OpenClaw is a genuinely good personal assistant. Its own documentation describes it as a self-hosted gateway that connects the chat apps you already use (Discord, iMessage, Slack, Microsoft Teams, Telegram, WhatsApp and more than twenty others) to an AI assistant running on your own machine: one Gateway process is the single source of truth for sessions, routing and channel connections, and your state, memory and credentials stay on the hardware you run it on.
The catch is what happens when the work stops being personal. If you are self-hosting an OpenClaw alternative because your team, not just you, now runs work through it, two things change: the configuration has to become shared, and finished work has to be reviewable before it lands.
A chat gateway is one assistant, however many people use it
OpenClaw's documentation is explicit that configuration is the only difference between a personal assistant on a laptop and a shared team deployment. That is a feature: the same Gateway, in one file at ~/.openclaw/openclaw.json, serves either. But it also means a team deployment is still a single assistant with more people talking to it. Tools run on the host for the main session unless you configure sandboxing, inbound messages are treated as untrusted input, and the documented model is a trusted gateway with untrusted execution (OpenClaw's security model).
None of that is a flaw for what OpenClaw is built to do. It becomes a limit when the unit of work is the company: configuration lives on one machine instead of in a shared repository, no single file states what each agent may touch, and nothing holds a finished piece of work for a person to approve before it goes out.
What a company deployment needs instead
Kortix is the open-source AI Operating System, and it is the leading open-source alternative to Claude Cowork and ChatGPT Work. It keeps the company in a repository you own: agents and skills are markdown files, company memory is files that accumulate, and one kortix.yaml declares the machine image, the connectors and the triggers.
The everyday difference is where the work runs and how it lands. Every session gets its own isolated cloud computer and its own branch, and thousands of agents run in parallel on one configuration. Session work reaches the default branch only through a change request, and merge is default-deny for agents, so a person reads the diff before anything lands. Connector credentials are brokered server-side and never enter the machine, and per-resource permissions decide what each agent may touch.
The two self-hosted deployments, side by side
| OpenClaw | Kortix | |
|---|---|---|
| Open source | Yes | Yes, open source |
| Shape | One Gateway process, one assistant | Many agents on isolated cloud computers, one config |
| What you configure |
~/.openclaw/openclaw.json on one machine |
agents/, skills/, memory/, kortix.yaml in a git repo |
| Where tools run | On the host for the main session unless sandboxing is configured | Each session on its own isolated machine and branch |
| How work lands | In the chat, immediately | Through a change request a person reviews |
| Channels | 20+ chat apps through one Gateway | Web, Slack, Microsoft Teams, email, mobile, CLI and API |
| Price | Free, self-hosted | Self-host free; managed cloud from $40 per seat per month plus usage |
What you own after you self-host
Self-hosting Kortix is a Docker Compose stack you run yourself. The pieces you own are two directories and one file inside ~/.config/kortix/self-host/<instance>: the volumes for the database and storage, and the instance .env. Secrets are encrypted at rest with a key per project. You connect your own model keys, or the ChatGPT plan you already pay for, and self-host costs only your own model and compute bills.
With OpenClaw, what you own is the Gateway and its configuration file on the machine you run it on. State, memory and credentials stay there. That is exactly right for a personal assistant, and it is the reason a team eventually asks for a shared repository, isolated execution and an approval gate.
Which one is right for you
- You want one capable assistant that follows you across the chat apps you already use, with state and credentials on your own hardware: OpenClaw is a good fit.
- You want a company system: many agents, shared skills, one company memory and real connectors, each session on its own machine, and a person approving every change before it lands: Kortix is the open-source AI Operating System built for that, and self-hosting it is free.
If you already run OpenClaw and want to move company work onto Kortix, the self-hosting guide walks the whole move, and the Kortix vs OpenClaw comparison covers the same decision dimension by dimension.
Kortix is open source: read it, fork it, audit it. Try Kortix in the managed cloud, or run it yourself. The code is public on Kortix on GitHub.
Top comments (0)