Chat
Chat surfaces, context, session configuration, and reviewing AI changes
Read time: 5 min
title: "Chat" description: "Chat surfaces, context, session configuration, and reviewing AI changes" section: "Copilot" readTime: "5 min"
Copilot Chat
Chat enables natural language AI assistance for coding tasks. Four surfaces are available, each optimized for different workflows.
Chat Surfaces
| Surface | Shortcut | Best for |
|---|---|---|
| Chat view | Ctrl+Alt+I | Multi-turn conversations, agents, multi-file edits |
| Inline chat | Ctrl+I | In-place edits in editor or terminal |
| Quick Chat | Ctrl+Shift+Alt+L | Quick questions without leaving current view |
| Command line | code chat | Starting chat from outside VS Code |
Configure Your Session
When starting a chat session, four settings shape behavior:
- Session type — where the agent runs (local, background, cloud)
- Agent — role/persona (Agent, Plan, Ask, or custom)
- Permission level — how much autonomy (Default Approvals / Bypass / Autopilot)
- Language model — which AI model powers the conversation
Adding Context
| Method | What it provides |
|---|---|
| Implicit | Active file, current selection, file name (automatic) |
#file / #folder / #symbol | Explicit file/folder/symbol reference |
#codebase | Semantic search across entire workspace |
#fetch | Content from a web URL |
@vscode / @terminal / @github | Domain-specific participants |
| Vision | Attach images (screenshots, UI mockups) |
| Drag & drop | Drop files from Explorer or editor tabs onto Chat |
Submit Your First Prompt
- Open Chat view (
Ctrl+Alt+I) - Select Local from the agent target dropdown
- Select Agent from the agent picker
- Type a prompt and press Enter:
Create a basic calculator app with HTML, CSS, and JavaScript
- Review suggested changes and Keep or Discard
Review and Manage Changes
After the AI edits files:
- Inline diffs — open a changed file to see diffs; use overlay controls to keep/undo individual changes
- Checkpoints — VS Code auto-snapshots files; roll back to a previous state anytime
- Stage to accept — staging changes in Source Control automatically accepts pending AI edits
Getting Better Responses
- Be specific: reference files, symbols, use
/commands - Add custom instructions for project-wide coding standards
- Use
/compactwhen the conversation grows too long - Start new sessions for unrelated tasks (context pollution degrades quality)
- Use the model picker (
Ctrl+Alt+.) to switch to a model better suited for the task
Troubleshooting Responses
Use Agent Logs (... menu → Show Agent Debug Logs) to inspect tool call sequences, LLM requests, token usage, and prompt file discovery. Use the Chat Debug view for raw system prompt and tool payloads.