Retellai Create Conversation Flow

MCP (Model Context Protocol) server configurations. Enables integration with external MCP-compatible services. Each MCP may include 'name', 'url', 'headers', 'query_params', and 'timeout_ms'.

Array of conversation flow nodes. Each node must have an 'id' (string), 'type' (string), and type-specific configuration like 'instruction'. Example: [{'id': 'greeting_node', 'type': 'conversation', 'instruction': {'type': 'prompt', 'text': 'Greet the user'}}].

Custom and built-in function definitions available in the conversation flow. Includes custom tools, end call tools, appointment booking tools, etc.

Knowledge base configuration for retrieval behavior. May include 'top_k' and 'filter_score' parameters.

Reusable sub-flows embedded within the conversation flow. Each component has its own nodes and tools.

Model configuration with model name and type. Example: {'model': 'gpt-4o', 'type': 'cascading'}. Required fields: 'model' (string) and 'type' (string).

System instructions applied across all nodes in the conversation flow. Used to define overall agent behavior and personality.

ID of the node where execution begins. If not specified, the first node in the nodes array is used.

Who initiates the conversation. Either 'user' or 'agent'.

Whether this conversation flow is used for transfer LLM scenarios. Affects how the flow is processed during call transfers.

Controls the randomness of the model's responses. Valid range: 0-1. Lower values make responses more deterministic. Default: 0.7.

Array of knowledge base IDs for RAG (Retrieval-Augmented Generation). Enables the agent to retrieve information from specified knowledge bases.