---
title: "Onboarding guide"
description: "From a new workspace to your agent’s first answer, step by step, and how to add more companies."
url: https://www.gatlas.ai/docs/onboarding
markdown_url: https://www.gatlas.ai/docs/onboarding.md
---

# Onboarding guide

Everything you need to go from a new account to your first answer in your agent. Each step names who does it.

## Who does what

Onboarding touches a few people, and each of them only does their own part.

- **You**, the person setting up Gatlas. You create the account, connect your agent, decide who sees what, and check the numbers. You are the owner of the workspace.
- **Whoever holds the logins**, usually the person in finance or operations who can sign in to your tools. If that is not you, they open a setup link you send, connect their tools, and are done. They never need a Gatlas account.
- **The people who ask questions**: partners, finance leads or your own team. They get a Gatlas login, an agent key, or both. An agent key is how an agent such as Claude connects to Gatlas.

## Before you sign up

Before you sign up, write down:

- The tools your business keeps its numbers in: accounting, commerce, payments, CRM, analytics. Check them against [the tools Gatlas connects](https://www.gatlas.ai/docs/sources.md); if one is missing, tell us.
- Who can sign in to each of those tools, with a name and an email address.
- The first questions you want answered. "Revenue by month" or "receivables older than 60 days" is a better start than "everything".
- Who should see the data once it is in.

If you run several companies, such as a fund with holdings or a studio with ventures, note them too and see Several companies.

> **Prompt for your agent:**
>
> Help me prepare before I set up Gatlas for my business. Fetch https://www.gatlas.ai/docs/playbooks/set-up-your-workspace.md and follow its "Gather what you need" section.

## Workspace and agent

1. **Sign up**: Go to [gatlas.ai/signup](https://www.gatlas.ai/signup) and fill in **Full name**, **Email**, **Password** (at least 8 characters) and **Workspace name**: the name of your business. Gatlas adds your business as the first company under that name, and you can rename and describe it later under Settings, **Company**.

**Create workspace** signs you in and opens **Welcome to Gatlas**, with two numbered steps: add Gatlas to your agent, then paste one prompt into a new chat. **Skip for now** goes to the dashboard instead; the steps after it are the same work done by hand there.

To sign up with your Google or Microsoft account, choose **Continue with Google** or **Continue with Microsoft** above the form. The next screen asks for the **Workspace name**, and you log in the same way from then on.

If a colleague invited you instead, open the invitation email and choose **Create account and join**. Invitations expire after 7 days.

2. **Connect your agent**: **Welcome to Gatlas** opens after sign-up. Pick your agent's tab (Claude, ChatGPT, Cursor or Other) and follow it; [Connect your agent](https://www.gatlas.ai/docs/connect-your-agent.md) has the full steps for each. Later, the same step is the first item of **Set up Gatlas** on **Overview**.

Then start a new chat and paste this:

> **Prompt for your agent:**
>
> Based on what you know about me and my business, show me what Gatlas can do for us and help me set up what is missing.

In Claude Code, run `/mcp__gatlas__gatlas_get_started` instead. Your agent starts from your business, recommends the tools to connect, and asks before it sends a setup link or changes anything.

> **Prompt for your agent:**
>
> Connect Gatlas to this agent for me. Fetch https://www.gatlas.ai/docs/playbooks/connect-your-agent.md and follow it.

## Tools and datasets

Connect each tool yourself, or send a setup link to the person who can sign in to it. Most people do both.

3. **Connect a tool yourself**: Open **Connect a tool**, under Sources in the sidebar, pick the tool, then either **Connect with** the tool (you sign in to it and approve access) or paste an **API key**. Each tool's requirements are on [Connecting tools](https://www.gatlas.ai/docs/sources.md).

After you approve access to a tool you sign in to, Gatlas shows **Choose tables** with every table selected, except the few a tool leaves for you to choose, such as Microsoft Teams message text. Choose **Save tables** to keep your selection; **Skip and sync every table** keeps the tables already selected.

> **Prompt for your agent:**
>
> Help me connect the tools my business uses to Gatlas. Fetch https://www.gatlas.ai/docs/playbooks/connect-tools.md and follow its "Connect a tool" section.

4. **Send a setup link**: Choose **Connect a tool**, then **Someone else connects it**. Choose **They choose** to let the contact pick their tools, or **You choose** to name them. The link expires in 7 days unless you pick 30, and it can be emailed straight to the contact.

The contact opens the link without an account and presses **Connect** for each tool. If they chose the tools, they finish with **That's everything**; if you named them, the link completes once each one is connected. It takes most people a few minutes. Follow progress in the **Status** column of the setup links table, for example **Pending**, **In progress** or **Completed**.

> **Prompt for your agent:**
>
> Help me send a setup link so the right contact can connect their own tools to Gatlas. Fetch https://www.gatlas.ai/docs/playbooks/connect-tools.md and follow its "Send a setup link" section.

