# Create Agent (Reference: https://docs.iqra.bot/build/agent)

The **Agent** represents the "Persona" of your Voice AI. While the Script handles the logic flow, the Agent defines *who* is speaking, what they know about your brand, and how they behave tonally.

Creating an Agent [#creating-an-agent]

Navigate to **Agent Studio** -> **Agents** and click **Add Agent**.

<Steps>
  <Step>
    General Information [#general-information]

    Give your agent an internal name (e.g., "Iqra - Sales") and a description. This helps you identify them when assigning them to campaigns later.
  </Step>

  <Step>
    Context Injection [#context-injection]

    Instead of manually typing your company address or pricing into every prompt, you can "Inject" structured data from your **[Business Context](/build/knowledge/context)**.

    * **Branding:** Company name, website, support email.
    * **Services/Products:** The list of items you sell (Name, Short Description).
    * **Branches:** Locations and working hours.

    <Callout type="info" title="Token Optimization">
      Only inject what is necessary. If an agent is for "Technical Support", do not inject the entire "Sales Product Catalog". This saves tokens and reduces hallucinations.
    </Callout>
  </Step>

  <Step>
    Personality (The System Prompt) [#personality-the-system-prompt]

    This is the most critical section. You are constructing the **System Prompt** that governs the LLM's behavior.

    * **Role:** Who is the agent? (e.g., "You are an official AI agent for Badal Technologies.")
    * **Capabilities:** What can it do? (e.g., "You have access to the knowledge base system.")
    * **Ethics:** The boundaries. (e.g., "You are a Muslim agent with the ethics of the Prophet (SAW)." or "You never ask for passwords.")
    * **Tone:** The vibe. (e.g., "Treat the user with care and respect. Start responses with a polite greeting like 'Salam'.")

    <Callout type="warn" title="Multi-Language Requirement">
      You must define these fields for **every language** enabled in your business. This ensures your Arabic agent has the correct cultural nuance (e.g., using "Salam") compared to the English agent.
    </Callout>
  </Step>

  <Step>
    Utterances (Greetings) [#utterances-greetings]

    Define how the call starts.

    * **Agent Speaks First:**
      * **Enabled:** Standard for Outbound. The agent speaks immediately.
      * **Disabled:** Standard for Inbound. The agent waits for the user to say "Hello?".
    * **Greeting Message:** The exact text spoken. Supports [Templating](/build/script/templating) (e.g., `Salam {{ name }}`).
  </Step>
</Steps>

Pro Tips: Prompt Engineering [#pro-tips-prompt-engineering]

How to write instructions that actually work.

<Accordions>
  <Accordion title="Write Instructions, Not Adjectives">
    **Bad:** `Tone: Polite`

    **Good:** `Tone: Treat the user with care and respect. Start responses with a polite greeting when appropriate (e.g., "Welcome to Badal Technologies", "Salam").`

    *Why:* LLMs follow instructions better than abstract keywords. Tell it *how* to be polite.
  </Accordion>

  <Accordion title="Define the Negative Space">
    **Bad:** `Be helpful.`

    **Good:** `Ethics: Do not engage in vague, idle, or spam conversations. If someone forces you to, end the call.`

    *Why:* Defining what the agent *should not* do is as important as what it *should* do to prevent jailbreaks.
  </Accordion>

  <Accordion title="Give it an Identity">
    **Bad:** `You are a bot.`

    **Good:** `Role: You are Iqra, an official AI agent for Badal Technologies Oman. Your role is to help the user while putting their needs first.`

    *Why:* Giving the agent a name and a specific employer helps ground its responses in a specific context.
  </Accordion>
</Accordions>

Environment (Background Audio) [#environment-background-audio]

To make calls feel more natural and less "robotic," you can inject a continuous background audio loop.

* **Use Cases:**
  * **Call Center:** Subtle typing and murmuring sounds.
  * **Office:** HVAC hum and distant phone rings.
  * **Cafe:** Clinking cups (for casual agents).
* **Configuration:** Upload an MP3/WAV file. The platform handles the loop mixing automatically.

<Callout type="tip" title="Psychological Effect">
  Adding a low-volume background track significantly reduces the "Uncanny Valley" effect and makes users more forgiving of small latencies.
</Callout>

Next Steps [#next-steps]

Once the Persona is defined, configure the "Senses" and "Brain".

<Cards>
  <Card icon={<Mic />} title="Interruption Engine" href="/build/agent/interruption">
    Configure **VAD** and **Turn-Taking** logic to handle interruptions naturally.
  </Card>

  <Card icon={<Settings2 />} title="Intelligence & Integrations" href="/build/agent/intelligence">
    Connect specific LLM models, TTS voices, and Fallback providers.
  </Card>

  <Card icon={<Library />} title="Knowledge Base" href="/build/agent/intelligence#knowledge-base">
    Connect RAG documents and define search triggers.
  </Card>
</Cards>