Commands & Shortcuts

Slash Commands

All commands start with / and can be typed at the > prompt.

/help, /h, /?

Show available commands and current mode status.

/clear, /c

Clear the conversation history and start fresh. Resets the system prompt and clears the terminal.

/model [name]

Without arguments, lists all available models on the connected LM Studio server. With a model name, switches to that model.

> /model                              # List available models
> /model mistral-7b-instruct-v0.3     # Switch to a specific model

/config

Opens an interactive configuration editor. Walks through each setting section and lets you modify values.

/session list|save [name]|load <id>|delete <id>

Manage conversation sessions.

> /session list           # Show saved sessions (up to 20)
> /session save           # Save current session (auto-generated title)
> /session save My Task   # Save with a custom title
> /session load abc123    # Load a saved session by ID
> /session delete abc123  # Delete a saved session

Sessions are stored in ~/.openorca/sessions/. See Sessions & Context for details.

/plan, /p [on|off]

Toggle plan mode. In plan mode, the model plans before executing and asks for approval.

> /plan          # Toggle plan mode
> /plan on       # Enable plan mode
> /plan off      # Disable plan mode

/compact [instructions]

Summarize the conversation to free up context window space. Optionally provide focus instructions.

> /compact                              # Compact with default summarization
> /compact focus on the auth changes    # Compact with specific focus

The compaction preserves the most recent turns (configurable via compactPreserveLastN). See Sessions & Context.

/rewind [N]

Remove the last N conversation turns. Defaults to 1 if no number is given.

> /rewind        # Remove last turn
> /rewind 3      # Remove last 3 turns

/context, /ctx

Show context window usage: estimated tokens, window size, usage percentage, auto-compact threshold, and a visual progress bar. Also shows message counts by role.

/stats, /cost

Show session statistics: duration, total turns, output tokens, messages in context, average tokens per turn.

/memory [show|edit|list|auto|clear-auto]

View or edit project instructions and manage auto memory.

> /memory           # Show current ORCA.md contents
> /memory show      # Same as above
> /memory edit      # Open ORCA.md in your $EDITOR (default: notepad on Windows, nano on Linux/macOS)
> /memory list      # List all auto-generated memory files with previews
> /memory auto on   # Enable auto memory (saves learnings at session end)
> /memory auto off  # Disable auto memory
> /memory clear-auto # Delete all auto-generated memory files

See Project Instructions for details on ORCA.md and Auto Memory for the auto memory system.

/doctor, /diag

Run diagnostic checks against your setup. Tests:

Check	Description
LM Studio connection	Verifies connectivity and lists loaded models
Model configured	Shows current model or "Auto-detect"
Tools registered	Shows count of registered tools
Config file	Checks if ~/.openorca/config.json exists
Log directory	Verifies write access to ~/.openorca/logs/
Session storage	Checks session directory accessibility
Default prompt template	Looks for ~/.openorca/prompts/default.md
Project instructions	Checks for ORCA.md in the current project
Native tool calling	Shows current setting

/copy, /cp

Copy the last assistant response to the clipboard. Strips <think> tags before copying.

• Windows: uses clip
• macOS: uses pbcopy
• Linux: uses xclip -selection clipboard

/export [path]

Export the full conversation to a markdown file.

> /export                             # Export to openorca-export-{timestamp}.md
> /export my-session.md               # Export to a specific file

The export includes system prompt (truncated), all messages by role, tool calls with arguments, and tool results.

/checkpoint [list|diff|restore|clear]

Manage file checkpoints. OpenOrca automatically snapshots files before any file-modifying tool (edit_file, write_file, delete_file, copy_file, move_file) executes.

> /checkpoint list              # List all checkpointed files with timestamps
> /checkpoint diff src/main.cs  # Show diff between checkpoint and current file
> /checkpoint restore src/main.cs  # Restore file to its checkpointed state
> /checkpoint clear             # Delete all checkpoints for the current session

See File Checkpoints for details.

/init

Scaffold a new .orca/ORCA.md project instructions file with a starter template. Uses ProjectInstructionsLoader.FindProjectRoot() to locate the project root, then creates .orca/ORCA.md with sections for overview, architecture, code style, testing, and common commands.

