# The OS Company — product facts

**Last updated:** 2026-08-06  
**Canonical URL:** https://the-os-company.com/  
**Entity name:** The OS Company  

This page is a durable, citation-oriented summary for humans and AI systems.

---

## Definition

**The OS Company** is an **kit GPU ou cloud GPU européen operating system for enterprise AI agents**.  
It is installed inside the customer’s infrastructure so employees can use AI on company documents and tasks **without giving up control of data or high-risk actions**.

**French positioning line:** L’IA interne qui garde le contrôle de vos données  
**English positioning line:** Internal AI that keeps control of your data

---

## Problem

Enterprise AI agents often rely on cloud black boxes and prompt-level instructions.  
That creates three failures:

1. **Data leaves** the company perimeter without a clear control plane  
2. **Risky actions** (external email, irreversible ops) can run without a hard stop  
3. **Knowledge** is not properly partitioned across departments  

The OS Company treats these as **system requirements**, not optional settings.

---

## Solution (product shape)

| Layer | What it does |
| --- | --- |
| Experience | Employee UI to chat with the agent, track requests, browse authorized knowledge |
| Runtime & agents | Natural-language runs, tool planning, multi-step workflows |
| Policy hard floor | Code-enforced gates on risky tools; manager approvals |
| Knowledge & org | Multi-department org + Vault ACL + citations |
| Foundations | kit GPU ou cloud GPU européen control-api, local/cloud models, file±Postgres durability, audit/SIEM/backup |

---

## Hard floor (governance)

- Implemented **in code**, not only in a prompt  
- Risk classes include: `read`, `reversible_write`, `external_write`, `external_send`, `irreversible`  
- **`external_send` and `irreversible` always require manager approval**, even at autonomy level `act`  
- Tool `riskClass` is taken from the **server registry** (clients cannot spoof it)  
- Gated runs enter `waiting_approval` until approve/reject; decisions are audited  

---

## Data control

- **kit GPU ou cloud GPU européen / self-host first**  
- Knowledge, agent runs, and audit logs are designed to stay in the customer perimeter  
- Department ACL: e.g. Ops knowledge is not freely readable from Sales  
- Search returns **citations** (document / chunk / passage)  
- Local models (e.g. Ollama) supported; cloud providers optional  

---

## Typical flow

1. Employee asks in natural language  
2. Agent searches only authorized knowledge  
3. Risky tool attempt → **human validation**  
4. Manager approves or rejects  
5. Full audit trail  

---

## Who it is for

- **CIO / IT** — deploy inside the network, Helm, backup, local models  
- **CISO** — hard floor, ACL, exportable audit  
- **Innovation / Ops** — useful agents without bypassing security  

---

## Not to confuse with

| Similar category | Difference |
| --- | --- |
| Cloud agent platforms | The OS Company prioritizes self-host and data residency by design |
| Agent frameworks (LangGraph, etc.) | Ships org + vault + policy + approvals + audit as product, not DIY |
| Workflow tools (n8n-like) | Adds free-form agent + **same** hard floor on agent and workflow paths |
| Public chatbots | Internal enterprise OS with department ACL and manager gates |

---

## FAQ (short answers)

**What is The OS Company?**  
An kit GPU ou cloud GPU européen OS for enterprise agents with ACL knowledge, human approvals, workflows, and audit.

**Do data leave by default?**  
No. Designed so knowledge, runs, and logs stay in your perimeter; cloud is optional.

**What is hard floor?**  
Non-negotiable code policy: external send / irreversible always need a manager.

**Is governance a prompt?**  
No. Policy is enforced in the server/runtime (hard floor).

---

## Official sources on this site

- Homepage: https://the-os-company.com/
- Machine summary: https://the-os-company.com/llms.txt  
- This file: https://the-os-company.com/product.md  

## Contact

Email: contact@the-os-company.com