Providers

LYRE is provider-agnostic. Mix and match STT, LLM, and TTS providers to optimise for latency, cost, quality, or data sovereignty. Configure providers in the dashboard or via the API.

How providers work

Each provider is registered with LYRE using your API key for that service. When you create an agent, you reference providers by name. LYRE routes audio and text through your configured providers at runtime.

curl -X POST https://api.lyre.au/v1/providers \
  -H "Authorization: Bearer $LYRE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "stt",
    "provider": "deepgram",
    "api_key": "dg_...",
    "models": ["nova-2"],
    "config": {
      "language": "en-AU",
      "smart_format": true
    }
  }'

STT providers (Speech-to-Text)

STT converts the caller's spoken audio into text for the LLM. Choose based on latency, accuracy, and data residency requirements.

Provider	Models	Streaming	Languages	Notes
Deepgram	nova-2, nova-2-general, nova-2-meeting	Yes	50+	Recommended default. Lowest latency, excellent en-AU.
Groq Whisper	whisper-large-v3, whisper-large-v3-turbo	No (batch)	99+	Fast batch transcription via Groq hardware. Good for post-call.
SCX Whisper	whisper-large-v3	No (batch)	99+	Australian-sovereign. Data stays in NEXTDC Sydney.
Google Cloud STT	chirp, chirp_2, long	Yes	125+	Sydney region available. Good multilingual support.
Azure Speech	whisper, custom	Yes	100+	Australia East region. Custom model training available.
Speechmatics	enhanced, standard	Yes	50+	UK-based. Excellent accuracy for accented English.
AssemblyAI	best, nano	Yes	Multiple	Real-time streaming. Built-in PII redaction.

LLM providers (Language Models)

The LLM generates conversational responses. For voice, latency to first token is critical — aim for sub-300ms TTFT.

Provider	Models	TTFT	Tool calling	Notes
Groq	llama-3.3-70b-versatile, llama-3.1-8b-instant, mixtral-8x7b-32768	~150ms	Yes	Recommended primary. Fastest inference via LPU.
SCX.ai	gpt-oss-120b, magpie-117b, deepseek-r1, llama-3.3-70b, qwen-2.5-72b	~400ms	Yes	Australian sovereign. NEXTDC Sydney. See sovereign AI guide.
Google Gemini	gemini-2.0-flash, gemini-2.0-pro, gemini-1.5-flash	~250ms	Yes	Sydney region via Vertex AI. Good fallback.
Mistral	mistral-large, mistral-medium, mistral-small, codestral	~300ms	Yes	EU-hosted. Strong multilingual.
Together AI	llama-3.3-70b, qwen-2.5-72b, mixtral-8x22b	~200ms	Yes	Good price/performance ratio.
Fireworks AI	llama-3.3-70b, mixtral-8x22b, qwen-2.5-72b	~180ms	Yes	Optimised inference. Competitive latency.
Cerebras	llama-3.3-70b, llama-3.1-8b	~100ms	Yes	Fastest raw inference. Limited model selection.

Important: OpenAI and Anthropic are not supported as LLM providers. LYRE uses open-weight and sovereign models exclusively.

TTS providers (Text-to-Speech)

TTS converts the LLM's text response into natural-sounding speech. Streaming TTS is essential for low-latency voice — the agent starts speaking before the full response is generated.

Provider	Models	Streaming	Voices	Notes
ElevenLabs	eleven_flash_v2_5, eleven_multilingual_v2, eleven_turbo_v2_5	Yes	1000+	Recommended. Best quality, flash model for speed.
Cartesia	sonic-2, sonic-english	Yes	50+	Ultra-low latency. Good for real-time voice.
Google Cloud TTS	neural2, studio, wavenet	Yes	400+	Sydney region available. Many Australian voices.
Azure Speech	neural, custom-neural	Yes	500+	Australia East. Custom voice training.
Deepgram	aura-asteria, aura-luna, aura-stella	Yes	12+	Simple, fast. Good for cost-sensitive deployments.

Dashboard configuration

The easiest way to configure providers is through the LYRE dashboard:

Navigate to Settings → Providers in the dashboard.

Click Add Provider and select the type (STT, LLM, or TTS).

Enter your API key for that provider. LYRE encrypts and stores it securely.

Select the models you want available. You can enable multiple models per provider.

API configuration

Method	Endpoint	Description
POST	/v1/providers	Register a new provider
GET	/v1/providers	List all configured providers
PATCH	/v1/providers/:id	Update provider config or rotate API key
DELETE	/v1/providers/:id	Remove a provider (fails if agents reference it)

Fallback chains

For production reliability, configure a fallback provider on each agent. If the primary provider returns an error or exceeds the latency threshold, LYRE automatically retries on the fallback.

Example: Groq primary, Gemini fallback

"llm": {
  "provider": "groq",
  "model": "llama-3.3-70b-versatile",
  "fallback": {
    "provider": "gemini",
    "model": "gemini-2.0-flash"
  }
}

Fallback is supported for LLM and TTS. STT fallback is on the roadmap.

Next: Calls →