Chat quickstart

Chat runs agents in a sandbox: a remote machine that belongs to one project. This page gets you from an empty project to a working conversation.

1. Open a project

Open Chat from the sidebar and pick a project.

2. Configure the sandbox

Select Configure Sandbox and choose what the sandbox starts with:

• Capabilities — the agents, tools, and skills to install. Pick at least one if you want an agent other than the default.
• Resources — CPU, memory, and Max idle time: how long the sandbox keeps running after you close the tab.
• Secrets — which of your saved secrets the sandbox receives. All of them by default.
• Variables — plain environment variables. Put API keys in Secrets instead.

Select Save. You can skip this step; the sandbox starts with defaults.

3. Send a message

Choose an agent and a model, type what you want done, and press Enter.

Your first message starts the sandbox, which takes anywhere from a few seconds to several minutes depending on what your capabilities install. The message sends once the sandbox is ready. Later messages send immediately.

Sandboxes are billed while running

A running sandbox uses credits whether or not the agent is working. When you’re done, select the sandbox status at the top of the session and choose Pause or Stop.

Next

• Runtimes in Chat — pause, stop, add capabilities, and troubleshoot.
• Agents & models — switch agent, model, or reasoning effort, and run autonomously.
• Commands — slash commands.