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

# Custom Assistants

> Create AI assistants with custom instructions and knowledge bases

Assistants are custom AI personas that you configure to behave in specific ways. Each assistant can have its own personality, instructions, and knowledge sources. Think of them as specialized team members -- one might handle customer support, another might review code, and a third might help write marketing copy.

## Creating an Assistant

<Steps>
  <Step title="Open the Assistants page">
    Click **Assistants** in the sidebar.
  </Step>

  <Step title="Create a new assistant">
    Click **New Assistant** to open the creation wizard.
  </Step>

  <Step title="Pick a starting point">
    Choose one of the two quick-start presets, or browse the full **template library** -- ready-made system prompts for common use cases like sales, support, and coding. Picking a template pre-fills the system prompt so there's less to write from scratch.
  </Step>

  <Step title="Review the instructions preview">
    Expand **System instructions (preview)** to see a read-only preview of the exact system prompt your assistant will start with. You can edit it any time afterward from the assistant's Instructions tab.
  </Step>

  <Step title="Configure the assistant">
    Fill in the remaining details described below, then save.
  </Step>
</Steps>

<Tip>
  To start from an existing assistant, open the **⋯** menu on its card and choose **Duplicate**. This copies the build configuration into a new inactive assistant (knowledge and channels are not copied). Activating the copy re-checks your plan limits.
</Tip>

### Configuration Options

* **Name** -- A clear, descriptive name (e.g., "Customer Support Bot" or "Code Reviewer")
* **Description** -- A brief summary of what the assistant does, visible to workspace members
* **System Prompt** -- The instructions that define how the assistant behaves
* **Temperature** -- Controls creativity. Lower values produce more focused, deterministic responses. Higher values produce more creative, varied responses.
* **Max Tokens** -- The maximum length of responses the assistant can generate

### Writing Effective System Prompts

The system prompt is the most important part of your assistant's configuration.

<AccordionGroup>
  <Accordion title="Be specific about the role">
    Instead of "You are a helpful assistant," try "You are a senior customer support agent for a SaaS company. You help users troubleshoot billing issues, account access problems, and feature questions."
  </Accordion>

  <Accordion title="Define the tone and style">
    Specify whether the assistant should be formal or casual, concise or detailed. For example: "Respond in a friendly, professional tone. Keep answers under 200 words unless the user asks for more detail."
  </Accordion>

  <Accordion title="Set boundaries">
    Tell the assistant what it should and should not do. For example: "Never make up information. If you do not know the answer, say so and suggest the user contact support."
  </Accordion>

  <Accordion title="Provide examples">
    Include example interactions in the system prompt to show the assistant exactly how you want it to respond.
  </Accordion>
</AccordionGroup>

## Visibility Settings

The public widget works whenever it's enabled on the assistant's **Embed** tab -- visibility has no effect on it.

On plans that include the internal **Chat** workspace, visibility controls who on your team can see and use the assistant there:

| Visibility    | Who Can Access                                |
| ------------- | --------------------------------------------- |
| **Workspace** | Active workspace members                      |
| **Private**   | Creator, admins, and explicitly granted users |

## Connecting to Knowledge

Assistants become much more useful with access to your organization's knowledge:

* **Projects** -- Link one or more projects to give the assistant access to uploaded documents. The assistant searches these documents for relevant context when answering questions.
* **Knowledge Bases** -- Connect external knowledge bases for broader document retrieval.

See [Projects](/features/projects) and [Knowledge Base](/features/knowledge-base) for setup details.

## Tool Configuration

Assistants can use external tools through MCP (Model Context Protocol) integrations:

* **Notion** -- Search and retrieve pages from your Notion workspace
* **GitHub** -- Search repositories, read issues and pull requests
* **Atlassian** -- Access Jira issues and Confluence pages
* **HubSpot** -- Look up contacts, deals, and CRM data

See [Integrations](/features/integrations) for setup details.

## Testing Your Assistant

Try your assistant in the **Sandbox** — the single home for testing. Open it from the **Sandbox** item in the navigation sidebar, or from the **Test Assistant** button below the nav (both land on Sandbox).

<Steps>
  <Step title="Chat with your assistant">
    The **Chat** tab embeds the real chat widget, so what you see is exactly what visitors get — forms, theming, file upload, and handoff all behave identically. Use **Reset** to start a fresh conversation after changing settings, or **Open full screen** to view the preview in a new tab.
  </Step>

  <Step title="Batch-test its answers">
    The **Batch tests** tab runs a bank of questions through your assistant at once and grades each answer against your knowledge. See [Sandbox](/features/testing) for the full walkthrough.
  </Step>
</Steps>

<Tip>
  Sandbox uses the same AI pipeline as production conversations, including any connected knowledge bases and tools — so what you see is exactly what your users will experience. Runs are free and never count against your credits.
</Tip>

## Assistant Analytics

Usage and performance metrics live on the workspace-level **Analytics** page, not in the assistant's own settings. Use the assistant filter on that page to narrow the view down to a single assistant, including its widget funnel.

## Deleting an Assistant

Open the assistant's **General** tab and scroll to the **Danger Zone**. Deleting is permanent -- to confirm, type the assistant's exact name into the field, then click **Delete**. This also removes its knowledge documents and disconnects any channels.

<Warning>
  Deletion cannot be undone.
</Warning>

## Example Assistants

<CardGroup cols={2}>
  <Card title="Customer Support Bot" icon="headset">
    Connect your FAQ documents and help articles. Set the system prompt to answer questions accurately and maintain a friendly tone.
  </Card>

  <Card title="Code Reviewer" icon="code">
    Instruct the assistant to review code for bugs, security issues, and best practices. Use a lower temperature for consistent feedback.
  </Card>

  <Card title="Content Writer" icon="pen-nib">
    Configure a higher temperature for creative responses. Provide brand guidelines in the system prompt.
  </Card>

  <Card title="Sales Assistant" icon="chart-line">
    Upload product documentation and pricing guides. The assistant can answer prospect questions and draft follow-up emails.
  </Card>
</CardGroup>

<Note>
  The number of assistants you can create may depend on your workspace plan.
</Note>