Start a conversation

Open Chat in the active workspace and write a request. Attach files from your computer or the workspace when the agent needs to read them.

The sidebar keeps your conversation threads. Start a new thread when the goal changes. Continue the existing thread when the agent still needs earlier decisions or corrections.

Choose a model

The model picker shows the models available to your account. Models differ in speed, cost, how much context they can read, and which tools they can use. Try another model when the current one does not suit the task.

Use /model in the composer to search and switch models. A model change is recorded in the thread so you can see where the conversation changed direction.

Attachments and artefacts

Attachments are files you send with a message. Artefacts are files the agent creates or uses while it works. Open the artefact panel to preview supported files, download them, or copy them into the workspace.

Copy an artefact to the workspace when you want to keep it. Temporary working files can stay in the thread.

Tools, skills, and subagents

Some models can use tools, follow specialised skills, and give independent work to subagents. The thread shows when the agent uses them and what they return.

  • Use $ to find an available skill.
  • Use /new to begin a clean thread.
  • Use the stop control to interrupt an active run.
Tool and agent availability depends on the selected model, your plan, and the current product configuration. Experimental tools may change or disappear.

Review the result

Models can misunderstand sources or invent details. Check important claims against the original material. Read generated files and test code before relying on them.

For high-stakes legal, medical, financial, or safety decisions, use a qualified professional rather than treating a generation as authoritative advice.