Overview
Loop’s interactive mode is a full-featured terminal UI (TUI) built with ratatui. It renders streaming markdown with syntax highlighting, provides tool execution cards, and gives you full control over the agent with keyboard shortcuts and slash commands.The Interface
The TUI is composed of several areas:- Status bar — Shows the active model, sandbox status, git branch, and session info
- Conversation pane — Streaming markdown responses with syntax highlighting
- Tool cards — Expandable/collapsible cards showing tool invocations and results
- Workflow task cards — Live multi-agent progress (with
--features orchestration) - Input area — Multi-line input with history and file mentions
Writing Messages
Basic Input
Type your message and press Enter to send. For multi-line input, use Shift+Enter or Ctrl+J to insert a newline.File Mentions
Reference files directly in your prompt with@:
Follow-up While Streaming
Press Alt+Enter to queue a follow-up message while the agent is still generating. It will be sent after the current turn completes.Local Shell Commands
Prefix a command with! to run it locally without sending it to the model:
Model Switching
Model Picker
Press Ctrl+L to open the interactive model picker. Browse available models, filter by typing, and select with Enter.Quick Cycle
Press Ctrl+P to cycle forward through your scoped models list. Ctrl+Shift+P cycles backward.Slash Command
Thinking Modes
Toggle AI reasoning/thinking visibility:
When enabled, you’ll see the model’s chain-of-thought reasoning before its response.
Tool Cards
When the agent invokes tools (reading files, editing code, running commands), each invocation appears as a tool card:- Cards show the tool name, arguments, and result
- Press Ctrl+O to expand or collapse tool, thinking, and workflow task detail, and to list an MCP server’s tools in
/mcp list - File edits display unified diffs with syntax highlighting
Workflow Progress
With--features orchestration, run a multi-agent workflow from the TUI:
WorkflowTask cards appear as each task starts (pending → success/error), with description and output preview. Failed tasks show a short error preview by default; Ctrl+O expands full detail. A final system line reports completed / failed / total counts.
See Multi-Agent Orchestration.
External Editor
Press Ctrl+G to open the current input in your external editor ($EDITOR). Save and close to send the content back to Loop.
Loop also supports opening diffs in VS Code or Cursor for file edit review:
Interrupting and Aborting
When you press Escape during generation, the agent stops and you can continue the conversation.
Session Navigation
Clipboard
Press Ctrl+X to copy the last assistant message to the clipboard. You can also use:Complete Keyboard Reference
Keybindings are fully customizable — see Keybindings.