Comparisons

The open-source ChatGPT Work alternative that lives in your repo

ChatGPT Work ships with every ChatGPT plan; Kortix is open source, and a company keeps its agents, grants and memory in a git repo of its own.

The Kortix Team
Kortix··6 min read

The open-source ChatGPT Work alternative that lives in your repo

ChatGPT Work is a capable agent inside OpenAI's product; a company that wants the same work on its own infrastructure runs Kortix, the open-source CompanyOS. Each agent lives in a markdown file under agents/, with its own block in kortix.yaml. A buyer asks who can read an agent's instructions and who approves a change to them. The other question is what each option costs before the first agent runs.

Compared here:
ChatGPT WorkChatGPT Work

What ChatGPT Work gets right

OpenAI launched ChatGPT Work on July 9, 2026 as an agentic mode inside ChatGPT, powered by GPT-5.6. OpenAI describes it as an agent that acts across your apps and files, stays with a project for hours and turns a goal into finished work (OpenAI announcement).

  • It returns a deliverable. Hand it a goal and a document, a spreadsheet or a presentation comes back, not a chat reply.
  • It holds a long task. A single piece of work can stay with a project for hours without a person driving each step.
  • It sits where the team already works. ChatGPT Work is a mode inside ChatGPT, so the login, the billing and the habits are the ones people have.
  • The model ships with the agent. GPT-6 Astra is available in ChatGPT Work (OpenAI). The GPT-6 Pro model needs Pro, Business or Enterprise (OpenAI Help Center). GPT-6 Astra for agent teams goes further on that.

Where the shape stops working

ChatGPT Work is closed source and stays in OpenAI's cloud. Nothing of it runs on hardware you control, and access comes through an OpenAI plan.

  • One vendor sets the models. Work runs OpenAI's models, and which one you get depends on the plan you bought.
  • Configuration lives in the OpenAI workspace. Agents, instructions and context are set up inside OpenAI's product and its admin console.
  • Access is metered. Usage draws on the plan's allowance, and credits buy more.

Work on every plan, and a two-seat floor for teams

Every plan carries Work, Free included. OpenAI's launch post puts the desktop app's three modes, Chat, Work and Codex, on all of them (OpenAI). Free and Go get a limited desktop version, and Plus adds desktop, web and mobile (ChatGPT pricing, checked October 2026).

Business is the team plan that carries Work for a company, and it is where the floor appears. Standard seats cost $20 a seat a month on annual billing, or $25 month to month (OpenAI Help Center, checked October 2026). Premium seats cost $100 and $125, with five times the usage and no five-hour cap. Every Business workspace needs at least two seats. The smallest real bill is two Standard seats: $40 a month annually, $480 a year, or $50 a month monthly, $600 a year. Usage sits on top, and OpenAI's API bills separately.

Free, Go and Plus suit one person trying the agent. A company that wants shared context, admin controls and central billing lands on Business, and Business starts at two seats.

Kortix puts the money on credits instead. Free is $0 with 500 credits per month on a person's first workspace (later workspaces start with 0) and every feature, on the user's own provider keys or ChatGPT subscription. Managed models are on the Credits plan only. Above the free tier, a hundred credits cost a dollar. The smallest paid plan is 5,000 credits a month for $50, and the workspace shares one credit pool at no per-member charge. Unused monthly credits roll over one month (pricing, checked October 2026). A credit buys the agent computer. The meter runs by the second while it works and stops when it idles; stored files are charged as storage. Two Business seats come to $40 a month billed annually. The Kortix paid plan starts at $50 a month in credits, so the two bills start close together.

Agents you can read, changes a person approves

One repo holds the workspace. Its manifest names the connectors, the triggers, the sandbox image and how far each agent can reach. The agent's behavior sits in its markdown file next to the manifest. The skills it draws on are markdown files too. A new workspace starts from the manifest kortix init writes. These are its lines from default_agent through the head of the worker block.

default_agent: mykortix

# Workspace skills: folders under skills/, registered by name. Each agent then
# enables them with its own `skills:` grant.
skills: {}

agents:
  worker:
    description: "General-purpose worker: coding, research, writing, ops and data work, done end to end in its own sandbox."
    mode: primary

The manifest has no version line. default_agent sets which agent a session or trigger runs as when it names none. A weekday digest that names no agent runs as mykortix here. The prompt on a trigger is the instruction it hands that agent. A trigger that runs an agent fails validation without one.

A manifest that leaves out skills, connectors, apps, secrets or kortix_permissions grants none of them. An agent opts in to skills with its own skills: grant. Three keys start open instead: tools is "*", permission defaults to allow, and codemode defaults to on. A grant widened from a session reaches main through a change request. The dashboard or kortix agents config writes the same grant into kortix.yaml and commits the default branch immediately. A wide grant can land there with nobody approving it first (docs). Either way the file is the record. Reading the repository answers the question a buyer asks after an incident: what was this agent told, which connectors could it call, and when did that last change.

Work lands the way it was configured. Session work reaches main through a change request. Merge is default-deny for agents: nothing merges itself unless the agent's grants include project.gitops.merge (docs). The starter agent ships with kortix_permissions: all, so it holds that grant out of the box. That grant is the setting to watch: it decides how much review a change gets. Remove it and no merge lands without a person's approval.

The model stays your choice: any provider, on keys you supply, or the ChatGPT plan you already pay for. Hosting too: run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free.

When ChatGPT Work is the better pick

For a person who wants a finished deck this afternoon, ChatGPT Work wins on the first minute. There is nothing to install, no repository to adopt, and no review process to agree with anyone. The agent, the model and the subscription arrive as one product. A team already paying for ChatGPT gets the mode without a migration.

What the two systems do differently

ChatGPT Work
Open source
No
Yes, the whole platform
Models
OpenAI models, plan-gated
Any model, your keys
Where it runs
OpenAI's cloud
Kortix Cloud, your VPC or on-prem
Self-host
No
Yes, free
Configuration
In the OpenAI workspace and admin console
Files in a repo you own
How work lands
A file returned to the chat
A change request; merge is default-deny for agents
Smallest team bill
Two seats, $40 a month billed annually
Free tier $0; paid credits from $50
Time to first result
Minutes, inside ChatGPT
A repo to set up first

Which one a company should run

Choose ChatGPT Work if…

You work inside ChatGPT and want finished deliverables from OpenAI's models on OpenAI's cloud, with no repo to run.

Choose Kortix if…

You want the platform your company owns, on any model, self-hosted or managed, with people reviewing what agents change.

Most teams end up running both, and the split follows the job. Quick thinking and one-off deliverables stay in ChatGPT, where they cost a seat. Work that repeats on a schedule, reaches the company's own systems and needs a record of who approved it moves to Kortix. That record lives in the repository next to the agent's instructions.

For the neighboring questions, Kortix vs Claude Cowork and what Kortix actually is go deeper. For the wider category, see the open-source Microsoft Copilot alternative.

Run the work on a platform you own

Free to start; hosting it yourself costs nothing.

Get started

Own your company’s AI.Start today, in our cloud or yours.

Start for free

Open source · SOC 2 Type I · SAML SSO and SCIM · Any model, your keys