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

# Creating an Agent

> Configure a new Bizzy agent with a system prompt, model, and LLM parameters

1. Go to **Agents** in the sidebar.
2. Click **New Agent**.
3. Fill in the form.

<Frame>
  <img src="https://mintcdn.com/thebizzycomapny/LTbGbhRRJUQlFsAe/images/user-guide/agents/create-form.png?fit=max&auto=format&n=LTbGbhRRJUQlFsAe&q=85&s=692c79b18c946c5245de83df4a7f3aec" alt="Create agent form" width="800" height="500" data-path="images/user-guide/agents/create-form.png" />
</Frame>

## Required fields

| Field      | Notes                                                                   |
| ---------- | ----------------------------------------------------------------------- |
| **Name**   | Short label — shown in chat and conversation history                    |
| **Type**   | `Assistant` (general-purpose) or `Specialist` (narrower, task-focused)  |
| **Status** | `Active` to enable chat; `Disabled` to park it                          |
| **Model**  | See [Models](/user-guide/agents/models). Defaults to Claude Sonnet 4.6. |

The type influences the default system prompt applied if you leave the prompt
empty.

## System prompt

Open the **System Prompt** section to customize behavior. Write in plain English
— describe the agent's role, tone, any rules, and what kind of output you want.

If left blank, a sensible default per type is used.

<Warning>
  The system prompt is the single biggest lever on agent behavior. Be specific
  about scope ("only answer support questions about our product X") and
  boundaries ("do not send email without explicit user confirmation").
</Warning>

## LLM settings

Advanced settings in the accordion — safe to leave at defaults for most cases:

| Setting               | Typical range       | What it does                                         |
| --------------------- | ------------------- | ---------------------------------------------------- |
| **Temperature**       | 0 – 2               | Randomness. Lower = more deterministic               |
| **Max tokens**        | Varies by model     | Cap on output length per turn                        |
| **Top P**             | 0 – 1               | Nucleus sampling cutoff                              |
| **Top K**             | Integer             | Candidate token cap                                  |
| **Frequency penalty** | -2 – 2              | Discourages repetition                               |
| **Presence penalty**  | -2 – 2              | Encourages new topics                                |
| **Max steps**         | Integer (default 5) | How many tool-call cycles the agent can run per turn |
| **Max retries**       | Integer             | Retries on provider errors                           |

## Saving

Click **Save**. The agent is created immediately. Tool permissions default to
the regular-user baseline — open the new agent's detail page and visit the
**Tools** tab to adjust. See
[Tool Permissions](/user-guide/agents/tool-permissions).

## Default agent

Each organization can have one **default agent**, marked with a badge in the
list. The default agent is what Bizzy suggests when you haven't picked another
one explicitly. To mark an agent as default, toggle the flag on its detail page.

## Next steps

<CardGroup cols={2}>
  <Card title="Models" icon="robot" href="/user-guide/agents/models">
    Available models and how to choose
  </Card>

  <Card title="Tool permissions" icon="shield-check" href="/user-guide/agents/tool-permissions">
    Decide what the agent can do
  </Card>
</CardGroup>