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

# Quickstart

> From zero to a working agent in a few minutes

By the end of this guide you will have an agent with a personality, connected to a real tool, doing work for you in a conversation.

## 1. Create your agent

1. Open the [Leme dashboard](https://app.leme.ai) and go to **Agents**.
2. Select **New agent**.
3. Give it a name and a photo — this is how it will appear everywhere, including Slack and email.
4. Write its personality instructions: who it is, how your company communicates, what it should prioritize.

A short, concrete personality beats a long, generic one. For example:

> You are Rui, the operations assistant at Acme. Be direct and warm. Answer in Brazilian Portuguese. When you present numbers, always say where they came from.

The agent is ready to use as soon as it's created. See [Agents](/agents) for everything you can configure.

## 2. Connect a tool

Agents become useful when they can reach your tools.

1. Go to **Integrations** and pick a tool your team uses — for example your CRM, Stripe, Notion, or Google Drive.
2. Select **Connect** and authorize access. Most tools use the provider's own sign-in screen; some ask for an API key.
3. Choose whether the connection is **shared** (the whole workspace can use it, subject to permissions) or **individual** (tied to your account only).

When the connection is created, review its **Agent permissions**: what the agent may do freely, what needs human approval, and what is off. Safe defaults are pre-selected — reading is allowed, risky actions require approval or are disabled. See [Permissions & approvals](/permissions).

## 3. Enable the connection on the agent

Open your agent, go to its connections, and enable the one you just created. An agent only sees the tools you explicitly give it.

## 4. Talk to it

Open a conversation with the agent in the dashboard and ask for something real:

> Pull our 10 largest invoices this month and summarize them in a table by customer.

The agent works with the connected tool, shows its progress, and delivers the result. If it needs to do something you configured as requiring approval, the conversation pauses and asks first — nothing sensitive happens behind your back.

## Where to go next

<CardGroup cols={2}>
  <Card title="Bring it into a channel" icon="messages" href="/channels">
    Let your team talk to the agent in Slack, Teams, WhatsApp, or by email.
  </Card>

  <Card title="Automate it" icon="bolt" href="/triggers">
    Run the agent on a schedule or when an external event arrives.
  </Card>

  <Card title="Tune permissions" icon="shield-check" href="/permissions">
    Decide exactly what each connection allows, per risk category.
  </Card>

  <Card title="Teach it your way" icon="graduation-cap" href="/skills">
    Add skills so the work comes out in your company's format.
  </Card>
</CardGroup>