Every dataset says what it means, so your agent reads it correctly.
Gatlas reads the tools each company runs and defines each figure once for the whole group. Each dataset is written for an AI agent with what it means and notes on its columns, and each company’s files become curated documents your agent can read.
Each dataset carries the write-up an agent reads first
A dataset is a table of company data, such as orders by week, with what it means and notes on its columns.
Your agent reads that write-up first, then asks only for the rows that answer the question.
Ready-made datasets start from the dataset catalog
The dataset catalog holds ready-made datasets for a connector, each with its write-up and the query behind it. Your agent publishes one as it is or adapts it to a company.
Connecting e-conomic, PostHog, QuickBooks, Shopify or Stripe adds a starter set of datasets for that company.
Customers
The customer master with lifetime booked invoice count, last invoice date and gross invoiced per invoice currency.
- Who are the customers, and when was each last invoiced?
- Which customers have an open balance?
balance is the customer's open balance as e-conomic reports it. Lifetime activity counts booked invoices only: lifetime_invoice_count, last_invoice_date, and lifetime_gross_by_currency, a JSON object of gross invoiced per invoice currency (for example {"DKK": 125000.00, "EUR": 4000.00}); amounts in different currencies are never added together.
- balance
- Open balance as e-conomic reports it.
- last_invoice_date
- Date of the latest booked invoice.
- lifetime_gross_by_currency
- Gross invoiced per invoice currency, as a JSON object of currency code to amount.
- data_as_of
- When this row was last synced.
Every read names its dataset and its date
A published dataset reaches your agents through the agentic interface, today MCP, the open standard agents use to connect to tools.
Each read carries the source line under the answer in Figure 3.1.
A dataset across several companies reaches only an agent granted every one.
Curated documents keep each company’s files readable
A curated document is a page, such as a company overview, kept current from the files a company syncs from Google Drive or SharePoint.
Written from the source
A prompt you write turns a company’s files, other documents and datasets into the document, on a daily or weekly schedule or whenever you run it.
Versioned
Each version lists the files it was written from, so your agent can open the file behind a sentence.
Published
A document reaches your agents once someone publishes it, scoped to the companies it covers.
Search
When no document covers a question, your agent searches the files underneath, and the search says how much of them it reached.
Ask about every company’s datasets from your own assistant
Create your workspace and connect your assistant. It publishes datasets over the data of each company you connect.