<!-- library-gap-audit: deepgram-python-sdk -->
Gap
The Deepgram Python SDK (deepgram-sdk on PyPI) has zero Braintrust instrumentation in this repo — no integration directory, wrapper, patcher, or auto_instrument() support. Deepgram is a voice-AI platform whose SDK spans speech-to-text, text-to-speech, and a full conversational Voice Agent loop; this repo already instruments comparable voice/agent-orchestration surfaces (LiveKit Agents, Pipecat) and has an open, unaddressed gap for ElevenLabs' voice SDK, but Deepgram — one of the highest-download libraries in this space — is not covered at all.
Weekly downloads: 657,610 (as of 2026-08-31; https://pypistats.org/packages/deepgram-sdk)
What is missing
Both DeepgramClient (sync) and AsyncDeepgramClient (async) expose the following execution surfaces, none of which are traced:
client.listen.v1.media.transcribe_file(...) — prerecorded/batch speech-to-text execution; response exposed via response.results.channels[0].alternatives[0].transcript.client.listen.v2.connect(...) / client.listen.v1.connect(...) — streaming/live transcription over WebSocket (connection.start_listening(), event handlers for OPEN/MESSAGE/CLOSE/ERROR).client.speak.v1.audio.generate(text=...) — text-to-speech generation.client.agent.v1.connect(...) — the Voice Agent API: a full conversational agent loop combining listen (STT) + think (LLM) + speak (TTS) providers, configured via AgentV1Settings/AgentV1SettingsAgent and driven with agent.send_settings(...) / agent.start_listening(). This is the library's agent-execution entry point and is analogous in scope to the agent-run surfaces this repo already traces for LiveKit Agents and Pipecat.client.read.v1.text.analyze(...) — LLM-backed text analysis (e.g. sentiment) over a transcript.
Braintrust docs status
not_found — Deepgram is not mentioned anywhere on https://www.braintrust.dev/docs/guides/tracing/integrations (checked 2026-08-31), which lists LiveKit Agents and Pipecat among voice/agent integrations but has no Deepgram entry.
Upstream sources
Braintrust docs sources checked
Local repo files inspected
py/src/braintrust/integrations/__init__.py — no Deepgram entry in the integration registrypy/src/braintrust/integrations/ — no deepgram/ directory (compare to existing livekit_agents/ and pipecat/ voice-oriented integrations)py/src/braintrust/wrappers/ — no Deepgram wrapperpy/src/braintrust/integrations/versioning.py — no Deepgram version entrypy/pyproject.toml — no deepgram-sdk entry in [tool.braintrust.matrix]py/noxfile.py — no test_deepgram session- Case-insensitive grep for
deepgram across py/src/braintrust/ — zero matches
<!-- library-gap-audit: deepgram-python-sdk -->Gap
The Deepgram Python SDK (
deepgram-sdkon PyPI) has zero Braintrust instrumentation in this repo — no integration directory, wrapper, patcher, orauto_instrument()support. Deepgram is a voice-AI platform whose SDK spans speech-to-text, text-to-speech, and a full conversational Voice Agent loop; this repo already instruments comparable voice/agent-orchestration surfaces (LiveKit Agents, Pipecat) and has an open, unaddressed gap for ElevenLabs' voice SDK, but Deepgram — one of the highest-download libraries in this space — is not covered at all.Weekly downloads: 657,610 (as of 2026-08-31; https://pypistats.org/packages/deepgram-sdk)
What is missing
Both
DeepgramClient(sync) andAsyncDeepgramClient(async) expose the following execution surfaces, none of which are traced:client.listen.v1.media.transcribe_file(...)— prerecorded/batch speech-to-text execution; response exposed viaresponse.results.channels[0].alternatives[0].transcript.client.listen.v2.connect(...)/client.listen.v1.connect(...)— streaming/live transcription over WebSocket (connection.start_listening(), event handlers forOPEN/MESSAGE/CLOSE/ERROR).client.speak.v1.audio.generate(text=...)— text-to-speech generation.client.agent.v1.connect(...)— the Voice Agent API: a full conversational agent loop combining listen (STT) + think (LLM) + speak (TTS) providers, configured viaAgentV1Settings/AgentV1SettingsAgentand driven withagent.send_settings(...)/agent.start_listening(). This is the library's agent-execution entry point and is analogous in scope to the agent-run surfaces this repo already traces for LiveKit Agents and Pipecat.client.read.v1.text.analyze(...)— LLM-backed text analysis (e.g. sentiment) over a transcript.Braintrust docs status
not_found— Deepgram is not mentioned anywhere on https://www.braintrust.dev/docs/guides/tracing/integrations (checked 2026-08-31), which lists LiveKit Agents and Pipecat among voice/agent integrations but has no Deepgram entry.Upstream sources
Braintrust docs sources checked
Local repo files inspected
py/src/braintrust/integrations/__init__.py— no Deepgram entry in the integration registrypy/src/braintrust/integrations/— nodeepgram/directory (compare to existinglivekit_agents/andpipecat/voice-oriented integrations)py/src/braintrust/wrappers/— no Deepgram wrapperpy/src/braintrust/integrations/versioning.py— no Deepgram version entrypy/pyproject.toml— nodeepgram-sdkentry in[tool.braintrust.matrix]py/noxfile.py— notest_deepgramsessiondeepgramacrosspy/src/braintrust/— zero matches