Chat Interface

The Chat tab is your primary interaction surface with Agentforce Vibes. From here you prompt the agent, switch execution modes, select models, start new projects, and manage conversation context.

The Chat interface includes:

  • Conversation tabs—manage multiple conversations simultaneously. Click + to start a new conversation.
  • Prompt input—the text field where you type requests. You can @mention files and run /commands directly from the input.
  • Mode selector—Toggle Plan mode from the bottom bar.
  • Model selector—choose which AI model handles your request.
  • Context indicator—shows current token usage (for example, 72.2k / 200k) with a breakdown of messages, autocompact buffer, and skills.

Each conversation opens in its own tab. To manage tabs:

  • Rename— click a tab and use the pencil to give the conversation a descriptive name.
  • Delete— Delete a tab by closing it. To search across your conversation history or export a conversation, use the history icon in the top bar.
  • Reorder—drag tabs to rearrange them.

Agentforce Vibes streams responses inline as the agent works, so you see output as it generates rather than waiting for a complete response. When the agent makes file changes, a diff view appears inline in the chat so you can review what changed before accepting.

To stop the agent mid-response, click the Stop button that appears in the input bar while the agent is generating.

When you start a new conversation, the Chat interface displays categorized prompt suggestions to help you get started:

  • CHAT—quick answers about your org, code, or configuration. Examples: “Trace an Apex error to its root cause,” “Explain the trigger execution order in this org,” “Review my LWC for accessibility issues.”
  • PLAN—multi-step workflows with approval before execution. Examples: “Add a validation rule, then deploy to sandbox,” “Build and deploy a change set across orgs,” “Migrate metadata between two sandboxes.”
  • PLUGINS—combine Skills and @mentions for targeted actions. Examples: /sf-apex @AccountTrigger.cls refactor the handler to use a dispatch pattern, /sf-deploy @package.xml validate against my production org.

Click + New Project in the top-right corner to scaffold a new application. Available project types:

  • React App
  • Agent
  • Lightning App
  • Mobile App

Each project type provides templates and starter prompts designed for quick onboarding. See Build Apps and Agents for details.

Click the context indicator (circle icon) in the bottom bar to view how your context budget breaks down:

  • Messages—the percentage of context consumed by conversation history.
  • Autocompact buffer—reserved space for summarizing earlier messages when the context grows large.
  • Skills—the percentage consumed by active skill instructions.

The agent manages context automatically through auto-compaction, but you can start a new conversation when context runs high to get a fresh window.