, "reports": [ { "body_sha256": "9cd2036c86d0980fa72cadef73d90accc28db9f966dbc11c9cb56f9a62d76db5", "content_license": "CC BY-SA 3.0", "fetched_at": "2026-07-25T17:51:26Z", "first_seen": "2017-05-30T17:13:45Z", "handle": "user3502355", "handle_url": "https://stackoverflow.com/users/3502355/user3502355", "quote": "Twilio answering machine detection not leaving message", "source": "stackexchange", "url": "https://stackoverflow.com/questions/44268185/twilio-answering-machine-detection-not-leaving-message" } ], "stacks": [ "twilio" ], "statement": "The system stops listening for the voicemail beep tone as soon as it detects silence after the greeting, so a beep arriving just after is missed and the tone-triggered action never fires.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/amd-tone-detection-premature-teardown/", "withdrawn": false }, { "activity": 0.070029, "affected": { "note": "Stacks this mode has a public report on. The atlas states where it has been reported, which is a different claim from where it can occur.", "stacks_reported_on": [ "twilio" ] }, "atlas_id": "VFA-2014-0001", "content_digest": "sha256:d38ab5cc91316ac1c7f78467a3e32915758302dafcb69a9ddebcd5358a30781f", "first_reported": "2014-09-22T19:12:33Z", "first_seen": "2014-09-22T19:12:33Z", "hotato_coverage": { "command": "", "detector_kind": "", "measured": false, "note": "" }, "id": "pat_d38ab5cc91316ac1c7f78467", "independent_sources": 1, "last_reported": "2014-09-22T19:12:33Z", "last_seen": "2014-09-22T19:12:33Z", "pattern_class": "amd-result-arrives-after-call-answered", "references": [ { "date": "2014-09-22", "handle": "Chris Dutrow", "role": "report", "source": "stackexchange", "url": "https://stackoverflow.com/questions/25981498/answering-machine-detection-amd-with-twilio-but-after-the-fact" } ], "report_count": 1, "reports": [ { "body_sha256": "9cd2036c86d0980fa72cadef73d90accc28db9f966dbc11c9cb56f9a62d76db5", "content_license": "CC BY-SA 4.0", "fetched_at": "2026-07-25T17:51:26Z", "first_seen": "2014-09-22T19:12:33Z", "handle": "Chris Dutrow", "handle_url": "https://stackoverflow.com/users/84131/chris-dutrow", "quote": "Answering Machine Detection (AMD) with Twilio.. But after the fact", "source": "stackexchange", "url": "https://stackoverflow.com/questions/25981498/answering-machine-detection-amd-with-twilio-but-after-the-fact" } ], "stacks": [ "twilio" ], "statement": "The system's answering-machine-vs-human classification only comes back after the call has already been answered and engaged, so the agent cannot act on whether it reached a person or a machine in time.", "status": "open", "status_history": [], "url": "https://hotato.dev/atlas/amd-result-arrives-after-call-answered/", "withdrawn": false } ], "generator": "hotato-atlas 0.1.0", "home": "https://hotato.dev/atlas/", "removal": { "email": "atlas-remove@hotato.dev", "honored_within_hours": 24, "method": "tombstone", "policy_url": "https://hotato.dev/atlas/remove/" }, "schema": "hotato.atlas-feed.v1", "schema_url": "https://hotato.dev/atlas/schema/feed.schema.json", "stack_hues": { "assemblyai": 115, "asterisk": 183, "azure-speech": 238, "cartesia": 145, "deepgram": 206, "elevenlabs": 345, "gemini-live": 256, "home-assistant": 273, "kokoro": 92, "livekit": 3, "nemo": 162, "openai-agents": 291, "openai-realtime": 53, "pipecat": 310, "piper": 329, "silero": 221, "telnyx": 19, "twilio": 38, "ultravox": 72, "whisper": 131 }, "title": "The Voice Failure Atlas" }