Skip to content
Open Source Claude Cowork

The self-hosting handbook

An open-source CompanyOS that you host yourself

Claude Cowork became part of Claude on September 16, 2026, and the question is where the work runs. Kortix is the open-source CompanyOS: your agents, their skills, your company memory, and every connector in one git repo you own, with the agents working on real cloud computers, and it is my pick; I work at Kortix, so weigh that accordingly.

Open source · One repo you can read · Your hardware or ours

how work lands
how work landshttps://kortix.com/docs/guide

The leading open-source alternative to Claude Cowork and ChatGPT Work

Your agents, skills and settings live in the project’s shared repository. Kortix keeps it safe from unreviewed changes:Each session works on its own branch, a private copy of the repository.When the agent finishes, it opens a change request: a request to merge that branch into the shared one.The change request appears on the Review page.A person reads the changes and approves, closes, or asks for changes.An agent can merge its own change request only when it holds the merge permission.

Session work reaches main through a change request. Merge is default-deny for agents.

The same three checks, applied to each option

  1. Open source

    YesYesKortix
    • Claude CoworkNoNo
    • ChatGPT WorkNoNo
    • OpenWorkYesYes
  2. What you can read

    The whole platform: agents, skills, memory, connectors, sourceKortix
    • Claude CoworkProduct page and the Help Center
    • ChatGPT WorkOpenAI's product and help pages
    • OpenWorkThe desktop app's source
  3. Where it runs

    Kortix Cloud, your VPC, or on-premKortix
    • Claude CoworkAnthropic's cloud, or Amazon Bedrock, Gemini Enterprise Agent Platform, or Microsoft Foundry
    • ChatGPT WorkOpenAI's hosted service, plus desktop and mobile apps
    • OpenWorkYour desktop; self-host or managed instance
  4. Self-host

    YesOne Docker Compose stack, freeKortix
    • Claude CoworkNoNot offered
    • ChatGPT WorkNoNot offered
    • OpenWorkSelf-host or managed private instance
  5. How work lands

    A change request a person mergesKortix
    • Claude CoworkAuto on Pro, Max, Ultra and Team; Manual asks first on Enterprise
    • ChatGPT WorkYou approve the plan in Plan mode
    • OpenWorkWaits for a yes before it posts

What a session hands back

Three jobs, three artifacts, from an ordinary workweek.

  • Customer support

    Triage the weekend's tickets before standup.

    A trigger starts the session on Monday morning: the agent groups the tickets, drafts the replies, and flags the two that need a person.

    support/triage-mon.md
    1. Bug reports18
    2. Refund requests6
    3. Feature asks9

    3 groups · replies in review

  • Legal review

    First pass on a folder of contracts.

    The agent reads each agreement against the playbook and returns a memo per contract with the clauses that depart from it.

    legal/contract-review.xlsx
    ContractStatusFlag
    Northwind renewalFlaggedAuto-renewal clause
    Globex MSAEscalatedNo liability cap
    Acme NDAClearedTerm within policy

    3 contracts · 1 escalated

  • Hiring ops

    Screen the application stack overnight.

    The session reads each application against the role's rubric, scores it, and drafts the follow-up for everyone who makes the shortlist.

    hiring/screen-tue.md
    1. Applications read41
    2. Strong fits5
    3. Follow-ups drafted5

    5 candidates advance

Start with three commands

The docs' own commands: an install, a first session, and a trigger.

terminal
curl -fsSL https://kortix.com/install | bashkortix init my-agentkortix ship
terminal
kortix sessions new --prompt "Add a skill that drafts the weekly status report"kortix chat <session_id>                      # talk to the agent; -p "…" sends one promptkortix cr ls && kortix cr diff 1              # read its change request
kortix.yaml
triggers:- slug: weekly-reporton:cron: "0 0 9 * * 1"timezone: Europe/Berlinrun:agent: supportprompt: Write the weekly support report for the last 7 days.

GitHub stars on the Kortix repo, October 2026

Open source. Read it, fork it, audit it.

The merge, in plain terms

Anthropic announced the change on 16 September 2026: one Claude answers a question and then takes on the task. What Cowork could do now happens in any conversation, with the context, skills, and connectors you already have. Existing Cowork work carries over: chats, projects, artifacts, connectors, and skills (the announcement).