> **Note:** The contact signs in to the tool themselves. You see the data, never a password or a key, and the contact can revoke access from the tool itself at any time. See the [privacy policy](https://www.gatlas.ai/legal/privacy) and [data processing agreement](https://www.gatlas.ai/legal/dpa) for where data lives.

5. **Watch the first sync**: The first sync starts by itself within a few minutes of connecting, and after that every 24 hours. Change the rhythm under **Schedule** on the connection's page, from every 5 minutes to daily or manual.

**Connections**, under Sources in the sidebar, shows each connection's status. **Connected** and **Syncing** are what you want. **Last sync failed** or **Reconnect needed** means a sync failed or access ran out; open the connection to see why, and see [Reconnect a connection](https://www.gatlas.ai/docs/sources.md#reconnect-a-connection). **Tool not answering** means the tool itself did not answer, and **Tool limiting reads** that it is slowing how fast Gatlas reads; Gatlas tries again by itself after either. A connection on a schedule that has not synced in 48 hours shows **No sync in 2 days**, on its row and on **Overview**.

> **Prompt for your agent:**
>
> Check whether the first sync has landed in Gatlas. Fetch https://www.gatlas.ai/docs/playbooks/connect-tools.md and follow its "Check the first sync" section.

6. **Add AI context**: Under Settings, **Company**, write in **AI context** what the business does, its fiscal year, its currency, and anything a newcomer would get wrong. Agents read this when they start working with your data, so a few good sentences here pay off on every answer. The same page holds the business's name, description, region and logo.

> **Prompt for your agent:**
>
> Help me write the AI context for my business in Gatlas, which I paste under Settings, Company. Fetch https://www.gatlas.ai/docs/playbooks/set-up-your-workspace.md and follow its "Write the AI context" section.

Synced data arrives as each tool stores it. A **dataset** turns it into a table that answers a question, such as monthly revenue, with a description an agent can read. Stripe and PostHog come with starter datasets. For everything else you build the datasets your questions need, and you do not need to write SQL to do it.

7. **Create a dataset**: Pick whichever suits you:

- Ask your agent. Once you have connected it (step 2), describe the dataset you want; Gatlas builds and approves it, usually in under a minute.
- In **Datasets**, under Definitions in the sidebar, choose **Create dataset** and use the **Visual builder** or the **SQL editor**.
- In **SQL**, under Library in the sidebar, **Run** a query and **Save as…** a dataset.

Set **Refresh schedule** to **Daily** or **Hourly** so the dataset picks up new data from each sync. A dataset left on **Manual refresh only** keeps the numbers it had when it was built. When you ask your agent to build a dataset, tell it to refresh the dataset daily. You can change the schedule later in the dataset's settings in **SQL**.

Name datasets after what they hold, such as "Northwind Monthly P&L". Agents find datasets by name.

More on this in [Datasets](https://www.gatlas.ai/docs/datasets.md).

> **Prompt for your agent:**
>
> Help me build a dataset in Gatlas that answers one question. Fetch https://www.gatlas.ai/docs/playbooks/datasets.md and follow its "Build a dataset" section.

8. **Describe it and approve it**: Give each dataset **AI context**: what a row is, what the numbers mean, and what to watch out for. Then turn on **Approved**. Agents are only offered approved datasets, so this is the switch that makes a dataset answerable.

Before you rely on it, compare one figure you already trust, such as last month's revenue, with what the dataset returns. Open the dataset under **Datasets**, choose **Open detail page**, then **Edit in SQL** and **Run**. A match is the quickest proof that both the connection and the dataset are right.

> **Prompt for your agent:**
>
> Help me describe a Gatlas dataset before I approve it. Fetch https://www.gatlas.ai/docs/playbooks/set-up-your-workspace.md and follow its "Describe a dataset before approving it" section.

## Give people access

9. **Invite your team**: Under Settings, **Members**, choose **Invite member**, enter an email address and pick a role. The roles:

- **Member**: explores all data, runs syncs, changes a dataset's settings and uses the agent keys an admin creates.
- **Admin**: full control of connections, datasets, briefs and members.
- **Lead admin**: an admin who can also add and remove admins in their companies.

You can invite anyone to a role below your own. Owner is not offered here: whoever creates the workspace is its owner, and an owner can pass that on to a member with **Make owner** in the member's menu. The invitation is emailed, and the link is also copied so you can send it yourself.

> **Prompt for your agent:**
>
> Help me plan who to invite to Gatlas and with which role. Fetch https://www.gatlas.ai/docs/playbooks/team-and-access.md and follow its "Invite your team" section.

