starters

Deepgram Starter Apps

Clone a working demo and start building. Every starter is a minimal, runnable app you can extend.

1. Pick Your Feature

What do you want to build?

• Transcribe a file → transcription — send audio/video, get text back (REST, Nova)
• Transcribe a live stream → live-transcription — real-time speech-to-text (WebSocket, Nova)
• Generate speech → text-to-speech — send text, get audio back (REST, Aura)
• Stream speech → live-text-to-speech — real-time text-to-audio (WebSocket, Aura)
• Analyze text or audio → text-intelligence — sentiment, topics, intents, summaries (REST)
• Build a voice agent → voice-agent — conversational AI agent (WebSocket, agent.deepgram.com)
• Conversational STT with turn detection → flux — Deepgram Flux STT for voice agents and interactive assistants (WebSocket, /v2/listen)
• Turn-based TTS for a voice agent → flux-tts — Deepgram Flux TTS, streaming synthesis with barge-in (WebSocket, /v2/speak)

Nova vs Flux STT for speech-to-text: use transcription or live-transcription (Nova, /v1/listen) for general-purpose transcription, captions, and batch workloads. Use flux (Flux STT, /v2/listen) when you need built-in turn detection for conversational audio. See the api skill for a full comparison.

Aura vs Flux TTS for text-to-speech: use text-to-speech or live-text-to-speech (Aura, /v1/speak) for one-shot synthesis, non-English voices, and compressed audio. Use flux-tts (Flux TTS, /v2/speak) when you're streaming LLM output to a speaker and need a turn lifecycle and barge-in. See the api skill for a full comparison.

Flux TTS starters exist for node, flask, fastapi, django, and java only — these are the five apps Deepgram officially publishes at Flux TTS template apps. There is no flux-tts starter for the other frameworks; don't construct those URLs. For an unsupported framework, start from the api skill's Flux TTS section and the SDK skills instead.

2. Pick Your Stack

3. Clone and Run

Every starter lives at https://github.com/deepgram-starters/{framework}-{feature}:

git clone https://github.com/deepgram-starters/{framework}-{feature}.git
cd {framework}-{feature}

Set your API key and follow the README:

export DEEPGRAM_API_KEY=your_key_here

Get an API key at https://console.deepgram.com.

Examples

"I want to build a voice agent in Python" → git clone https://github.com/deepgram-starters/fastapi-voice-agent.git

"I need live transcription in my Node app" → git clone https://github.com/deepgram-starters/node-live-transcription.git

"I want to add text-to-speech to my Go service" → git clone https://github.com/deepgram-starters/go-text-to-speech.git

"I want to analyze audio for sentiment in C#" → git clone https://github.com/deepgram-starters/csharp-text-intelligence.git

"I want streaming TTS with barge-in for my Node voice agent" → git clone https://github.com/deepgram-starters/node-flux-tts.git

All Starters

Need something more specific?

• Focused feature snippets (one feature, one language, < 50 lines) → recipes skill → https://github.com/deepgram/recipes
• Third-party integrations (Twilio, LiveKit, LangChain, Vercel AI SDK, Discord, etc.) → examples skill → https://github.com/deepgram/examples
• SDK-specific code skills (idiomatic imports, async patterns, gotchas) → npx skills add deepgram/deepgram-{lang}-sdk — see the api skill for the full list of 9 SDKs.

Related Deepgram skills

• api — consolidated REST + WebSocket API reference
• recipes — minimal runnable feature snippets per language
• examples — full integration examples with third-party platforms
• docs — documentation finder
• setup-mcp — Deepgram MCP server installation