Creating Call Flows

Using the Call Flow Builder

Call Flows let you create sophisticated call handling with multiple branches and outcomes. Every Voice AI Agent requires a deployed Call Flow to handle conversations. The fastest way to get started is to use Create with AI first, then edit the generated flow to match your business.

1. Navigate to Call Flows

   Click Call Flows under the Phone section in the sidebar.

2. Start with Create with AI

   Click 'Create with AI' first. Describe your business, what the caller needs, and the main outcomes you want, such as lead capture, booking, routing, or voicemail.

3. Review the generated flow

   Let the AI generate the first draft, then inspect the full flow before publishing anything. This is your starting point, not the final version.

4. Edit the Start node and greeting

   Make sure the greeting sounds like your business and clearly tells the caller what happens next.

5. Refine Collect and routing logic

   Update the generated questions, response paths, and fallback handling so callers can reach the right outcome even if they say something unexpected.

6. Add or adjust Say nodes

   Use Say nodes for explanations, confirmations, and next-step instructions. Single outgoing edge = auto-advance. Multiple outgoing edges = caller input is required.

7. Set clear End or Transfer outcomes

   Make sure every branch ends correctly, whether that means ending the call, transferring to a person, sending to voicemail, or creating a CRM action.

8. Rename nodes clearly

   Give nodes descriptive names so the flow stays easy to maintain after the AI creates the initial draft.

9. Test in the simulator

   Click 'Test Call Flow' and walk through all major scenarios: ideal responses, invalid responses, silence, transfer paths, and dead ends.

10. Deploy to Production

    After successful testing, click 'Deploy' to move your Call Flow to Production status.

11. Assign it to a phone number

    Go to Phone Numbers, open your number, set the Production Call Flow, choose the AI engine and voice, then save.

Tips

• Create with AI is the best starting point for most users because it gives you a usable first draft faster
• Treat AI-generated call flows as a draft that you refine, not something you publish without review
• Start simple: greeting → collect input → route → end. Add complexity once the base flow works
• Always include fallback/error handling paths for unexpected responses and timeouts
• Use descriptive node names ('Lead Qualification', 'Sales Transfer', 'Booking Confirmation') so you remember what each does
• Tool Call nodes execute backend logic (look up customer, create CRM record, check inventory) silently without the caller knowing
• Use IVR Menu nodes for traditional phone menus — callers can press keys on their keypad OR say the digit out loud
• Multi-route nodes (Say with 2+ outputs) require caller input; single-route auto-advances smoothly
• Test all scenarios: valid input, invalid input, silence/no response, transfers, errors
• Call Flows are reusable—create one and assign it to multiple phone numbers

Important notes

• A Call Flow must be deployed to Production before it can handle real calls
• Do not publish an AI-generated flow without reviewing every branch and fallback path
• Changes to a deployed Call Flow won't affect live calls until you re-deploy
• Test thoroughly before deploying—callers experience every node you create
• Ensure all Collect nodes have proper error handling (what if caller doesn't respond?)

Understanding Node Types

Call Flows are built from different node types. Each serves a specific purpose in the conversation:

1. Start Node

   Entry point for every call. Defines your greeting and first message. Only one Start node allowed per flow.

2. Say Node

   Agent speaks a message to the caller. Single outgoing edge = auto-advances to next node. Multiple edges = caller must respond/press a key to route.

3. IVR Menu Node

   Creates a phone menu (press 1 for Sales, press 2 for Support, etc.). Callers can press keys on their phone keypad or say the number out loud — both work automatically. Toggle which digits to include (1-9, *, 0, #), give each a label, then connect each digit's output handle to the next node. Also connect the 'timeout' and 'invalid' handles for error paths.

4. Collect Node

   Gathers input: caller presses digit (1-9) or speaks answer (voice recognition). Always has multiple exits for different responses.

5. Tool Call Node

   Silent backend execution: look up customer data, create CRM record, check inventory, send email. Caller doesn't hear anything—ideal for data operations.

6. End Node

   Terminates the call. Can speak a closing message ('Thanks for calling!') or just hang up. Supports CRM updates and data capture at hangup.

7. Transfer Node

   Transfers to another destination: human agent, extension, or different call flow.

