Skip to main content
TurnCall lets developers deploy AI voice agents over phone calls, WebRTC browser calls, WhatsApp, and SMS — with configurable providers, tool calling, call transfers, voicemail detection, and bidirectional server events. Define an agent via API, bind it to a phone number, and callers talk to your AI.

​
Features

Twilio PSTN Calls

Inbound and outbound phone calls via Twilio Voice + Media Streams.

WebRTC Browser Calls

Talk to agents directly from the browser with peer-to-peer audio.

Video Avatar

Optional lip-synced video avatar (HeyGen or Tavus) on WebRTC cascade calls.

WhatsApp Business

Voice calls and text messages via WhatsApp Cloud API.

SMS / Chat

Text-based conversations with agents via Twilio SMS or the Chat API.

Multi-Provider

Mix and match STT, LLM, and TTS providers per agent — Deepgram, OpenAI, Anthropic, ElevenLabs, Cartesia, Ollama.

Speech-to-Speech

~300ms processing (vs ~800ms cascade) via OpenAI Realtime, Gemini Live or Amazon Nova Sonic.

Tool Calling

Built-in tools (transfer, handoff, DTMF) + webhook tools + MCP servers.

Connected Apps

Slack and Google Calendar on an agent without handling a token — during a call, or when one ends.

Knowledge Base (RAG)

Upload documents with three retrieval modes: prompt, auto, and tool.

Agent Versioning

Publish immutable versions, auto-promote phone numbers, rollback instantly.

A/B Testing

Weighted traffic routing on phone numbers, deterministic by caller.

Post-Call Analysis

Automatic summary, sentiment, success evaluation, and structured extraction.

Takeaways

Define a JSON schema once, attach it to any agent, and get it filled from every call.

Evals

Behavioural tests that run against the real pipeline — scripted or simulated, text or audio, with tools mocked. Gate a PR on whether your agent still behaves.

Server Events

Bidirectional hooks — call-init, function-call, call-end, transcript updates.

Observability

OpenTelemetry traces and pipeline observers — per-stage latency, turn timing, startup.

​
How It Works

A caller connects over any transport; TurnCall runs a per-call Pipecat pipeline in one of two modes and streams audio back. See Architecture for the full pipeline, call flows, and data model.

​
Tech Stack

​
Build agents by describing them

Prefer not to hand-write agent configs? The Agent Builder turns a plain-English description into a deployed agent — it asks follow-up questions, generates the config, and creates it in your instance. It’s a separate, source-available app that runs alongside this engine.