Retell AI
Type: low-code · Vendor: Retell AI · Language: Web product / API · License: proprietary · Status: active · Status in practice: emerging
Hosted platform for building, testing, deploying and monitoring AI phone agents, with single-prompt, multi-prompt and conversation-flow agent shapes, function calling and call/agent transfers.
Description. Retell AI is a phone-agent platform. Agents can be built as a Single-prompt agent (one open prompt), a Multi-prompt agent (state machine of prompts with transition conditions), or a Conversation Flow agent (explicit node graph with transition conditions). Function Calling, custom functions and code tools let the agent perform real-world actions; Transfer Call and Transfer Agent nodes hand off to humans or to other Retell agents. The dashboard exposes per-language selection with explicit accuracy trade-offs across the supported speech-recognition and voice providers.
Agent loop shape. Hosted phone-agent loop. The platform owns the STT, LLM orchestration, TTS and turn-taking layers; the developer authors agent behaviour as a Single-prompt, Multi-prompt (state machine of prompts with transition conditions) or Conversation Flow (node-and-transition graph) agent. Function calls and code tools execute external actions; Transfer Call hands off to humans and Transfer Agent hands off to another Retell agent.
Primary use cases
- lead qualification voice agents over telephony
- customer support and dispatch phone agents
- receptionist agents with call transfer
- multilingual phone agents with language selection
Key concepts
- Agent types (docs) — Single-prompt, Multi-prompt (state machine) and Conversation Flow (node graph) agents.
- Conversation Flow (docs) — Node-and-transition graph; transitions decide which node fires next.
- Function calling → tool-use (docs) — Custom functions, code tools and built-in functions for transfers and digits.
- Transfer Agent → conversation-handoff (docs) — Hand off the ongoing call to another Retell agent.
- Multilingual selection → multilingual-voice-agent (docs) — Pick exact language combinations; multilingual mode trades accuracy.
- Language constraints (docs) — Language must be supported by both the voice provider and a speech-recognition provider.
Patterns this low-code implements —
- ★★Conversation Handoff to Human
Transfer Agent is a documented action that modularises tasks and re-uses specialised agents without traditional phone transfers; Conversation Flow has agent/call transfer nodes.
- ★Multilingual Voice Agent Stack
Per-agent language picker with documented accuracy trade-offs; language must be supported by both the chosen voice and a speech-recognition provider.
- ★★Tool Use
Function calling covers built-in actions (transfer, press digit, send SMS) and custom functions / code tools that call external APIs.
- ★Topic-Based Routing
Multi-prompt agents and Conversation Flow agents are state machines that route the live conversation between prompts/nodes by transition conditions.
- ★★Human-in-the-Loop
Transfer Call is a documented action for handing the live call off to a human via PSTN with warm/cold transfer modes.
Neighbourhood
Click any neighbour to follow the lineage. Scroll to zoom, drag to pan.