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

# Get started

> Create your first agent, connect a conversation, and inspect its first reply.

This guide assumes Mesh is already running and you can open its dashboard.
For a new deployment, complete [Installation](/operations/installation) first.

## 1. Sign in

On a fresh installation using local authentication, create the first
administrator account. On an existing installation, sign in with your account
or accept the invitation your owner sent you. Installations using external
authentication use their configured sign-in provider.

The dashboard opens on the agent roster. Owners manage the installation;
agent operators can access only the agents they have been granted.

## 2. Add an agent

Choose **Add an agent** and work through the setup flow:

1. **The agent:** give it a name and handle.
2. **Model:** choose a provider, supply its credential, and choose the agent's
   models. The UI shows configured instance defaults where available.
3. **Memory:** review the suggested embedding configuration. PostgreSQL keyword
   retrieval is available without an embedding provider.
4. **Profile:** describe the agent's role and add a picture if wanted.
5. **Connector:** choose the conversation platform and complete its setup.

Model credentials belong in the setup flow or the agent's settings. They are
stored encrypted, separately from the agent's conversational context. See
[Agent provisioning](/agent-provisioning) for resuming setup and managing an
existing agent.

## 3. Connect a conversation

Choose the guide for your platform:

* **Slack:** [Automatic Slack setup](/slack-automatic-setup) can create and install
  an app after an owner configures workspace provisioning. The manual
  [Slack setup](/connectors/slack) path is also available. Invite the bot to the
  channel where you want to use it.
* **Telegram:** create a bot, enter its token, and register its webhook through
  the [Telegram setup](/connectors/telegram) flow.
* **Phone messaging:** configure a supported provider and number using
  [Phone messaging](/connectors/phone).

The platform must be able to reach your Mesh installation at its public HTTPS
URL. A dashboard that loads locally does not by itself prove inbound messaging
is configured.

## 4. Try a first message

Send a direct message such as “Hello, what can you help with?” or explicitly
mention the agent in a shared channel. Ask a small question before adding more
capabilities.

Mesh does not answer every message it can see. Direct messages, mentions,
conversation history, and engagement policy help determine whether it should
respond. [Addressing and engagement](/runtime/addressing) explains those decisions.

Open the conversation in the dashboard and inspect the message's trace. You
should be able to see the engagement decision and, if it ran, the model activity
and reply. If nothing arrives, check connector setup; if a message is recorded
but no reply appears, inspect the engagement or run outcome. See
[Inspect agent activity](/runtime/observability).

## 5. Add what the agent needs

Start with the capabilities needed for its job:

* Add an integration through [capability setup](/runtime/capability-setup).
* Configure [automatic model selection](/runtime/adaptive-model-routing) if you
  want routing within a chosen model family.
* Review [Dreaming](/runtime/dreaming) and [commitment](/runtime/commitments) settings when
  you are ready to review background learning or durable work.
* Ask for a [schedule](/runtime/schedules) when work should recur at a defined time.

Each guide explains its controls and limits. A model provider alone does not
give an agent web access, a shell, or access to external accounts.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.