Run multiple Codex sessions
and find the one you need at a glance.

Codex in the terminal or in VS Code works well while you have one task. But the tasks add up. When they run across different projects, you can only tell the chats apart by their messages, because in the list they all look alike.

Buzzpot is a Mac app that puts every Codex session on a canvas, one shared board, as a card next to its project. Each card has a name and a status. So you find the session you need by looking, not by rereading conversations.

In the morning you give Codex a task. Then another one, in a different project. By lunch you have ten chats, their names look alike, and the names no longer tell them apart.

It’s like hunting for the right thread among dozens of Slack chats. With one difference: in each of these threads, an agent is at work.

A project card is a folder on your disk. Next to it sit that project’s tasks, each on its own card with its own Codex session. Each project has its own area of the canvas. To switch projects, move the canvas.

Buzzpot saves the canvas after every change, so tomorrow, even after you restart your Mac, the cards will be where you left them.

A new card takes its name from the first line of your message. When Codex replies for the first time, it comes up with a title for the conversation, and the card takes it, unless you’ve already named the card yourself. Don’t like the title? Rename the card.

Cards on one topic can be grouped under a name and a color. You can also stick a note to the group. That way these tasks stay together and don’t get mixed up with the rest.

Every card has a status: Working, Waiting for input or Done. The agent changes the status itself when it starts working, asks a question or finishes. A card that’s waiting for an answer stays in Waiting for input until you reply.

When Codex asks for permission, your Mac shows a notification with the card’s name. With Remote control, the same notification reaches your phone, and you can answer right from there.

A card turns Done only when Codex itself reports that the work is finished, so an agent that just went quiet stays in Working.

Two Codex sessions in one folder can end up editing the same files. To run Codex agents in parallel without that, turn on Isolated workspaces for the project. Then each card works in its own copy of the project, and when Codex reports the work is done, Buzzpot brings its changes back into the project.

Link the tasks that should run one after another. The next card starts when the previous one turns Done.

How parallel tasks work →

Buzzpot has no model of its own. It runs the Codex CLI already installed on your Mac, under your account, so your AGENTS.md, settings and MCP servers work the same way they do in the terminal.

A Codex card has two pickers: Plan or Work, and, separately, the access level. When Codex approves an action without you, a gray line stays in the chat. In it, Codex says what it approved and why.

The terminal and the VS Code extension stay with you, because that’s where you read and edit code. Buzzpot shows all your Codex sessions at once.

When you switch from Codex to Claude and back, the windows, commands and habits change along with the agent. And every switch means getting used to things again.

In Buzzpot, the agent is one field on the card. The canvas, statuses and chat stay the same with any of the three agents. So you put Codex where it’s needed, without moving to another app.

You need a Mac, a Codex CLI you’re already signed in to, and the projects you’re working on now.

After that, all your Codex sessions are in their places. You don’t need a new OpenAI account or an API key, because Buzzpot works with the Codex you’re already signed in to. Buzzpot is free during the beta, and your code stays on your Mac.