Agent Builder

Agents are the AI personas that handle conversations. Configure their personality, capabilities, voice, and which tools and knowledge bases they have access to.

Agent Configuration

System Prompt

The system prompt is the most important setting. It defines who the agent is and how it behaves. Write it clearly:

• Identity: "You are Maya, a billing support agent for Acme Corp."
• Task scope: "Help customers check balances and make payments."
• Boundaries: "Do not discuss competitors. If asked about a refund, transfer to the billing team."
• Voice style: "Keep every response under 2 sentences. Be warm and direct."

AI Model

Choose the language model that powers the agent's reasoning. Different models offer different trade-offs between speed, capability, and cost. The default is optimized for conversational voice — fast responses with high accuracy.

Voice Language

Select the language your callers will speak in. This affects both speech recognition accuracy and the voice output. Supported languages include English, Hindi, Tamil, Telugu, Malayalam, Kannada, Marathi, Spanish, French, German, and Arabic.

Voice

Pick a voice from the library. Voices are organized by language and gender. You can preview any voice from the Playground → TTS tab before assigning it. For brand consistency, you can also clone a custom voice.

Greeting

The first thing the agent says when a call connects. Keep it short: "Hello, this is Maya from Acme support. How can I help you today?" If you're using a Flow with a Prompt node before the Agent node, the greeting from the Flow node takes priority.

Built-in Tools

Enable tools the agent can use during a call. Each tool is available by toggling it on and optionally configuring it:

• CRM Lookup: Automatically looks up the caller's full CRM profile — contact details, open deals, recent interactions, owned assets, pending tasks, and AI insights. Called at the start of every interaction to personalize the conversation.
• Update CRM: Record information the caller shares during the call — lifecycle changes, custom field values, tags, or entity data (vehicle registration, policy number, property address).
• Create Deal: Create a new deal/opportunity in your pipeline when the caller expresses purchase intent or requests a quote.
• Book Appointment: Check calendar availability and book a meeting during the call. Integrates with Google Calendar and Calendly.
• Transfer Call: Transfer the caller to a phone number, queue, or human agent with full context handoff.
• Schedule Callback: Book a callback for the caller at a specific time.
• Send SMS / WhatsApp: Send a text message to the caller mid-call (confirmation links, documents, directions).
• Knowledge Base Search: Look up facts from your uploaded documents to answer questions accurately.
• Create Ticket: Log a support ticket in your connected helpdesk (Freshdesk, Zendesk, HubSpot).
• Check Availability: Query calendars or external APIs for appointment slots or inventory.
• End Call: Gracefully terminate the call with a reason.

CRM Context Injection

Before the agent speaks its first word, the platform automatically fetches the caller's CRM record and injects it into the AI's context. This means the agent already knows:

• The caller's name, lifecycle stage, and lead score
• Previous call summaries and topics discussed
• Open deals and their stages
• Pending tasks and follow-ups
• Custom field values and entity data
• AI-generated insights and notes

This happens in under 700ms and never delays the call. If the lookup takes too long, the call proceeds without it — reliability always wins.

Campaign Context

When an agent handles a campaign call, it receives additional campaign-specific context:

• Objective: What the campaign is trying to achieve.
• Qualifying criteria: What to evaluate during the conversation.
• Fields to collect: Specific data points to extract naturally.
• Conversation memory: What happened in previous attempts with this contact (objections raised, topics discussed, promises made).
• Objection handling: Pre-configured responses to common pushback.
• In-call actions: Whether calendar booking, SMS sending, or warm transfer are available.

Agent Actions (AI-driven routing)

Beyond built-in tools, you can configure Agent Actions — custom behaviors the AI decides to trigger autonomously based on the conversation:

• Escalate to Human: When the caller asks for a manager or is frustrated, route to the human handoff queue.
• Route to flow: Exit this agent and jump to a specific point in the flow (e.g., "billing", "sales").
• Call API: Hit an external endpoint mid-conversation and use the result in the response.
• Set Variable: Store a piece of information from the conversation for use later in the flow.

Each action has a "When to use" description you write in plain English. The AI reads this and decides when to trigger the action — no keyword matching needed.

Knowledge Base Assignment

Assign specific knowledge base documents to an agent. When the agent has the Knowledge Base Search tool enabled, it only searches documents assigned to it — not all documents in your workspace. This ensures data isolation between agents.

Custom Tools Assignment

Custom tools (HTTP APIs, webhooks, code snippets) you create in the Knowledge & Tools section can be assigned to specific agents. An agent only sees tools explicitly assigned to it.

Agent status

• Draft: Agent is being configured. Not yet live on any channel.
• Active: Agent is deployed and can handle calls.
• Archived: Agent is disabled and hidden from the active list.