Documentation
Getting StartedRetell configuration

Retell configuration

Voice, response engines, analysis and tools for Retell agents.

Retell configuration includes ambient sound, speech recognition, denoising, pronunciation, webhook events, privacy, retention and post-call analysis settings. Language accepts a language code or an array of codes. Existing voice model and language values remain accepted when editing imported agents.

Choose a response engine: Retell LLM, custom LLM or conversation flow. Custom LLM uses its WebSocket URL; conversation flow uses its flow ID. The general prompt and begin message belong to the Retell LLM. Its ID may be omitted when creating a new LLM through deployment.

Tool references are resolved into the LLM's general tools. End call, transfer, agent swap, DTMF, SMS, custom, code, variable extraction, bridge/cancel transfer and MCP tools use their current vendor schemas. Existing custom_function tools remain accepted and are sent as custom tools. Legacy transfer numbers remain accepted; new configurations use transfer_destination and transfer_option.

JSON parameter names, HTTP headers and response-variable names are preserved verbatim during deployment.

For Retell LLM agents, deploying an explicitly empty tool-reference list clears the general tools. Omitting both tool references and inline tools leaves general tools unspecified in the update.

Editing an existing agent preserves its configured Retell agent name independently of the PRM display name. Imported vendor field names are normalized for the form without renaming user-defined keys in maps or extension fields.