10. **Connect people's agents**: On **Agent access**, under Agents in the sidebar, choose **New key**. Name it after the person and their agent, choose **Read-only** or **Read/write**, and under **Approved datasets** limit it to certain datasets if you like. **Create** gives you an **Agent install link**, valid for 30 days, which walks them through adding Gatlas to Claude, ChatGPT, Cursor or another agent in about a minute.

> **Important:** New keys start as **Read/write**, which lets an agent create datasets and change settings for the companies it can read. Choose **Read-only** for anyone who only asks questions.

> **Prompt for your agent:**
>
> Help me plan who gets an agent key to Gatlas and what each key reaches. Fetch https://www.gatlas.ai/docs/playbooks/team-and-access.md and follow its "Plan agent keys" section.

## Ask the first questions

Once your datasets are in place, ask your agent what you wrote down at the start. For example:

- "What was revenue for each of the last six months?"
- "Which invoices are more than 60 days overdue?"
- "Which datasets can you see, and what does each one answer?"

## Setup checklist

Admins see the same list as **Set up Gatlas** on **Overview**, numbered in this order and ticked off as the work happens:

- **Add Gatlas to your agent**
- **Paste this into a new chat**
- **Connect your tools**, with **Connect yourself** or **Send a setup link**, steps 3 and 4 above
- **Bring in your data**, the first sync
- **Add AI context**, step 6 above; with several companies it opens the first company's page
- **Create a dataset**
- **Approve a dataset**

To test it, have someone who was not part of the setup ask your agent a question about the business, and check the answer against your books.

## Your first month

A typical first month looks like this.

- **Day one**: sign up, connect your agent, and connect your tools or send a setup link to whoever holds the logins.
- **The first week**: the setup links come back and the first syncs run.
- **The second week**: build the datasets your first questions need, check one number against your books, then describe the datasets and approve them.
- **The rest of the month**: invite the people who ask questions, connect their agents, and turn on alerts in Automations. By then the checklist above is done.

## Sync alerts

On **Automations**, under Agents in the sidebar, choose **New automation**, then **Alert me when something happens**. An owner can send alerts to Slack or by email, and an admin by email. Gatlas then tells you when a sync fails, a connection stops syncing or access is about to expire, so a broken connection gets fixed before anyone asks a question it cannot answer.

> **Prompt for your agent:**
>
> Help me decide where Gatlas alerts should go and who acts on them. Fetch https://www.gatlas.ai/docs/playbooks/connect-tools.md and follow its "Keep syncs healthy" section.

## Several companies

A group that runs several companies, such as a fund with holdings or a studio with ventures, adds each one beside the first. Each company keeps its own connections and data, and everything above works the same for each of them.

Under Settings, **Company**, choose **Add another company**. With two or more, Settings shows **Companies** in place of **Company**, and each company is managed on its own page there; add the rest with **Add company**. Every company has a tier:

- **Group**: the top tier, which every other company sits under.
- **Sub-group**: holds a set of companies beneath it.
- **Company**: a company or business unit, where data actually syncs.

Your workspace stands for the whole group, so it needs no company of its own. If you want the holding company or fund in the tree, add it as a **Group**, put a **Sub-group** under it for each fund or studio, and nest each company under its sub-group with **Part of**. A flat list of companies works too. If your group calls these tiers something else, rename them under Settings, **Companies**, **Tier names**, which appears once you have a second company.

> **Prompt for your agent:**
>
> Help me plan the companies I add to Gatlas beside my business, which it already holds as my first company, in my workspace's tiers Group, Sub-group and Company. Fetch https://www.gatlas.ai/docs/playbooks/set-up-your-workspace.md and follow its "Plan your companies" section.

With several companies, a few steps above ask one more question:

- **Connect a tool** asks which company the tools belong to, whoever connects them. Each company's page under **Companies** opens it for that company, from **Connect a tool** or **Send a setup link**.
- Each company is described on its own page, on its **AI context** card, and the checklist step reads **Add AI context**.
- **Invite member** and **New key** ask which companies the person or the agent can see: **All companies**, or specific ones (see [Team and access](https://www.gatlas.ai/docs/team.md)).
- **Companies** rolls sync health up per company, showing the worst of its connections: for example **Healthy**, **Needs attention**, or **No sync in 2 days** when one of them has not synced in 48 hours. A company with a connection that has never synced shows **Not ready** (or **Syncing** while that first sync runs), unless another connection is worse.
- Your agent needs to know which company you mean: name it in your question, and it asks when you do not.
- Under **Automations**, each company has its own Slack channel; an admin can connect one for a company nested under another.

## Get help

Write to [team@gatlas.ai](mailto:team@gatlas.ai) or use **Feedback** in the app's sidebar. We usually reply within a business day. A missing tool, a number that does not match, or a question the docs do not answer are all worth an email.

[Next: best practices](https://www.gatlas.ai/docs/best-practices.md)
