> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wittify.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Build a Hybrid Agent

> Step-by-step walkthrough for creating an agent that handles both chat and voice on the same brand. The wizard has 7 steps in this order: Text Agent Settings, Voice Agent Settings, Appearance, Knowledge Base, Tickets & Leads, Integrations & Tools, Deployment.

<Note>
  Hybrid agents handle **both chat and voice** on every channel: web widget, WhatsApp, Messenger, Instagram, X, Telegram, plus phone lines via Twilio / Telnyx / SIP Trunk / WhatsApp Calls. To enter, click **Create Agent** on the dashboard topbar, then **Hybrid Agent** in the picker. The wizard opens at **Step 1: Text Agent Settings**.
</Note>

## Pre-flight

Hybrid is the most-capable but most-setup variant. Have ready:

* A **name** for your agent.
* A **chat system prompt** and a **voice system prompt** (you'll write both).
* A **logo or avatar**.
* **Knowledge sources** (files / FAQs / URLs).
* A **telephony provider account** with at least one provisioned number (for the voice side).
* Connected social accounts (WhatsApp Business, Facebook Pages, Instagram, X, Telegram bot) you want to deploy to.

The wizard runs at `/{lang}/create/hybrid/{step}`. The first step is `text-settings`.

## Step-by-step

<Steps>
  <Step title="Step 1: Text Agent Settings">
    Name your agent, pick the model, write the **chat system prompt**, tune Creativity / Max Tokens / Top-P / Reasoning Level / Stop Sequences.

    **Required to advance**: agent name (max 100 characters), Core Instructions (chat prompt, max 2,500 characters).

    [Full step reference →](/en/Systems/AI-Agents/Wizard/Text-Settings)
  </Step>

  <Step title="Step 2: Voice Agent Settings">
    Pick a voice, set the voice language, write the **voice system prompt** (separate from the chat prompt), tune Speaking Speed / Voice Temperature / 7 advanced fields, pick a Background Sound.

    **Required to advance**: Core Instructions (voice prompt, max 2,500 characters). The agent name from Step 1 is reused (no separate Agent Name field on this step for Hybrid).

    [Full step reference →](/en/Systems/AI-Agents/Wizard/Voice-Settings)
  </Step>

  <Step title="Step 3: Appearance">
    Brand both widgets at once: colors, logo, avatar, launcher icon, position, font, welcome banner, post-chat / post-call ratings, minimized state. Per-language entries cover both Chat Screen Text and Voice Screen Text.

    [Full step reference →](/en/Systems/AI-Agents/Wizard/Appearance)
  </Step>

  <Step title="Step 4: Knowledge Base">
    Upload files, add FAQs, crawl URLs, optionally enable public web search. Knowledge is **shared** between chat and voice, both run against the same trained sources.

    This step is **optional**.

    [Full step reference →](/en/Systems/AI-Agents/Wizard/Knowledge-Base)
  </Step>

  <Step title="Step 5: Tickets & Leads">
    **This is where Hybrid differs most.** Two tabs at the top: **Text Agent** and **Voice Agent**. Configure data collection independently per channel. Each has its own toggle, instructions, fields, categories, priority. Notifications are shared.

    The **Copy from Text Agent** button on the Voice tab deep-clones the Text config into Voice as a starting point, then they evolve independently.

    [Full step reference →](/en/Systems/AI-Agents/Wizard/Tickets-and-Leads)
  </Step>

  <Step title="Step 6: Integrations & Tools">
    Same as Text and Voice variants: marketplace, MCP, agent tools, custom API tools, webhooks. Connections apply to both chat and voice (e.g. Human Handoff fires for both).

    This step is **optional**.

    [Full step reference →](/en/Systems/AI-Agents/Wizard/Integrations-and-Tools)
  </Step>

  <Step title="Step 7: Deployment">
    All three tabs visible: **Website**, **Social** (WhatsApp / Messenger / Instagram / X / Telegram), **Telephony** (Twilio / Telnyx / SIP Trunk / WhatsApp Calls). The **Agent Status** card at the bottom has the **Activate** button.

