Summary
On Telegram direct chats, inbound voice transcription succeeds reliably, but the separate echoTranscript message is intermittent.
When the failure happens, logs show:
media: echo-transcript delivery failed: Error: Outbound not configured for channel: telegram
Important: ordinary Telegram outbound messaging still works in the same runtime, so this is not a full Telegram outage.
Reproduction
- Configure Telegram audio transcription with:
tools.media.audio.enabled: truetools.media.audio.echoTranscript: true
- Run OpenClaw on
2026.3.23-2
- Send Telegram voice notes in a direct chat
- Observe that STT succeeds every time, but transcript echo is inconsistent
Expected behavior
If audio transcription succeeds and echoTranscript is enabled, OpenClaw should always deliver the transcript echo back to the same Telegram chat.
Actual behavior
- STT succeeds
- transcript reaches the agent/session
- but transcript echo is flaky:
- sometimes:
media: echo-transcript sent to telegram/telegram:<chat_id>
- other times:
media: echo-transcript delivery failed: Error: Outbound not configured for channel: telegram
Key evidence
Manual foreground run works
Running manually as the openclaw user in verbose mode:
sudo -u openclaw -H bash -lc 'openclaw gateway run --verbose'
Then sending a Telegram voice note produced:
Media understanding audio: success (1/1) via openai/gpt-4o-mini-transcribe
media: echo-transcript sent to telegram/telegram:<chat_id>
Normal service mode fails intermittently
With the systemd service running normally, the same voice-note flow produced:
Media understanding audio: success (1/1) via openai/gpt-4o-mini-transcribe
media: echo-transcript delivery failed: Error: Outbound not configured for channel: telegram
In the same service runtime, normal Telegram outbound sends were still succeeding, e.g.:
telegram sendMessage ok chat=<chat_id> message=<message_id>
So the bug appears specific to the echoTranscript outbound resolution path, not general Telegram connectivity.
Environment
- OpenClaw:
2026.3.23-2
- Install method:
pnpm
- Channel: Telegram direct chat
- Audio model observed in logs:
openai/gpt-4o-mini-transcribe
- Service mode: systemd
- Also tested in manual foreground verbose run
Likely direction
This may be a context/account/outbound-resolution bug in the echoTranscript path, or a runtime/plugin-registry mismatch that affects the echo send path but not ordinary Telegram outbound.
Possibly related:
Summary
On Telegram direct chats, inbound voice transcription succeeds reliably, but the separate
echoTranscriptmessage is intermittent.When the failure happens, logs show:
Important: ordinary Telegram outbound messaging still works in the same runtime, so this is not a full Telegram outage.
Reproduction
tools.media.audio.enabled: truetools.media.audio.echoTranscript: true2026.3.23-2Expected behavior
If audio transcription succeeds and
echoTranscriptis enabled, OpenClaw should always deliver the transcript echo back to the same Telegram chat.Actual behavior
Key evidence
Manual foreground run works
Running manually as the
openclawuser in verbose mode:sudo -u openclaw -H bash -lc 'openclaw gateway run --verbose'Then sending a Telegram voice note produced:
Normal service mode fails intermittently
With the systemd service running normally, the same voice-note flow produced:
In the same service runtime, normal Telegram outbound sends were still succeeding, e.g.:
So the bug appears specific to the
echoTranscriptoutbound resolution path, not general Telegram connectivity.Environment
2026.3.23-2pnpmopenai/gpt-4o-mini-transcribeLikely direction
This may be a context/account/outbound-resolution bug in the
echoTranscriptpath, or a runtime/plugin-registry mismatch that affects the echo send path but not ordinary Telegram outbound.Possibly related:
echoTranscriptdelivery)