Skip to content

Your AI doesn’t know your company.Gatlas gives it the context.


Gatlas keeps your companies’ data and what it means in one place, called a context layer. Your people read it, and so do your AI agents, such as Claude or ChatGPT. It works for a group of companies or for one.

01ProblemFigure 1.1

Where company knowledge sits

A company’s tools hold its records. How to read those records is in its people’s heads.

Across a group, every company has its own tools and its own way of counting. Explaining the same things again takes time every month.

An AI agent is only as useful as what it knows about your company.

  • Records, in tools

    Ledgers, CRMs, inboxes and drives hold what happened: every invoice, deal and message.

  • Know-how, in people

    Most of it was never written down.

  • An AI agent, with the tools only

    Without what your people know, it guesses.

Tools

  • e-conomic
  • HubSpot
  • Shopify
  • Google Drive
  • Stripe
Your AI agent

People

  • Which number is current
  • How margin is defined
  • Which customers are exceptions

Reads every record

Cannot read what people know

A column of tools on the left, an AI agent in the middle and a column of people on the right. Solid lines run from every tool to the agent. Dashed lines from the agent toward the people stop short.

Figure 1.1What an AI agent can and cannot readIllustrative company
02GapFigure 2.1

Three layers of a company

Of a company’s three layers, two run on software and the third is carried by people.

AI agents can only read software, so what a company knows has to be written down in software too.

No existing tool can hold it, because each tool sees only its own part of the company.

  • CustomersWebsite, shop and support
  • OperationsAccounting, sales and projects
  • ContextWhat the company knows: its data and what the data means
  • Runs on software
  • Carried by people
  1. CustomersWebsite, shop and supportRuns on software
  2. OperationsAccounting, sales and projectsRuns on software
  3. ContextWhat the company knows: its data and what the data meansCarried by people

Three stacked layers. Customers and Operations are drawn solid. Context, underneath them, is a dashed outline with people standing on it.

Figure 2.1The three layers of a companyContext has no software of its own
03GatlasFigure 3.1

What Gatlas is

Gatlas is the context layer for a company or a group of companies.

It connects the tools each company already runs and keeps a current copy of their data. Your people say what the data means, and an agent writes it down.

No tool is replaced. People, agents, dashboards and documents all read from that one place.

Your tools

  • e-conomic
  • Shopify
  • HubSpot
  • Google Drive
  • Xero
  • Stripe
  • Attio
  • QuickBooks

Gatlas

Approved corrections come back as notes

Where it is used

Your own agents

  • Claude
  • ChatGPT
  • Cursor

Run by Gatlas

  • Gatlas agentIn Slack
  • Automations

Built in Gatlas

  • Dashboards
  • Documents
  • WorkflowsPlanned

Lines run from a column of tools into Gatlas, drawn as three stacked parts: Data, Engine and Action. From Action, lines run out to agents, dashboards and documents, with a dashed line to workflows. One line returns to Engine.

Figure 3.1From each tool, through Gatlas, to each useDashed: planned
04PartsFigure 4.1

Data, engine, action

One part connects the data, one holds what it means, and one puts it to use.

Data
  • Accounting and banking
  • CRM and calls
  • Files and documents
Engine
  • Datasets
  • Descriptions
  • Notes
Action
  • Agents
  • Dashboards
  • Documents
  • Automations

Three stacked plates: Data at the bottom, Engine in the middle and Action on top. Each plate carries pins that name what it holds.

Figure 4.1The three parts of GatlasIllustrative
  1. Data

    Connects the tools each company runs.

    Gatlas reads each tool on a schedule and keeps a current copy of its data. The tool keeps its data, and no company changes how it works.

  2. Engine

    Holds what the data means.

    A dataset is a table built from your data, such as a monthly P&L, with a written description. Corrections your team approves are saved as notes. Every agent that reads a company reads both.

  3. Action

    Puts it to work for people and agents.

    Your own agent answers from it. Dashboards, documents and decks are built on it. Alerts and scheduled posts run on it.

05Definitions

Who writes the definitions

Today your own agent writes the definitions with you. Gatlas drafting them for your approval is planned.

Your agentYour own agent, as you work with itYou, in the conversationToday
Co-pilotGatlas, from a survey of your dataYour people, before anything is savedPlanned
AutonomousGatlas, on its ownYou set the limits and read what changedPlanned
06Example

How a correction is saved

A correction is made once and kept.

  1. Ask

    A finance lead asks an agent for gross margin by company.

  2. Correct

    The answer is wrong for one company, which books freight under overhead. The finance lead tells the agent.

  3. Approve

    The agent offers to save the correction as a note. The finance lead says yes.

  4. Reuse

    The note reaches every agent that reads that company.

07Outcome

Answered once

Count the work in your company that only moves information: finding a number, explaining it again, forwarding the file. With Gatlas each answer is written down once, and every person and agent reads it from there.

  • Exports
  • Figures re-keyed into a report
  • The same question answered again
  • Status updates
  • Handovers
  • Forwarded threads
08Status

Today and planned

What each part does today, and what is planned.

Data
  • 35+ tools across files and documents, CRM and calls, commerce and payments, work tools and HR, analytics and databases, and accounting and banking.
  • You connect a tool yourself, or send a setup link to whoever holds the login.
Planned Files no tool holds, such as brand files.
Engine
  • Your own agent, such as Claude or ChatGPT, writes datasets and notes with you.
Planned Gatlas drafts the definitions for you to approve, including where one company counts a figure its own way.
Action
  • Answers in your own agent. It reaches Gatlas through MCP, the open standard that AI agents such as Claude and ChatGPT use to connect to other software.
  • The Gatlas agent in Slack.
  • Dashboards, documents and decks.
  • An alert when a sync fails, a dashboard or document posted in Slack on a schedule, and a dataset kept current in a spreadsheet.
Planned Agents and workflows you set up in Gatlas.
09Trust

Guarantees

What holds for every answer, in every company.

Sources
Every number traces back to its sources
Freshness
Each dataset says when its data was last refreshed
Access
Each person and each agent sees only the companies you grant
Log
Every request an agent makes is logged
Storage
Each company’s data syncs into its own set of tables in Gatlas

Build your context layer

Start with one company and its tools. Add the rest when you are ready.