Flow Builder

Provider Nodes

Configure LLM, Text-to-Speech, and Speech-to-Text providers for your flow.

Provider nodes supply AI services to your flow. Every flow requires one LLM provider, one TTS provider, and one STT provider.

LLM Providers

The LLM powers conversational intelligence — understanding caller intent, generating responses, and evaluating transitions.

OpenAI

Google Gemini

Anthropic

Grok (xAI)

Azure OpenAI

For organizations using Azure-hosted OpenAI models. Requires your Azure endpoint URL.

SettingDescription
EndpointYour Azure OpenAI endpoint URL (required)
ModelDeployed model name
Temperature0.0-2.0

LLM configuration

SettingDescription
ModelSelect the specific model
TemperatureResponse randomness. Range: 0–2. Lower = more deterministic. Not supported by every model — see below.
Max tokensMaximum tokens per response
Priority tier2x pricing for faster inference and higher availability

Temperature support

Some models reject a custom temperature and use reasoning defaults instead. When a restricted model is selected, temperature is stripped from the flow definition before it reaches the bot and the UI — setting a value has no effect.

The restriction is driven by model metadata, not a hardcoded list. Currently restricted: gpt-5, gpt-5-mini, gpt-5.1-chat-latest, gpt-5.2-chat-latest, and claude-opus-4-8.

The model's llmCapabilities.temperature field in the provider list response indicates support. The model table above renders this capability inline when showCapabilities is set.

TTS Providers

The TTS provider converts agent text responses into spoken audio.

ElevenLabs

ElevenLabs settings

SettingRangeDescription
VoiceSelect from premade, cloned, generated, or professional voices
Stability0.0–1.0Voice consistency. Higher = more stable, lower = more expressive.
Similarity0.0–1.0How closely output matches the original voice sample
Style0.0–1.0Style variation.
Speaker booston/offEnhances clarity and presence.
Speed0.7–1.2Playback speed.

Cartesia

Cartesia settings

SettingRangeDescription
VoiceSelect from voice library. Supports masculine, feminine, and gender-neutral.
Emotion54 optionssonic-3 only. See emotion list below.
Speed0.6–1.5Playback speed. Default: 1.0.
Volume0.5–2.0Output volume. Default: 1.0.

Supported emotions (sonic-3)

Positive — Neutral, Happy, Excited, Enthusiastic, Elated, Euphoric, Triumphant, Amazed, Surprised, Flirtatious, Curious, Content, Peaceful, Serene, Calm, Grateful, Affectionate, Trust, Sympathetic, Anticipation, Mysterious, Proud, Confident

Negative — Angry, Mad, Outraged, Frustrated, Agitated, Threatened, Disgusted, Contempt, Envious, Sarcastic, Ironic, Sad, Dejected, Melancholic, Disappointed, Hurt, Guilty, Bored, Tired, Rejected, Nostalgic, Wistful, Apologetic, Hesitant, Insecure, Confused, Resigned, Anxious, Panicked, Alarmed, Scared

Other — Distant, Skeptical, Contemplative, Determined

Google TTS

Gemini TTS models (gemini-2.5-flash-tts for low latency, gemini-2.5-pro-tts for highest quality). 30 named voices, 23 languages.

Google TTS settings

SettingDescription
VoiceNamed Gemini voice (e.g., Kore)
LanguageOutput language
Style promptNatural-language instruction that steers delivery style

xAI TTS

Streaming synthesis with word-level timestamps. 5 voices, 20 languages.

xAI TTS settings

SettingRangeDescription
VoicexAI voice. Default: eve.
LanguageOutput language. Default: en.
Speed0.7–1.5Playback speed.

Rime

mistv2 for low-latency streaming (300+ voices, English/Spanish), arcana for premium expressiveness (English/Spanish/French/German).

Rime settings

SettingDescription
VoiceRime voice. Default: cove.
LanguageOutput language. Default: en.

STT Providers

The STT provider transcribes caller speech into text for the LLM.

Deepgram

Deepgram settings

SettingDescription
LanguageTranscription language
KeywordsBoost recognition of specific words or phrases

Cartesia

Cartesia STT settings

SettingDescription
LanguageTranscription language
ModelSTT model to use

Soniox

Soniox settings

SettingDescription
LanguageTranscription language
ModelSTT model to use

ElevenLabs STT

ElevenLabs STT settings

SettingDescription
LanguageTranscription language
ModelScribe v2 model

Google STT

Streaming recognition with chirp_3 (best accuracy, 100+ languages), chirp_2, or latest_long (long-form conversation model).

Google STT settings

SettingDescription
Language hintsOne or more expected languages to bias recognition
LocationGoogle Cloud recognition location. Default: global.

xAI STT

grok-stt — real-time streaming with word-level timestamps and telephony support across 25 languages.

xAI STT settings

SettingDescription
LanguageTranscription language, or auto-detect
DiarizationLabel speech per speaker

Amazon Transcribe

Streaming recognition with word-level timestamps and telephony (8 kHz) support across 30+ languages.

Amazon Transcribe settings

SettingDescription
LanguageTranscription language
RegionAWS region for the streaming endpoint

Telephony

Transport layer for connecting calls.

On this page