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

# How it works

> The end-to-end pipeline behind every Clara buyer conversation - from first hello to booked meeting.

Every Clara conversation follows the same pipeline. Each stage is configurable per agent, and every stage feeds the next.

## The conversation pipeline

<Steps>
  <Step title="Capture: Clara engages">
    A visitor opens your website and Clara starts the conversation - through the chat widget, a voice session, or a full avatar interaction. You choose the experience.
  </Step>

  <Step title="Understand: intent + context">
    Clara reads what the visitor says and does. Spoken input is transcribed in real time and any data you pass to the widget (like `userName` or page context) shape the conversation.
  </Step>

  <Step title="Reason: LLM + your training">
    The language model generates responses grounded in your knowledge base, qualification framework, personalization rules, and connected tools.
  </Step>

  <Step title="Qualify: Evalute your customers">
    Clara asks your qualifying questions - from BANT, MEDDIC, CHAMP presets or a custom set - and logs every answer.
  </Step>

  <Step title="Act: book and sync">
    Clara books meetings on your calendar, writes the lead and transcript to your CRM, and fires webhooks for every downstream workflow.
  </Step>
</Steps>

## What Clara can be?

The same agent can present itself in several modes. The pipeline is identical - only the render layer changes.

| Mode            | How it looks                                    | Best for                                    |
| --------------- | ----------------------------------------------- | ------------------------------------------- |
| **Chat**        | Text conversation in the widget                 | High-intent self-serve buyers, low friction |
| **Voice-only**  | Pulsing audio visualizer with Clara's voice     | Fast, low-bandwidth, phone-like flows       |
| **Avatar**      | A lifelike face that sees, hears, and speaks    | Premium, human-feeling product experiences  |
| **With slides** | Clara presents relevant slides while she speaks | Demos, walkthroughs, objection handling     |

See [Agent widget](/docs/capabilities/widget) for modes and [Avatars & voice](/docs/capabilities/avatar-gallery) for the visual options.

## Configuration layer

Clara's behavior comes from the agent configuration, edited in your Clara dashboard:

| Setting              | Controls                                                            |
| -------------------- | ------------------------------------------------------------------- |
| Company Description  | What your company offers, who it serves, and your value proposition |
| Goal                 | What Clara drives the conversation toward                           |
| Tone                 | How Clara speaks                                                    |
| Closing Behavior     | How Clara ends conversations and suggests next steps                |
| Knowledge base       | What Clara can answer from (docs, FAQs, website)                    |
| Visual presentations | The deck Clara can present during conversations                     |
| Qualifying questions | BANT / MEDDIC / CHAMP presets or a custom question set              |
| Meeting booking      | Where qualified buyers get scheduled                                |
| CRM connection       | Where leads, transcripts, and meetings sync                         |

Everything attaches to the same agent, and changes take effect on the next conversation - no redeploy.

## The sales loop

Every conversation produces structured output your team can act on:

* **Qualification record** - every question, every answer, logged in detail
* **Meeting record** - what was booked, with which rep
* **Conversation transcript** - the full context attached to the CRM record
* **Webhook events** - `conversation.completed`, `lead.qualified`, `meeting.booked`, and more

[See the full event list →](/docs/operations/webhooks-events)