Make Your Office Smart! Installing Claude and Codex Agents Directly on Your Server?

A futuristic graphic showing AI agents for different departments operating within a company server.
AI Summary

A 'self-hosted Company OS' has emerged, allowing businesses to run AI agents like Claude Code and Codex on their own servers, enabling department-specific AI workflows without security concerns.

Imagine this: You arrive at the office in the morning and tell your AI assistant, “Summarize last month’s sales data and write a draft report by department.” But instead of sending your material to an external cloud server, this AI processes everything within your company’s secure underground server room, training only on your company’s data to generate the results. The worry of data leakage disappears, and your company’s unique workflow remains intact.

Recently, a “Company OS” project—an AI-integrated system designed to assist with corporate tasks—unveiled by a developer named Dimitris on the Hacker News community, has been garnering significant attention. Reference 1 This environment allows companies to install powerful tools we know, such as “Claude Code” (an AI agent that assists with development), directly on their own servers to be freely used by various departments. Reference 1, Reference 10

Why is this important?

Until now, many companies have hesitated to adopt AI due to “data security” concerns. They do not want core corporate secrets transmitted to external cloud services. However, the newly introduced Company OS utilizes “self-hosting” (a method of operating software by installing it directly on your own server rather than renting an external service). Reference 1, Reference 4

Simply put, it creates “your own secure AI island” where company data never leaves the premises. Each department can have its own AI agents to manage work schedules, use necessary tools, and accumulate their own institutional memory. Reference 10

Easy Understanding: “An AI Factory Dedicated to Our Company”

This system can be compared to an “AI factory dedicated to our company.”

  1. Agents (AI Assistants): The smart, skilled workers in the factory. “Claude Code” or “Codex” (another AI model that helps with coding) fulfill this role. Reference 5, Reference 10
  2. Sandbox (Isolated Safety Zone): The safety fence inside the factory. Using a technology called “bubblewrap,” it thoroughly isolates the AI workers so that no matter how hard they work, they cannot leak information outside the factory, and outside hackers cannot break in. Reference 10
  3. Rule Files (CLAUDE.md / AGENTS.md): The factory work manuals. If you place a file named CLAUDE.md in the project folder, Claude Code reads this manual every morning when it “clocks in” to work. Codex performs the same tasks using a manual called AGENTS.md. Reference 6

In other words, once you provide the manual stating, “Our company works by these rules,” the AI processes tasks safely within the company server while adhering to those rules.

What can it do?

Currently, this system provides independent workspaces for each department. Each agent moves independently with its own work memory and schedule. Reference 10 Specifically for enterprise environments where security is critical, it uses a network isolation technology called “pasta” to completely block unnecessary external connections. Reference 10

“Claude Code” is widely known as a tool that helps developers understand and edit code, and it can be installed and used directly based on open source. Reference 5, Reference 12 However, it is worth noting that basic technical knowledge regarding server construction is required to utilize it properly as a Company OS.

What lies ahead?

Moving forward, it is expected that more companies will opt for direct installation of models suited to their server specifications instead of complex cloud subscription models. Because the project released is open source—free for anyone to use—it is highly likely to develop into an easier and more powerful management tool as more developers contribute to it. Reference 1 The era of hiring and managing your company’s own “AI assistant team” is drawing near.

MindTickleBytes AI Reporter’s Perspective

The advancement of technology is returning from the “public space” of the cloud back to the “private space” of the company. Ultimately, just as important as how smartly you use AI is how safely you protect your company’s valuable data while collaborating with it—this will be the core of future competitiveness.

References

  1. Show HN: Self-hosted company OS, Claude Code and Codex agents in departments
  2. [VueHN 2.0 Show HN: Self-hosted company OS, Claude Code and…](https://vue-hackernews-ssr-5cavbdjcta-ew.a.run.app/item/49630606)
  3. Show HN: Self-hosted company OS, Claude Code and Codex agents in departments
  4. Self-hosted company OS, Claude Code and Codex agents in departments
  5. Claude
  6. Codex in VSCode: How to connect GPT-5.6 and configure an AI agent
  7. [Show HN: Self-hosted company OS, Claude Code… HackerNews](https://news.ycombinator.com/item?id=49630606)
  8. [Natural 20 — AI News in Real-Time The Bloomberg Terminal for AI](https://natural20.com/c/1bbk9g)
  9. [Claude Code by Anthropic AI Coding Agent, Terminal, IDE](https://claude.com/product/claude-code)
AD
Test Your Understanding
Q1. What is a key feature of the recently introduced Company OS?
  • It only works on cloud servers
  • Anyone can install and self-host it for free
  • It cannot be used without a paid subscription
This system is provided as open source, allowing companies to install and operate it directly on their own servers for free.
Q2. Which technical methods are used to maintain AI agent security?
  • Password strengthening
  • Applying sandbox kernel and network isolation technologies
  • Maintaining constant internet connectivity
Each agent runs safely inside the server, secured through sandboxing (bubblewrap) and network isolation (pasta) technologies.
Q3. How are project rules or commands transmitted to the AI agent?
  • Entering them only in a dedicated app
  • Creating rule files like CLAUDE.md or AGENTS.md in the project folder
  • Typing into a chat window every time
Claude Code learns and executes project rules via the CLAUDE.md file, while Codex uses the AGENTS.md file.
Make Your Office Smart! Ins...
0:00