The open source Claude Cowork alternatives worth self-hosting

Claude Cowork is one of the best agents you can put on a desktop, and when a company needs to own that same power, the open-source AI Management System Kortix is the one built for it. Most searches for a Claude Cowork alternative come from one of three walls: you want to self-host, you want a model other than Claude, or you want the setup shared across a team instead of sitting inside a vendor's product. The projects that can replace Claude Cowork clear those walls in different ways, and Kortix clears all of them.

What Claude Cowork is, and what it does not hand you

Claude Cowork is Anthropic's agent for non-coding knowledge work. Give it a goal, point it at folders and connected tools, and it plans the steps, opens a browser when it has to, and returns finished work for your review. It runs on web, desktop and mobile, and it runs on a Claude account or your own cloud provider: Amazon Bedrock, Google Cloud, or Microsoft Foundry (Anthropic's Cowork page).

For a single knowledge worker, Cowork is strong and polished. The limit for a team is ownership. Your plugins, skills and connectors are configured inside Anthropic's product, not in a repository you clone. The model is Claude. The runtime stays in Anthropic's cloud or a cloud account you rent, never on hardware you control. Teams looking to leave usually want those three things back.

What an open source alternative has to give you

Replacing Cowork means taking back four things. A project that returns only some of them still leaves part of your setup outside your control.

• A repo you own. Agents, skills, memory and connector config should be files in a repository you can clone, diff and roll back, not records in a vendor's database.
• Any model with your keys. The point of leaving is picking the model per job, so bring your own key to Anthropic, OpenAI or Gemini, or point at your own OpenAI-compatible endpoint.
• A runtime you control. Self-hosting on your laptop, a VPS, your VPC or on-prem keeps your data and your credentials inside your own walls.
• A review gate. Agents should propose changes and stop, so a person reads the diff and merges before anything lands in production.

The open source alternatives, compared

Six open source projects can replace Claude Cowork for a specific job. Kortix is the one built to carry a whole company.

Kortix is the only project in the list built as a company platform; the rest each hand you a single agent or a single interface. Agents, skills, company memory and connector configuration are files in one git repo you own, and every change to them moves through a change request a person reads as a diff.

OpenHands is the choice for an engineering team that wants a control center for coding agents. It runs OpenHands, Claude Code, Codex or Gemini across local, Docker, VM or cloud backends, with any LLM, and it self-hosts cleanly (OpenHands on GitHub). Its focus stays inside the software lifecycle, and the agent gets full access to the host filesystem unless you sandbox it.

Open WebUI is a self-hosted AI interface for a team's own models. It puts Ollama and any OpenAI-compatible API behind one UI, with users, groups and permissions, and its Open Terminal gives an agent a terminal and filesystem inside the chat (Open WebUI on GitHub). Reach for it when you want a chat and search surface over your models.

n8n is a fair-code workflow automation platform. You build each job on a visual canvas, run it self-hosted or in n8n Cloud, and connect 1500+ integrations with native AI steps (n8n on GitHub). It fits when the job is the same steps in the same order every time, and it fights you when the work is open-ended.

OpenClaw is a personal assistant that meets you in the chat apps you already use. It is MIT-licensed, runs on your own machine, and keeps state, memory and credentials on your hardware (OpenClaw on GitHub). It is built for one person; its tools run on the host with broad access by default, which is fine on your own laptop and a problem once several people can steer it.

Hermes is Nous Research's personal agent, MIT-licensed and built around a learning loop that writes reusable skills as it works. Run it on a $5 VPS or a serverless backend, talk to it from Telegram, Slack or the terminal, and point it at any model endpoint (Hermes on GitHub). It is a strong always-on personal agent, and team roles, tenant isolation and org-wide audit are not what it is built for.

OpenWork is the closest desktop clone of Cowork. It is built on OpenCode, runs on macOS, Windows and Linux, works with any of 50+ providers or local models through Ollama, and lets a team share skills and MCP servers (OpenWork on GitHub). The desktop app is MIT-licensed and local-first; the team control plane, OpenWork Den, is under the OpenWork EE License and needs a subscription to run in production.

Why Kortix is the one to own

Every alternative here hands you one layer. Kortix is the whole stack in one repo you control.

Kortix is open source (Elastic License 2.0): self-host, read and modify the code. Run it on your laptop, a VPS, your VPC or on-prem, or use the managed cloud, so your data and your models stay where your security team wants them. Any model works, per agent, per session or per message, with your own keys, the ChatGPT subscription you already pay for, or your own OpenAI-compatible endpoint (Kortix docs).

The repo is the company. Agents and skills are markdown, memory is files that accumulate, and connectors and triggers are declared in kortix.yaml. You can grep the whole company, diff any change and roll any part of it back. Connector credentials are brokered server-side and never enter the machine, and you set each tool call to allow, ask or block, down to the arguments it was given (Kortix on GitHub).

Every session gets its own isolated Linux machine and its own branch. The agent can install, run and break anything, and only what it commits survives. Work reaches the main branch through a change request a person approves, so the company improves one reviewed change at a time. Agents reach 3,000+ apps plus any MCP, OpenAPI, GraphQL or HTTP API, and you can start the same session from the web app, Slack, Teams, email, the CLI or a scheduled trigger, with the result landing back in the repo you started from.

The migration path: install, init, ship

Moving from Cowork to Kortix is three commands. Install the CLI, scaffold a project, and ship it.

# 1 · Install the CLI
curl -fsSL https://kortix.com/install | bash

# 2 · Scaffold a project: kortix.yaml, agents, skills and runtime config
kortix init

# 3 · Ship it: push the repo and bring the whole thing live
kortix ship

From there, hand a session a real task and read what it proposes before it lands:

kortix sessions new --prompt "Summarize this week's commits and open a change request"
kortix cr ls   # read the proposed change; merge to keep it

Nothing an agent does reaches your main branch until you merge the change request. If you would rather not install anything, sign up at kortix.com, create a project and start a session in the browser.

When to pick which

Cowork and Kortix solve different problems, and they can coexist. Keep Cowork for one person's desktop tasks and run the company's shared, governed workforce on Kortix.

Next reads: Kortix vs Claude Cowork walks the head-to-head, and personal AI agents vs a company OS covers where OpenClaw and Hermes fit.

Try open-source Kortix at kortix.com and keep the whole setup in a repo your team owns.