Global Prompt and Personality

The Global Prompt controls the overall personality, tone, and response style of your call flow. Think of it as the behavior layer that applies across the whole conversation, while individual nodes control the specific steps and routing.

1. Open the call flow editor

   Start from your call flow and look for the global prompt or personality controls in the builder.

2. Set the overall tone

   Define how the AI should sound, such as friendly, warm, professional, calm, direct, or concierge-style.

3. Use the AI global prompt builder if needed

   If you want help writing the prompt, use the built-in AI prompt builder to generate a personality prompt from your call flow structure.

4. Keep node copy aligned with the prompt

   Make sure the wording in your Say and Collect nodes matches the overall style you want the assistant to deliver.

5. Test the tone in the simulator

   Run a few realistic calls to hear whether the assistant sounds right across multiple branches, not just in the greeting.

Tips

• Use the global prompt for personality and delivery style, not for detailed branching logic
• A strong global prompt can make the same flow sound more polished without rebuilding the node structure
• If the AI sounds too robotic or too casual, update the global prompt before rewriting all of your nodes

Important notes

• Do not rely on the global prompt to replace node-level routing or call outcomes
• If your global prompt conflicts with node instructions, the caller experience can feel inconsistent

Global Nodes

Global Nodes are reusable topic handlers that can be triggered from anywhere in the conversation. They are best for common caller interruptions or cross-flow questions such as pricing, business hours, location, refund policy, or how to talk to a person.

1. Pick a reusable topic

   Choose a topic callers may ask about at any time, even if the current node is about something else.

2. Open the node inspector

   Select the node you want to reuse globally and enable the Global Node option.

3. Set the Global Label

   Give the node a clear topic label such as Pricing, Hours, Office Location, or Speak to Human.

4. Choose auto-return behavior

   If Auto Return is enabled, the assistant answers the global topic and then returns to the previous part of the conversation. If disabled, you can send the caller somewhere else after the global topic is handled.

5. Test interruptions in the simulator

   While testing a normal flow, interrupt it with a question like 'What are your hours?' or 'How much does this cost?' and confirm the assistant moves into the global node correctly.

Tips

• Use Global Nodes for topics that can interrupt almost any conversation path
• Hours, pricing, location, transfer-to-human, and FAQ topics are strong candidates for global nodes
• Auto Return is useful when the caller asks a side question but should continue the original path afterward

Important notes

• Do not overuse global nodes for topics that should stay inside the main flow
• If the global label is unclear, the assistant may not trigger the right topic reliably during conversation

Global Knowledge Base

Global Knowledge Base lets the call flow reference selected knowledge across the whole conversation. This is useful when business information should be available in multiple parts of the call without manually repeating it in every node.

1. Open the left panel in the builder

   Find the Global Knowledge Base section in the call flow builder.

2. Enable Global KB

   Turn on the global knowledge setting for the flow.

3. Choose search behavior

   Set the search scope and any auto-search topics that should trigger knowledge lookup across the conversation.

4. Keep source material current

   Make sure the linked business information is accurate before depending on it in live calls.

5. Test real questions

   Ask common business questions in the simulator to confirm the assistant finds and uses the right information.

Tips

• Use Global KB for business facts that may come up in multiple branches, such as pricing, hours, policies, or service coverage
• Global KB works well with a strong global prompt and a few well-designed global nodes
• If callers ask the same question across many branches, move that information into global knowledge instead of duplicating it everywhere

Important notes

• Outdated knowledge content leads to outdated answers
• Global knowledge should support the flow, not replace clear routing and node design

Advanced Best Practices

The best phone flows combine AI-generated structure, clear routing, strong personality, reusable global topics, and accurate business knowledge.

Tips

• Start with Create with AI, then refine the draft with your own business logic
• Use the global prompt for personality, global nodes for reusable interruptions, and global KB for shared facts
• Keep the main path simple even if the flow has advanced capabilities behind the scenes
• Test with real interruptions, side questions, silence, and caller confusion — not just ideal responses
• Re-test the full flow after changing the global prompt, global nodes, or shared knowledge behavior

Important notes

• Advanced features can improve the experience, but they also increase the number of paths you need to test before production