Skip to main content

Open the chat panel

Click the CodePilot icon in the activity bar. CodePilot main chat screen

Choose a mode

From the dropdown to the left of the input box, choose the mode that fits your goal. Mode selection dropdown For the full breakdown, see the Chat modes documentation.

Try your first request

Understanding code (ASK)

ASK mode only answers — it doesn’t modify code. Try it first when getting your bearings in an unfamiliar codebase.

Building a feature (CODE)

CODE mode has the AI create and modify files directly.

Handing off a goal (AGENT)

AGENT mode takes a goal, breaks the work down on its own, and follows through with verification after editing files. If you use a frontier model, we recommend AGENT mode.

Planning first (PLAN)

PLAN mode presents an implementation plan first without touching any files. It’s useful for checking the scope of impact before a large task.

Attaching context

Type @ in the input box to pick the context you want to hand to the AI. @ menu
Type / to use shortcut commands such as commit, review, and session management. Slash command menu

Approving / undoing changes

When the AI modifies a file, the changes are shown in the editor as an inline diff. Diff screen
  • Keep — Apply the change.
  • Undo — Revert the change.
You can also handle everything at once in the Pending changes area at the bottom of the chat panel.
CODE and AGENT modes change actual files. Check your Git commit state before any important work.

Running multiple tasks at once

Large requests are automatically split into several subtasks and processed in parallel. You can track progress in real time in the task list in the chat panel. Task progress

Reviewing the results

When the AI finishes, it shows a summary of what it did and a list of the files that changed. Task complete Below the response, you can confirm this turn’s changes all at once or expand the list of files the AI referenced. Referenced documents

Next steps

Chat modes

How to use CODE / AGENT / ASK / PLAN

Key features

Automatic error fixing, security guardrails, RAG

Settings

AI models, build verification, security settings