Voiceflow
Type: no-code · Vendor: Voiceflow · Language: N/A · License: proprietary · Status: active · Status in practice: mature
SaaS visual platform for designing AI customer-experience agents that mix agentic playbooks (LLM-driven goal-based reasoning with tools) and deterministic workflows (visual step graphs), deployed across chat, voice, and call-centre channels.
Description. Voiceflow is a hosted platform for building AI agents for support, lead generation, and other customer experiences. Builders mix two primitives: Playbooks (goal-based LLM agents with tools and instructions that decide how to reach an outcome) and Workflows (visual, deterministic step graphs with branches, conditions, and integrations). Agents have a global prompt, instructions, a knowledge base for RAG, and global tools available everywhere. The Crew step coordinates multiple playbooks with direct handoffs between them. Tools include API tools, MCP tools, Function tools (custom JS), system tools (knowledge base, call forwarding, web search, UI artefacts), and pre-built integrations. Deployment surfaces include chat and voice widgets and call-centre IVR.
Agent loop shape. Hybrid agentic/deterministic loop. The agent has a global prompt + instructions and may be in an agentic context (top-level agent, playbook, or crew step — LLM reasons freely with global tools) or a scripted context (workflow step — deterministic, global tools disabled). Playbooks are goal-based: the LLM decides which tool to call (API tool, MCP tool, function tool, integration, system tool) toward the configured outcome. The Crew step nests multiple playbooks that hand control between each other based on each playbook's LLM description. Workflows can embed playbooks via the Playbook step for agentic-inside-scripted hybrids.
Primary use cases
- AI customer support agents with playbook + workflow design
- lead generation and qualification conversational agents
- voice / IVR agents for call-centre automation
- knowledge-base RAG agents grounded in uploaded documents and Zendesk/Shopify connectors
- multi-playbook crews coordinating sales, billing, and support sub-agents
Key concepts
- Playbooks → tool-use (docs) — Goal-based agentic primitives — LLM decides how to reach the configured outcome with tools.
- Workflows → visual-workflow-graph (docs) — Visual, deterministic multi-step builder with branches, conditions, integrations.
- Crew step → handoff (docs) — Coordinates multiple playbooks that hand off tasks to each other within one workflow step.
- Knowledge base → agentic-rag (docs) — Upload URLs/sitemaps/documents or connect Zendesk/Shopify/Kustomer/Salesforce for RAG grounding.
- Tools (API / MCP / Function / Integration / System / Global) → tool-use (docs) — Six tool kinds the agent can call; Global tools are agent-level and available in any agentic context.
- Global prompt / instructions (docs) — Per-turn agent prompt (personality/goal/tone/guardrails) plus routing instructions for skill selection.
Patterns this no-code implements —
- ★★Agentic RAG
Knowledge base ingests URLs, sitemaps, documents, and external sources (Zendesk, Shopify, Kustomer, Salesforce) for RAG grounding; explicitly called out in the docs as 'ground responses in real infor…
- ★★Conversation Handoff to Human
Crew step lets playbooks hand off tasks between each other directly; system tools include 'forward calls' for voice channels.
- ★Multilingual Voice Agent Stack
Documented as a native capability across voice and chat channels.
- ★★Tool Use
Six documented tool kinds (API, MCP, Function, Integration, System, Global) callable from playbooks; tool description plus LLM signal drive selection.
- ★★Visual Workflow Graph
Workflows are explicitly a visual builder; playbooks and workflows are the two named primitives of the build surface.
- ★★Plan-and-Execute
Playbook = goal-based plan; workflow = explicit scripted execution; the two are mixed by embedding a Playbook step inside a Workflow.
- ★★Hierarchical Agents
Crew step nests a root playbook with child playbooks that hand control between each other and back to the root.
- ★Topic-Based Routing
Instructions define routing across skills; each playbook's LLM description tells the top-level agent or crew when to route into it.
References
Provenance
- Last analyzed:
- Last updated:
- Verification status: verified