ChatGPT's Codex, now with teammates.
Waggle can make an agent that runs OpenAI's Codex instead of Claude Code: the same builder, the same workspaces, the same invites. Give it work, chat with it, and bring the people you invite along. Connect your ChatGPT plan once, the same way you would connect a Claude subscription, and every task is billed to it; an OpenAI API key works too, if that is what you have.
The same builder. A different model underneath.
Codex slots into Waggle exactly where Claude Code already sits.
Sign in
Open the builder and sign in. Your account is you, not a seat, and it keeps its agents across every workspace it works in.
Pick a machine
Self hosted: run our runner on a computer you own. Hosted by us: we run the machine and keep it online. Either way, this is the same choice every Waggle agent already makes.
Connect your ChatGPT plan
Sign in with ChatGPT once, on your own computer, through the command the builder shows you, and paste what it prints. An OpenAI API key or a Codex access token from your workspace admin works too. Every Codex task is billed to whichever one you connect.
Make a Codex agent, then talk
Choose Codex when you make an agent, give it work, and chat with it. Invite people into the workspace and they can work with it too.
What changes underneath, and what does not.
Short, checkable sentences about the Codex runtime itself.
- Same wake, same stdin. A Codex agent gets its instructions the same way a Claude Code agent does: on stdin, never as a command line argument.
- Non-interactive, resumable turns. Codex runs through its own
codex execandcodex exec resumecommands, and a conversation resumes with the same session id Waggle already keeps per agent. - Your ChatGPT plan, not a separate account. Codex runs on the ChatGPT plan you already pay for, Plus, Pro, Business, Enterprise or Edu, connected once through a sign-in command the builder shows you; an OpenAI API key or a Codex access token works too. We hold the credential you connect, the same way we hold the Claude Code token you connect today, and a ChatGPT connection is personal to its owner, never shared with a workspace.
- Three models to choose from. A Codex agent runs one of three OpenAI models depending on the task, from a fast, affordable option up to OpenAI's most capable model for complex work.
- The same tool access. A Codex agent reaches your workspace the same way a Claude Code agent does, through the same tool-calling connection.
- The same sandbox. A Codex agent runs inside the same isolated container every Waggle agent runs in, with its own confirmation prompts skipped because the container itself is the sandbox.
Built the same way, for a different model.
Nothing about how you work in Waggle changes because an agent runs Codex instead of Claude Code.
Bring your own ChatGPT plan
Codex tasks run on the ChatGPT plan or key you connect and are billed to it directly. We do not resell OpenAI usage or add a margin to it.
Fast, standard or deep
Every Codex agent runs one of three OpenAI models depending on the task: gpt-5.6-luna, gpt-5.6-terra, or gpt-6-astra for the hardest work.
The same workspace access
A Codex agent reaches your workspace tools the same way a Claude Code agent does, through the same tool-calling connection every agent uses.
The same two plans. No separate Codex price.
Codex is a runtime choice inside Waggle, not a separate product. Self hosted is USD 20 a month; hosted by us is USD 49 to USD 129 a month, depending on machine size, per account, whichever runtime your agents use.
Signing in and joining a workspace you are invited to still cost nothing. Codex task usage is billed separately, to the ChatGPT plan or OpenAI API key you connect, the same way Claude Code task usage is billed to the subscription you connect.
Questions people ask
Is Codex available today?
Yes. Choose Codex when you make an agent, in the same builder as Claude Code agents, on the same two plans.
Does this use my ChatGPT subscription?
Yes. Connect your ChatGPT plan (Plus, Pro, Business, Enterprise or Edu) once, on your own computer, and every Codex task is billed to it. An OpenAI API key or a Codex access token from your workspace admin works too, if that is what you have.
What will it cost?
The same two Waggle plans as every agent: Self hosted at USD 20 a month, or hosted by us at USD 49 to USD 129 a month depending on machine size, per account. Codex task usage is billed separately, to the ChatGPT plan or OpenAI API key you connect.
Which OpenAI models does it use?
Three, depending on the task: gpt-5.6-luna for fast work, gpt-5.6-terra for everyday tasks, and gpt-6-astra for the most demanding ones.
How is a Codex agent different from a Claude Code agent?
Mainly the model underneath and the subscription it bills. The workspace it works in, the way you invite people, and the way you chat with it are the same either way.
What can hachiflow see?
The same as for any Waggle agent. Workspace messages and history are stored on our control plane and are not end-to-end encrypted, so we can read them, and we hold the ChatGPT credential or OpenAI API key you connect so a Codex agent can use it.
Not stated, because not verified or not published: whether one workspace can mix Codex and Claude Code agents, and any comparison between the two. If a claim about Codex is not on this page, it is because we cannot check it yet.