If an ORCA.md already exists, it tells you and suggests /memory edit instead.

/diff

Show uncommitted git changes in formatted Spectre.Console panels:

• Staged changes — shown in a green-bordered panel with stat summary + full diff
• Unstaged changes — shown in a yellow-bordered panel with stat summary + full diff

If there are no changes, prints "No uncommitted changes."

/undo

Revert AI-made (or any uncommitted) changes with a confirmation step:

1. Shows a panel with git diff --stat of all staged + unstaged changes
2. Offers three choices via an interactive prompt:
   • Revert all — runs git reset HEAD + git checkout .
   • Stash changes — runs git stash
   • Cancel — does nothing

/rename <name>

Rename the current session. Requires an active session (save first with /session save if needed).

> /rename Refactoring auth module

/add <file1> [file2] ...

Add file contents to the conversation context so the LLM can reference them. Supports glob patterns.

> /add src/Program.cs                    # Add a single file
> /add src/*.cs                          # Add all .cs files in src/
> /add src/Models/User.cs tests/UserTests.cs  # Add multiple specific files

Each file is injected as a user message with a path header. Files are truncated at 50K characters.

/ask [question]

Without arguments, toggles Ask mode — a persistent mode where all input is sent to the LLM without tools. The prompt shows [ask] ❯ when active. Toggle off by running /ask again or pressing Shift+Tab.

With arguments, performs a one-shot ask — sends the question without tools and returns to the current mode.

> /ask                                                    # Toggle ask mode on/off
> /ask what is the difference between Task and ValueTask in C#?  # One-shot ask
> /ask explain the strategy pattern with an example              # One-shot ask

/exit, /quit, /q

Exit OpenOrca.

Custom Commands

You can define project-specific or global custom slash commands by creating .md files in command directories.

Command Directories

Location	Scope
.orca/commands/	Project-level (checked first)
~/.openorca/commands/	Global (used as fallback)

Creating a Custom Command

Create a markdown file named after the command. For example, .orca/commands/review-pr.md:

Review PR #{{ARG1}} thoroughly. Check for:
- Code style consistency
- Security issues
- Test coverage
{{ARGS}}

Template Variables

Variable	Description
{{ARGS}}	All arguments joined with spaces
{{ARG1}}, {{ARG2}}, ...	Positional arguments

Usage

> /review-pr 123                # Runs the review-pr.md template with ARG1=123
> /deploy staging --verbose     # Runs deploy.md with ARGS="staging --verbose"

Custom commands inject the expanded template as a user message and run it through the agent loop. Built-in commands always take priority over custom commands with the same name.

See Custom Commands for more details.

Shell Shortcut

Prefix any command with ! to run it directly in the shell, bypassing the LLM:

> !git status
> !ls -la
> !python script.py

Output is displayed in a panel with exit code. Commands time out after 120 seconds. Output is truncated at 5000 characters.

On Windows, commands run via cmd.exe /c. On Linux/macOS, via /bin/bash -c.

Keyboard Shortcuts

Shortcut	Action
Shift+Tab	Cycle input mode: Normal → Plan → Ask → Normal. The prompt indicator updates immediately (❯, [plan] ❯, [ask] ❯).
Ctrl+O	Toggle thinking visibility. When hidden, you see a token counter. When visible, you see the full streaming output in cyan. Can be toggled mid-generation.
Ctrl+C	First press: Cancel the current generation. Second press within 2 seconds: Exit the application.
Escape	Clear the current input line.

Input Modes

OpenOrca has three input modes, cycled with Shift+Tab:

Mode	Prompt	Behavior
Normal	❯	Full agent loop with all tools available
Plan	[plan] ❯	Model plans without executing, then asks for approval
Ask	[ask] ❯	Chat without tools — faster, cheaper responses

You can also switch modes with commands:

• /plan or /plan on|off — toggle or set Plan mode
• /ask (no args) — toggle Ask mode

Multi-Line Input

End a line with \ to continue on the next line. In non-interactive mode (piped input), Console.ReadLine() is used as a fallback.