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

# Custom Agents

> Create focused assistants with their own instructions, tools, and model settings.

Custom agents let you save the setup work behind a repeatable role. Instead of retyping instructions and selecting tools every time, configure an agent once and invoke it from chat with `@`.

<Frame>
  <img src="https://github.com/user-attachments/assets/d0a325c0-ff1e-4038-b6bf-fcf57659a5c1" alt="Navigator custom agent example" />
</Frame>

## Agent ingredients

<CardGroup cols={2}>
  <Card title="Instructions" icon="message-square">
    Define the agent's role, constraints, tone, and working style.
  </Card>

  <Card title="Tools" icon="wrench">
    Limit the agent to the MCP servers and built-in tools it needs.
  </Card>

  <Card title="Model" icon="brain">
    Pick the provider and model that fit the agent's cost, latency, and reasoning needs.
  </Card>

  <Card title="Sharing" icon="users">
    Make useful agents available to a team or keep them private for personal workflows.
  </Card>
</CardGroup>

## Example agents

<AccordionGroup>
  <Accordion title="GitHub manager" icon="code-branch">
    Give the agent repository context and issue/PR tools. Use it to triage issues, draft PR summaries, or prepare release notes.
  </Accordion>

  <Accordion title="Research analyst" icon="search">
    Combine web search, source extraction, and table generation for market scans or competitor summaries.
  </Accordion>

  <Accordion title="Operations assistant" icon="briefcase">
    Connect calendar, CRM, spreadsheet, and messaging tools through MCP or Composio-backed integrations.
  </Accordion>

  <Accordion title="Data helper" icon="chart-line">
    Give the agent code execution and visualization tools for repeatable analysis tasks.
  </Accordion>
</AccordionGroup>

## Good agent design

Keep each agent narrow. A smaller set of instructions and tools makes it easier for the model to choose the right action and easier for users to predict what the agent will do.