    [Full step reference →](/en/Systems/AI-Agents/Wizard/Deployment)
  </Step>
</Steps>

## What makes Hybrid unique

* **Two prompts.** The chat prompt and the voice prompt are separate. Phrase the chat one with markdown formatting and longer answers (chat surfaces render markdown). Phrase the voice one for speech: no markdown, shorter sentences, natural pacing, written for the ear.
* **Independent Tickets & Leads channels.** Text and Voice tabs are fully independent. Field requirements, custom fields, categories, and priorities can differ per channel. The Copy button on Voice is a one-time deep-clone, after that they evolve independently.
* **All 10 deployment channels.** Hybrid is the only variant that can deploy to web + every social channel + every telephony provider in one agent.
* **Shared knowledge base.** A single set of trained files, FAQs, URLs answers both surfaces. No need to duplicate sources.
* **One agent identity.** Same name, same voice, same brand across chat and call. Customers experience continuity.

## Hybrid step order

The 7-step order is deliberate:

1. Text Agent Settings (the chat side first, since chat is more common)
2. Voice Agent Settings
3. Appearance (after both prompts so you can write voice screen text with full context)
4. Knowledge Base
5. Tickets & Leads
6. Integrations & Tools
7. Deployment (last, so all your config is saved before you go live)

## Test before you ship

The **Test Your Agent** button on Manage Agent pages opens a slide-in preview. The preview shows the chat view by default. Click the call button inside the preview to switch to the voice view. Both run against your real config. [Read more →](/en/Systems/AI-Agents/Manage-Agent/Test-Your-Agent)

## Activate

When all steps you care about are saved, go to **Step 7: Deployment** and click **Activate Agent** on the Agent Status card. One click flips the agent online for **both chat and voice** across every connected channel. Toast: *Agent activated.*

## Common questions

<AccordionGroup>
  <Accordion title="My chat prompt and voice prompt are basically the same. Do I really need both?">
    For best results, yes. Models speak differently than they write. A prompt with markdown bullets reads great in chat but stumbles when spoken aloud. Voice-style prompts (short sentences, natural transitions) sound human, but become overly terse in chat. Worth writing both even if 80% of the content overlaps.
  </Accordion>

  <Accordion title="I configured fields in Tickets & Leads on the Text tab. Why aren't they showing on Voice?">
    The two channels are independent. Click the **Copy from Text Agent** button on the Voice tab to copy your Text config over. After copying, you can refine the Voice version (e.g. drop email collection on calls since people rarely spell email correctly aloud).
  </Accordion>

  <Accordion title="The agent answered a chat in a different tone than my voice prompt asked for.">
    Different prompts are at play, the chat prompt drives chat answers, the voice prompt drives calls. They don't share. If your voice tone differs from chat by design, that's working as intended. If you want the same tone, copy the prompt from one to the other.
  </Accordion>

  <Accordion title="Activating Hybrid feels heavier than activating Text or Voice.">
    It is, slightly. Hybrid wires both surfaces at once. But the Activate button is the same single click. Channels that aren't connected just stay dormant until you connect them.
  </Accordion>

  <Accordion title="Can I deploy Hybrid to only chat for now and add voice later?">
    Yes. Connect chat channels in Step 7 and skip Telephony. The voice side stays inactive until you wire a telephony provider, all your voice prompt and voice settings are preserved for whenever you're ready.
  </Accordion>

  <Accordion title="My Knowledge Base is shared, but my chat agent answers from one document and my voice agent from another. Is that a bug?">
    No. Both query the same knowledge base, but each query is shaped by the prompt + the customer's question. If the chat prompt says *prefer concise written answers* and the voice prompt says *prefer step-by-step verbal instructions*, the model will pick different snippets to ground each answer.
  </Accordion>
</AccordionGroup>

## Next

<CardGroup cols={2}>
  <Card title="Open Step 1" icon="play" href="/en/Systems/AI-Agents/Wizard/Text-Settings">
    Start at Text Agent Settings.
  </Card>

  <Card title="Manage your agent" icon="chart-line" href="/en/Systems/AI-Agents/Manage-Agent/Manage-Agent-Overview">
    Where the live agent lives.
  </Card>
</CardGroup>