Speech in and out
Speech recognition and voice playback run in the caller's browser, so they add no cost to the platform.
Speech, understanding and answers
Voice AI
Callers speak, the assistant understands, answers from your knowledge, and acts through tools, with safe fallbacks at every step.
Pipeline
From speech to answer
Each stage can be swapped without changing the assistant.
Built-in understanding
Intent detection, slot filling and confirmation steps, so booking and lookup flows are predictable.
Answers with sources
Retrieval over your documents using the platform's own embeddings and vector index. Answers cite where they came from.
Optional language models
Connect an OpenAI-compatible, Groq, Gemini or OpenRouter model for open-ended questions. External models are off until you enable them.
Many languages
Switch the conversation language, with optional machine translation.
Safety rules
Policies for blocked topics, redaction, emergency phrases and maximum turns.
Fallbacks
If something is unsure, a person takes over
Callers can always ask for a human.
- Hand-off queue the caller's request opens a queue item with a short summary.
- Callback tasks if nobody answers in time, a callback task is created automatically.