The rollout is staged. Pro, Max and Ultra plans get it first, in the Claude app on web, desktop, and mobile over the coming weeks. Team plans have it in beta behind a switch an owner turns on, Enterprise plans have it where an owner turns it on, and in existing Enterprise organizations it turns on by default on 16 November 2026 (the Claude Help Center).

The Cowork product page is still live, and it still sells the capability with paid plans. It also carries the note that Cowork consumes limits faster than Chat. Its own FAQ draws the line between Anthropic's two tools: Cowork is for non-coding knowledge work, and Claude Code is for software engineering. The capability stayed inside Claude. What went away was the separate Cowork surface.

The question the merge leaves open

On a Claude plan the model is the vendor's, and the files and connectors sit in the vendor's product. Anthropic's third-party deployment mode sends inference and billing to a provider the organization configures (the 3P overview), and the app stays Anthropic's either way. None of that configuration is yours to copy out. The merge added no new switch: the files and connectors stayed in Anthropic's product. The question an alternative has to answer is direct: where does the system that does the work run?

This handbook answers it with three checks: what you can read, what you can host, and what you can review. None of them requires asking a vendor for permission.

What you can read

Kortix is open source, so the readable part is the whole system. Agents are markdown files, skills are folders beside them, and company memory is plain files that accumulate. At the root, kortix.yaml names what each session gets. The starter repo lays it out plainly: kortix.yaml with agents/, skills/ and memory/ beside it.

For the wider field, how the open-source options differ sorts frameworks, visual builders, and full platforms by the job each one actually fits.

What you can host

Self-hosting Kortix is one Docker Compose stack on a Linux box you control. The self-host commands install and run the full stack on your own infrastructure. The same configuration runs unchanged on Kortix Cloud if you would rather start there. The self-hosting guide takes a bare box from nothing to a running stack, and the runtime, the data, and the configuration stay wherever you put them.

What you can review

The six lines in the hero artifact are the whole gate, read top to bottom. A session works in its own sandbox on a branch of its own, and the agent commits there before opening a change request against the default branch. A person with the merge permission reads it and merges it. Nothing merges itself: work reaches main through a change request a person approves, unless an admin granted the merge capability. That capability is the project.gitops.merge grant, and an admin is the one who grants it. Without that grant, the merge waits for a person. Anthropic's side of that gate depends on the plan. Auto is the default on Pro, Max, Ultra and Team, where Claude keeps working and automated safety checks run before each action. Manual is the default on Enterprise, where it asks before each action and the person allows each one (the Claude Help Center). The difference is where the gate sits: here it is the repo's own merge path, on hardware you hold.

Where the choice stands

Reaching the capability got easier on September 16: open Claude and hand it a task. Where the system around the capability runs is still a decision about hardware and a repo you keep. On Claude Cowork and ChatGPT Work the readable part is a product page and the approval is a setting inside the app, while OpenWork keeps the files on your desktop. A stack you host puts the review in a merge you own, and Kortix vs Claude Cowork takes the two products apart point by point.

Questions before you choose where the work runs

  • Is there an open-source alternative to Claude Cowork?

    Yes. Kortix is the open-source CompanyOS, and the whole system is readable: agents, skills, company memory and connectors are files in one git repo, and the code itself is public. The same comparison sets it beside Claude Cowork and the open-source desktop options.

  • Can Kortix be self-hosted?

    Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free. The self-hosting guide takes a bare Linux box from nothing to a running stack.

  • What happened to Claude Cowork in September 2026?

    Anthropic merged Cowork and chat into one Claude on September 16, 2026: the capability now runs in any conversation, with your existing context, skills, and connectors (the announcement). Pro, Max and Ultra went first in the rollout, Team plans have it in beta behind an owner's switch, and existing Cowork work carries over.

  • Is there a free alternative to Claude Cowork?

    Claude Cowork works with a paid plan today: Pro starts at $17 a month, and usage limits apply (the product page, checked October 2026). On Kortix, Free is $0 with 500 credits per month, every feature and 1 project, on your own provider keys or ChatGPT subscription. Self-host is free too, so the open-source route adds no subscription beyond the hardware.

  • Who reviews an agent's work?

    You do, by default. The agent works on its own branch and opens a change request, and a person with the merge permission reads the diff and merges it into the default branch. Nothing merges itself: work reaches main through a change request a person approves, unless an admin granted the merge capability.

opensourceclaudecowork.com

© 2026

Put the work where you can read it

Start on Kortix Cloud, or self-host the whole stack on your own box.