> ## 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.

# Agents

> Create agents with a name, a personality, and exactly the capabilities they need

An agent is an AI coworker configured for your company: it has a name, a face, a way of communicating, and a specific set of tools it's allowed to use. A project can have as many agents as it needs — a support agent, a finance agent, a sales agent — each with its own configuration.

## Personality

Every agent has personality instructions: a prompt that shapes who it is and how it communicates. This is where your company's voice lives.

Good personality instructions are concrete:

* **Identity** — its name, role, and what it's responsible for.
* **Tone** — formal or casual, which language to answer in, how your company talks.
* **Priorities** — what it should always do ("cite the source of every number") and what it should avoid ("never promise deadlines on behalf of the team").

You can change the personality at any time; new conversations pick it up immediately.

<Note>
  Personality shapes *how* the agent communicates. *What it can do* is governed by its connections and [permissions](/permissions) — instructions never override those.
</Note>

## Capabilities

An agent starts with general work skills — it can analyze data, write documents, generate spreadsheets and PDFs. Everything beyond that is explicit:

* **Connections** give it access to external tools (your CRM, Stripe, Notion…). Enable each connection on the agent; what it may do within each tool is controlled per connection in [Permissions & approvals](/permissions). See [Connections](/connections).
* **Skills** teach it your company's way of doing things — templates, procedures, formats. See [Skills](/skills).

An agent only sees the connections and skills you enable on it. Two agents in the same project can have completely different capabilities.

## Conversations

Each person talks to the agent in separate conversations, and one person can have many — just like a chat app:

* In the **web app**, you have a conversation list and can open a new one anytime.
* In **Slack and Teams**, each thread is its own conversation.
* By **email**, each email thread is a conversation.

The agent keeps context within a conversation. Start a new one when you switch topics — it keeps things focused and easy to find later.

## Multiple agents

Prefer several focused agents over one that does everything. A focused agent gets a tighter personality, fewer tools, and permissions that match its job — which makes it both better at the work and safer. Exclusive channels also attach per agent: your support agent can own Slack while your finance agent owns email. See [Channels](/channels).

## Archiving

Archiving an agent takes it out of service immediately: it leaves its channels, stops responding, and stops running triggers. Nothing is deleted — conversations and history are preserved, and you can reactivate it at any time.

Use it as a pause button: while you rework an agent's configuration, or when something looks wrong and you want it offline *now*.

Managing agents — creating, editing, channels, archiving — requires the `Owner` or `Admin` role. See [Workspace](/workspace).