Conversation AI v3 introduces a modern, node based builder for chat and voice. Map user journeys with triggers, conditions, and actions, manage context, and hand off to agents when needed. Built in testing, versioning, and analytics help you iterate faster and ship reliable assistants.
Visual flow canvas with drag and drop nodes for intents, messages, conditions, and actions
Context management for variables, memory, and slot filling
Entity extraction and validation with prompts and retries
Multichannel support for website chat, SMS, WhatsApp, and more
Built-in test console with live previews and transcript capture
Versioning, change history, and safe publish controls
Analytics for turn-by-turn metrics, drop-off points, and goal completion
Human handoff nodes for seamless transfer to live agents
Reusable subflows and templates for common conversation patterns
Integrations to trigger workflows, update records, and call webhooks
Start with a single happy path, then add branches for common exceptions
Name nodes with verbs and outcomes like Collect Email or Confirm Appointment
Use entity validation with friendly reprompts to reduce dead ends
Store important values in variables early and reuse them downstream
Add a universal Fallback with guidance and quick replies to recover gracefully
Review transcripts weekly to spot confusing prompts and refine wording
Use sub-flows for shared tasks like authentication or FAQs to avoid duplication
Step 1
Open AI Agents, then click on Conversation AI and choose New flow. Select your primary channel to set defaults.

Step 2
Add triggers such as keyword, intent, or channel entry. Configure greeting and initial options.

Step 3
Create entities for items you must capture. Add validation rules and default values.

Step 4
Drag message, question, and condition nodes to design the happy path. Connect actions like webhook or record update.

What channels are supported?
Web chat, SMS, WhatsApp, and similar messaging channels with consistent flow behavior across each
How does entity validation work?
Define required entities with rules and reprompt messages; the flow retries or routes to fallback if validation fails
Can I reuse parts of a conversation?
Yes, create subflows or templates to centralize common tasks like authentication, FAQs, or data capture
How do I integrate with backend systems?
Use action nodes to call webhooks, update records, trigger workflows, or fetch data for dynamic replies
Is there version control?
Yes, save versions with change history, test safely, and publish when ready to ensure stable deployments
How do I handle low confidence inputs?
Add a confidence check and route to clarification, a richer prompt, or a human handoff based on thresholds