Open-source CompanyOS comparison: which of these names runs the agents

The review a session's work goes through

Five names, five different objects

Every product in this comparison is sold under the CompanyOS name, and each one promises something different by it. companyos.cc shares knowledge, company-os.ai hands a team a local Office app and a template, companyos.com runs a vertical platform, Aqtos BOSS* keeps records and projects with an agent attached. The tomevault Tome is a folder of skills you fork. Put them side by side and the phrase stops being the useful part of the comparison.

companyos.cc shares what the team already knows

companyos.cc puts the same facts, procedures, and templates in front of the AI account each person opens, inside Claude, ChatGPT or Codex. Files and connected tools stay read-only to the AI, so the product is a shared reference: the team stops answering one question four ways, and no agent gets a machine out of it. It is free during beta, then €29 per seat a month excluding VAT, with the model provider's own plan on top.

company-os.ai hands you an Office app to run yourself

company-os.ai calls itself a free template for running companies AI-natively: one Office where a team and its AI agents keep goals, decisions, services and money. It arrives as an open-source template plus a local Office app you start yourself with python3 companyos.py serve. It costs nothing, starts no agent session and hosts nothing for the team.

companyos.com sells a vertical platform

companyos.com is the Operational Intelligence Platform for Manufacturers and Distributors. It captures processes, decision rules and exceptions, connects them to ERP data, so employees and AI read one source. Its agents carry out defined tasks in connected systems, inside the access and limits the customer approves, and the vendor runs a paid pilot end to end. A factory whose orders already move through an ERP is the buyer this fits.

Aqtos BOSS* sells records with an agent attached

Aqtos sells a SaaS that connects the work your team does with the clients, people, money and decisions behind it, over five areas: Work, People, Money, Office and Tech. An AI agent reaches it through an MCP connection and uses the connected account's role and permissions. DATA is in beta, and BCG and Aqtos AI are in alpha as of October 2026. The agent sees exactly what that account sees, which is a clean boundary and a narrow one.

The tomevault Tome ships skills and no runtime

The tomevault repository describes itself as a complete, opinionated, fork-friendly operating system for an AI-instructed company: thirteen markdown skills, a justfile template and constitutional documents, all open source and free to fork. For a team already running a coding agent it is the cheapest useful starting point in the field, and it stops where a runtime would begin. Nothing in it starts a session, provisions a machine, or holds a change until a person reads it.

The two questions that decide the purchase

Ask whether the software runs an agent. companyos.cc and company-os.ai answer no, and that is a fair design: they hand context to an AI that runs somewhere else, and they are judged on the quality of what they hand over. companyos.com and Aqtos answer yes, inside limits the customer approves. Kortix answers yes and provisions the machine the agent works on.

Ask where the memory lives once the trial ends. In the two knowledge products it sits in an account owned by Claude, ChatGPT or Codex, or in a .zip on one laptop. At companyos.com and Aqtos it sits in the vendor's platform. In the Tome and in Kortix it sits in a git repository, and a repository clones.

Who each one suits

• No engineers and no agents this year. companyos.cc at €29 a seat is the cheap fix for a team whose AI gives four answers to one question.
• A founder writing the manual before choosing software. company-os.ai, or the Tome if a coding agent is already in the loop.
• A manufacturer or distributor with an ERP. companyos.com, run as a pilot by the vendor.
• A services firm that wants clients, money and office in one place. Aqtos BOSS*, with the agent inside the account it was granted.
• An engineering team whose agents need procedures. The Tome first, then a runtime when the skills outgrow a laptop.

Two further results sell implementation services rather than software: a consultancy will build and run one of these with you, priced by the hour or the project. That is a different purchase from a product and belongs in a separate budget line.

Why Kortix is the recommendation

Kortix is the one name here that answers both questions with the same repository. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free. The agents, the skills they follow and the memory they read are files in a git repo the company owns. A change of host, or of vendor, starts with a clone taken whenever you like.

The other half of the question is what lands in that repo while agents work. Session work reaches main through a change request. Merge is default-deny for agents. Someone reads the diff before it counts, which is the job a knowledge layer never asks anyone to do.

What that asks of you is a short list: the host or the cloud account, the model keys, the change requests waiting for a person, and the rules you set for each agent. The checklist page names the parts in the order a buyer meets them, and the self-hosting page covers what stays on your own network.

Set beside those five rivals, the line is narrow and it is the one that decides the purchase. A knowledge layer makes an AI better informed. A template makes a team write things down. A vertical platform fits one industry, and a forkable Tome gives a coding agent its procedures. Kortix runs the agents itself, on machines it provisions, and keeps the record where the company can carry it out. Read the docs is the shortest way to check that.