> ## 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.

# Connections

> Give agents access to the tools your company uses

An **integration** is a tool Leme knows how to work with — Slack, Stripe, Gmail, Notion, your CRM. A **connection** is your authorized account with that tool: you sign in once, and agents can then work with it on your behalf, within the permissions you set.

One integration can have several connections. You might connect two Stripe accounts (production and a subsidiary), or have each teammate connect their own Gmail.

## Creating a connection

1. Go to **Integrations** in the dashboard and pick the tool.
2. Select **Connect**. Most tools open the provider's own sign-in and consent screen; some ask for an API key instead.
3. Choose the connection type — **shared** or **individual** (see below).
4. Review the connection's **Agent permissions** before agents start using it. Safe defaults are pre-selected. See [Permissions & approvals](/permissions).

Credentials are stored encrypted. They are never shown to agents, and never shown to other users — not even admins.

## Shared vs. individual

<CardGroup cols={2}>
  <Card title="Shared" icon="users">
    Belongs to the workspace. Multiple agents and users can work through it, subject to permissions. Best for company accounts: the team's Slack, the company's Stripe.
  </Card>

  <Card title="Individual" icon="user-lock">
    Belongs to one person. When an agent needs it, it uses the connection of whoever is talking to it — your agent reads *your* inbox for you, never someone else's. Best for personal accounts: email, calendar.
  </Card>
</CardGroup>

The difference matters most for individual connections: the agent acts as the person in the conversation. If you ask the agent to check email and you haven't connected your account, it will tell you — it cannot borrow a colleague's connection, and admins cannot grant access to someone else's individual connection either.

## Using a connection

Enable the connection on each agent that should use it — agents never gain access implicitly. For individual integrations, you enable the integration on the agent, and the concrete account is resolved from whoever is in the conversation.

What the agent may *do* through the connection — read, write, message people, and so on — is controlled per connection in [Permissions & approvals](/permissions).

## Managing connections

Shared connections and their permissions are managed by `Owner` and `Admin` roles. Individual connections are managed by their owner.

A connection can be disabled at any time: agents lose access immediately, and nothing is deleted. If a connection expires or is revoked on the provider's side, it shows as needing attention in **Integrations** — reconnecting restores it.