Overview

Multi-Channel AI Agents use the voicebot@v1 action to create phone agents that understand natural language, call external tools, and follow up across channels. Define a system prompt, connect your LLM (OpenAI, Claude, Mistral), and declare HTTP tools — the agent handles the rest.

Architecture

voicebot@v1 streams caller audio to your chosen LLM via real-time speech recognition. The model reasons over the conversation history and can invoke tools you define — each tool is an HTTP endpoint with typed parameters the AI fills automatically. When the caller hangs up, the hangup branch processes the full $bot.messages history for summarization, CRM push, and notifications.

Tool calling

Declare tools as HTTP endpoints with method, URL, headers, and JSON Schema parameters. The AI decides when to call them and fills the parameters from conversation context. Examples: query a calendar API for available slots, POST a booking, look up a customer in your CRM, check order status. No code required — just the endpoint spec in YAML.

Multi-channel follow-up

After the call, use openai@v2 to extract structured data from the conversation — summaries, appointment details, lead qualification. Send results via email@v1 to your team, sms@v1 to the caller (confirmation, links), and fetch@v2 to your backend (CRM update, webhook).

Configuration

Engine selection: OpenAI, or any OpenAI-compatible endpoint (custom URL supported). Temperature, max tokens, and model are configurable. pleaseWait plays filler phrases when the LLM takes more than N seconds. Built-in tools include dial (transfer to human) and hangup.