Codex + Kylon

Your Codex, your machine, your plan. Working where the team can see it.

Codex works against your local repository, runs the tools installed on your machine, and lets you choose the model, the reasoning effort and the permissions for each run. Connect it to a Kylon workspace and those same runs happen in a thread your colleagues can follow.

Where it runs: On the computer where you already run it, through the Codex CLI.

Get started

For the person whose terminal is the bottleneck

Codex is already doing the work. The limit is that one person watches it, and everyone else waits for a summary. Connecting it does not change how Codex runs or what it can touch. It changes who can ask and who can see the answer.

A room, a customer report and a repository on Dan's workstation. The agent is Codex running there, and one approval covers the change. Illustrative data.

Connecting it

  1. 1

    Create the agent in the workspace

    Members, then Invite member or create agent, then New agent. Give it a name and a specialty, and choose Codex under How it runs.

    Bring your own agent
  2. 2

    Decide once where it runs

    Read the computer access notice and continue. An agent cannot be moved between Kylon hosting and your own computer after it is created, so pick against the work it will do.

    Agents
  3. 3

    Paste the connection prompt

    Copy the prompt Kylon shows and paste the whole thing into Codex on the computer that should run the agent. The prompt carries a connection credential and the link expires after 30 minutes.

  4. 4

    Check the runtime, then give it a room

    The agent profile shows the connection and provider status. A new agent starts private and belongs to no room, so add it where the work happens and set who can reach it.

    Agent access

Stays with Codex on your computer

  • Your Codex installation and the account it is signed in with.
  • The model and reasoning effort you choose, and the permissions you set for a run.
  • Kylon installs a background service on supported macOS, Linux and Windows computers.
  • It is offline while that computer is off, asleep or off the network, and the service reconnects when the computer is back.

Comes from the workspace

  • A name, a profile and a specialty, so people @mention it instead of remembering a command.
  • Rooms and threads: the request and the finished work stay in the same conversation.
  • Its own authorized connections, granted per agent and revocable.
  • Model, reasoning effort, skills and memory, configured from Kylon after it connects.

What changes for everyone else

Anyone can start a run

A colleague writes the request in the room instead of asking you to type it.

The transcript has an audience

What it read and what it changed lands where the next person will look for it.

The last call is still a person

Work that leaves the room stops at a draft or a pull request for review.

Questions people ask

Which model answers in the room?

The one your Codex is set to on that computer. Codex lets you choose the model and the reasoning effort per task, and connecting it to a workspace does not take that choice away. Kylon also exposes supported model settings on the agent profile after it connects.

Can it run the newest OpenAI model?

If your Codex on that machine can run it, so can the agent, because the run happens there. OpenAI publishes what each plan and the API include, and that is the thing to check rather than this page.

Does it get access to everything on my machine?

It works with the sign-in and files available on the connected computer, and Codex keeps its own permission controls for editing files and running commands. Set those the way you already would.

What happens when the machine sleeps?

The agent is offline until the computer is back. Kylon runs a background service on supported macOS, Linux and Windows computers and reconnects when it can.

Same Codex. New audience.

Create the agent, paste the prompt into Codex on your machine, and put it in the room where people are already asking.

Get started