/push docs

Useful work, from your first session to your whole team.

Browse the docs
Get started · Tutorial

Your first useful session

Start an agent, give the work a home, and return to the same conversation later.

Push gives your agent conversations a place to live: work in a terminal, find the result on the board, and return when you need to continue.

Before you start

You need an Apple Silicon Mac on macOS 26 or later and an installed, signed-in coding agent: Claude Code, Codex, OpenCode, or Hermes. Push uses your agent subscription or provider account. See pricing for Push's current offering.

If someone invited you to their team's workspace, start with Push in Slack or the browser. You do not need to set up another Mac.

Open your first conversation

  1. Download Push, move it to Applications, and open it.
  2. Choose an available coding agent during setup. If one is missing, use Settings → Terminal → Re-probe runtimes after installing it.
  3. Open a terminal tab with ⌘T, choose your project folder, and start your agent. You can also run its usual command, such as claude or codex, in the shell.
  4. Give it a small, useful task: review a document, explain a project, or prepare a plan.

Give the work a home

In a fresh conversation that is not already tracked, ask:

Ask your agent
Use the Push skill to track this conversation as "Prepare the launch". Tell me where I can find the work in Push.

The Push skill guide explains /p and Codex's $p. If your conversation already belongs to a Push Session, keep working there.

For direct CLI use, the alternative is push track "Prepare the launch" in a shell inside Push. Do not type it into the agent conversation expecting the shell to execute it, and do not run both alternatives to create the same work.

The command applies to its own tab. For the conversation you already opened, asking the agent is the straightforward path; a command in another tab would track different work.

See the result

Find Prepare the launch on the work board. Open it to see the work's thread and connection to its terminal conversation. Ask the agent to explain its result and surface any files it produced.

After the agent has finished its turn, quit and reopen Push. Return through the work's Open or Reopen control. Conversation history and terminal scrollback serve different purposes; the transcript lets you return to earlier discussion.

If something stops

If the agent cannot find the skill or push, follow Set up the skill. If a conversation does not resume, keep its work reference and use troubleshooting.

Next, track and return to work, find your results, or set up Push for your team.

Updated