Skip to main content

​
Interactive Mode

Pi’s interactive mode provides a powerful terminal interface for working with AI agents. The interface consists of four main areas: startup header, message history, input editor, and status footer.

​
Interface Overview

┌─────────────────────────────────────────────────────┐
│  Startup Header                                     │ Shows shortcuts, loaded context files, skills
├─────────────────────────────────────────────────────┤
│                                                     │
│  Messages                                           │ Conversation history, tool calls, results
│                                                     │
├─────────────────────────────────────────────────────┤
│  Editor                                             │ Input prompt area (border color = thinking)
├─────────────────────────────────────────────────────┤
│  Footer                                             │ Working dir, session, tokens, cost, model
└─────────────────────────────────────────────────────┘

​
Starting Interactive Mode

# Basic interactive mode
pi

# With initial prompt
pi "List all TypeScript files in src/"

# Continue previous session
pi -c

# Resume from session selector
pi -r

# With specific model
pi --model anthropic/claude-sonnet-4

# With thinking level
pi --thinking high "Solve this complex problem"

​
Editor Features

1

File References

Type @ to open a fuzzy search for project files. Navigate with arrow keys and press Enter to insert the file path.
@src/core/agent.ts  # References the file
2

Path Completion

Press Tab to autocomplete file paths as you type.
3

Multi-line Input

Press Shift+Enter (or Ctrl+Enter on Windows Terminal) to add newlines without submitting.
4

Image Input

  • Ctrl+V: Paste images from clipboard
  • Drag & Drop: Drag image files onto the terminal
5

Bash Commands

Execute bash commands directly from the editor:
  • !command - Runs command and sends output to the LLM
  • !!command - Runs command without sending output
!git status          # Sends git status to LLM
!!npm install        # Runs silently

​
Built-in Commands

Type / in the editor to trigger commands:

​
Authentication & Models

CommandDescription
/loginOAuth authentication for providers
/logoutSign out from OAuth providers
/modelOpen model selector
/scoped-modelsEnable/disable models for Ctrl+P cycling

​
Session Management

CommandDescription
/resumeBrowse and select from previous sessions
/newStart a new session
/name <name>Set display name for current session
/sessionShow session info (path, tokens, cost)
/treeNavigate session tree and branch points
/forkCreate new session from current branch
/compact [prompt]Manually compact context with optional instructions

​
Export & Sharing

CommandDescription
/copyCopy last assistant message to clipboard
/export [file]Export session to HTML file
/shareUpload as private GitHub gist with shareable link

​
Configuration

CommandDescription
/settingsOpen settings UI (thinking level, theme, message delivery)
/reloadReload extensions, skills, prompts (themes auto-reload)
/hotkeysDisplay all keyboard shortcuts
/changelogShow version history

​
Exit

CommandDescription
/quit, /exitExit Pi

​
Keyboard Shortcuts

Press /hotkeys to see all shortcuts. Customize in ~/.pi/agent/keybindings.json.

​
Essential Shortcuts

KeyAction
Ctrl+CClear editor
Ctrl+C (twice)Quit Pi
EscapeCancel/abort current operation
Escape (twice)Open session tree
Ctrl+DExit Pi
Ctrl+ZSuspend (returns to shell)

​
Advanced Keyboard Shortcuts

For full editing shortcuts (delete word, undo, etc.), see the keybindings documentation.

​
Message Queue

Send messages while the agent is working:
1

Steering Messages

Press Enter to queue a steering message. It’s delivered after the current tool execution, interrupting remaining tools.Use this to redirect the agent mid-turn:
> Agent is running bash commands...
> You type: "Actually, use Python instead"
> Press Enter
> Agent stops after current command and receives your message
2

Follow-up Messages

Press Alt+Enter to queue a follow-up message. It’s delivered only after all work is complete.Use this for new requests:
> Agent is working...
> You type: "Also check the test coverage"
> Press Alt+Enter
> Agent finishes current work, then receives your message
3

Managing Queued Messages

  • Escape: Abort and restore queued messages to editor
  • Alt+Up: Retrieve queued messages back to editor

​
Queue Configuration

In /settings or ~/.pi/agent/settings.json: 
{
  "steeringMode": "one-at-a-time",  // or "all"
  "followUpMode": "one-at-a-time",  // or "all"
  "transport": "auto"                // or "sse" or "websocket"
}
  • one-at-a-time (default): Waits for response between messages
  • all: Delivers all queued messages at once
  • transport: Connection type for providers with multiple options

​
Session Tree Navigation

Press Escape twice or use /tree to navigate your conversation history: 
┌─ Session Tree ────────────────────────────────────┐
│ 1. User: Initial request               [main]    │
│ 2. Assistant: Response                           │
│ 3. User: Follow-up question                      │
│ 4. Assistant: Response                           │
│ └─ 5. User: Alternative question      [branch-1] │ <- Select to continue here
│    6. Assistant: Response                        │
└──────────────────────────────────────────────────┘

​
Tree Controls

  • Type: Search/filter messages
  • ←/→: Page through results
  • Ctrl+O: Toggle filter modes (default → no-tools → user-only → labeled → all)
  • L: Label entry as bookmark
  • Enter: Continue from selected point

​
Example Workflows

# Start Pi with high thinking for planning
pi --thinking high

# In Pi:
> "Create a new TypeScript library for parsing markdown"
> Agent creates files...
> /tree  # Review what was created
> /name markdown-parser
> /session  # Check session location and stats

# Resume previous session
pi -c

# Or select from list
pi -r

# In Pi:
> "Continue implementing the parser"
> /model  # Switch to faster model for iteration
> Ctrl+P  # Cycle between configured models

# In Pi:
> "Let's try a different approach"
> /fork  # Creates new session from current point
> Agent continues in new branch
> /export experiment.html  # Save for later review

​
Tips & Tricks

Use Ctrl+P for quick model switching. Configure your favorite models with --models flag:
pi --models "claude-sonnet-4,claude-haiku,gpt-4o"
Paste large code blocks. Pi automatically detects large pastes and offers to collapse them to save tokens.
Label important points in /tree view with the L key to create bookmarks for easy navigation.
Use thinking levels strategically:
  • off/minimal: Quick tasks, file operations
  • medium: Standard coding tasks
  • high/xhigh: Complex debugging, architecture decisions

Build docs developers (and LLMs) love

Get started for freeTalk to us