EASYCOMPANY

Confidential — Investor Access

This deck is shared for investor discussion only. Enter the access phrase you were given.

EASYCOMPANY
← Back to the deck

AtlasOS — product and integrations

As of 23 August 2026 · what runs today · what comes next →

What it is

An AI agent with real write access to the systems a business actually runs on. It does not answer questions about the work, it does the work: files invoices, updates records, prepares meetings, triages the inbox, runs scheduled operations overnight. What separates it from a chat window is not language quality but what happens afterwards — access to the real systems, and the accountability that comes with it.

Operates rather than answers

Every action runs against the real systems, with real credentials, and produces a result you can check.

Reviewable

Each run happens in isolation and ends in a diff a human approves — never a silent change.

Extended by skills

A skill is a documented procedure. New capability comes from writing, not from engineering.

Existing integrations

SystemForAccess
Microsoft 365Outlook mail and calendar — Teams, SharePoint and OneDrive follow in August 2026read & write
GmailMail, labels, drafts, sendingread & write
NotionTasks, projects, notes, journal, personnel filesread & write
ExxasInvoicing and ERP — receivables, payables, offers, subscriptionsread & write
TopalAccounting — chart of accounts, payables, VAT periodsread & write
smallinvoiceInvoicing for mandates outside Exxasread & write
GitHubIssues, pull requests, code, releasesread & write
Google DriveFiles and backupsread & write
ownCloud / oCISFile storage on Swiss hostingread & write
Home AssistantBuilding and device controlread & control
TelegramRemote operation by voice and textbidirectional
iMessageMessages and contacts (macOS)read & send

Twelve production integrations, built by one person. That says less about the person than about the tool: AtlasOS is built with itself. A new connector is a documented skill plus credentials, not a project. It is why the plan treats an engineering team of five to seven as sufficient where a conventional estimate would arrive at four or five times that.

The customer brings their own AI subscription

Supported backends

  • Claude CodeAnthropic
  • CodexOpenAI
  • CursorCursor
  • Ollamalocal, offline

What that means economically

  • Inference cost to usCHF 0
  • Gross margin 2031 (model)92.7%
  • Credentialsper user
  • Vendor lock-innone

AtlasOS never resells model access. That removes the largest variable cost almost every AI product carries, and is the actual reason gross margin sits above 90% — not pricing, but an absent cost line. It also makes the customer indifferent to which model leads in two years: they change their subscription, not their platform.

From Q4 2026 there is a second option: Swiss-hosted LLMs. A customer will be able to choose a model that never leaves Switzerland. Today that is the one place where the sovereignty argument has a hole — the subscription a customer brings is usually with a US provider, and it is the first thing a data-protection officer asks about.

Scale today

37
operational skills in daily use
20
skills in the public catalog
12
production integrations
3
private skill repositories
17
production blueprints in the Marketplace

Private skill repositories are encrypted catalogs shipped with the product: a customer unlocks internal procedures with their own credentials, and those procedures are not even visible to anyone else. Three are already running — EasySystems, EasyCompany Admin, EasyAcademy.

How it is operated

Hosted in the Marketplace

Multi-tenant on the EasySystems Marketplace, data in Switzerland, no US hyperscaler touching customer data.

Three ways in

Web interface, command line and Telegram. The same agent with the same permissions, reached differently.

Scheduled operations

Recurring work runs unattended on a schedule — overnight housekeeping, briefings, research, billing runs.

What is still missing

Two things a prospect would notice immediately. Onboarding your own repository is not yet self-service — a customer buying today needs help with it, which is a demo risk rather than only a documentation gap. And the customer course in the Marketplace (7 chapters, 29 lessons, German and English) is written but deliberately unpublished, because it lacks screenshots and a reviewed render. Both are tracked and scheduled, and both are part of what the bridge financing pays for.

← Back to the deck

Confidential — prepared for investor discussion only. Not an offer to sell securities.