Skip to content

Command Reference

Complete reference for all DSPy Code slash commands available in interactive mode.

Overview

All DSPy Code commands are slash commands executed in interactive mode. Simply type / followed by the command name.

Project Management

/init - Initialize Project

Initialize a new DSPy project or scan an existing one.

Usage: 

/init [--fresh]

Options: - --fresh - Create a complete project structure with all directories

Examples: 

/init              # Minimal setup (just dspy_config.yaml)
/init --fresh      # Full project structure

What it does: - Creates dspy_config.yaml configuration file - Indexes your codebase for RAG features - Sets up project structure (if --fresh) - Displays entertaining messages during indexing

Code Generation

/create - Generate DSPy Components

Generate DSPy Signatures, Modules, or complete programs.

Usage: 

/create <type> <description>

Types: - signature - Create a DSPy Signature - module - Create a DSPy Module - program - Create a complete DSPy program

Examples: 

/create signature sentiment analysis with text input and sentiment output
/create module email classifier with subject and body inputs
/create program question answering system with RAG

Model Connection

/model - Interactive model selection

Interactively choose and connect to a model (local via Ollama or cloud).

Usage: 

/model

Use this if you're not sure which model to pick or don't want to type the full /connect command.

/connect - Connect to LLM

Connect to a language model for code generation and execution when you already know the model name.

Usage: 

/connect <provider> <model>

Providers: - ollama - Local models via Ollama - openai - OpenAI models (GPT-4, GPT-5 family) - anthropic - Claude models - gemini - Google Gemini models

Examples: 

/connect ollama gpt-oss:120b
/connect openai gpt-5-nano
/connect anthropic claude-sonnet-4.5
/connect gemini gemini-2.5-flash

Note: You may need to install provider SDKs separately. DSPy Code will guide you if needed.

Code Operations

/save - Save Generated Code

Save the last generated code to a file.

Usage: 

/save <filename>

Examples: 

/save sentiment_analyzer.py
/save modules/email_classifier.py

Note: Code must be generated first using natural language or /create.

/validate - Validate Code

Validate DSPy code for best practices and correctness.

Usage: 

/validate [filename]

Examples: 

/validate                    # Validate last generated code
/validate my_module.py       # Validate specific file

What it checks: - Syntax correctness - DSPy best practices - Signature definitions - Module structure - Import statements

/run - Execute Program

Run a DSPy program in a sandboxed environment.

Usage: 

/run <filename> [--input key=value] [--test-file data.jsonl]

Options: - --input - Provide input values as key=value pairs - --test-file - Run with test dataset

Examples: 

/run sentiment_analyzer.py
/run my_program.py --input text="This is great!"
/run classifier.py --test-file test_data.jsonl

Optimization

/optimize - Optimize with GEPA

Optimize a DSPy program using GEPA (Genetic Pareto).

Usage: 

/optimize <program.py> [training_data.jsonl]

Examples: 

/optimize my_program.py
/optimize classifier.py training_data.jsonl

Requirements: - A DSPy program file - Training data (JSONL format) with input/output examples - Connected LLM model

What it does: - Evolves prompts using GEPA - Improves reasoning steps - Tests on validation set - Reports performance improvements

Data Generation

/generate data - Create Training Data

Generate training examples for optimization.

Usage: 

/generate data <count> for <task>

Examples: 

/generate data 20 for sentiment analysis
/generate data 50 for email classification

Output: Creates a JSONL file with training examples.

MCP Integration

/mcp-add - Add MCP Server

Add an MCP server configuration.

Usage: 

/mcp-add <name> --transport <type> [options]

Transport Types: - stdio - Standard input/output - sse - Server-Sent Events - websocket - WebSocket connection

Examples: 

/mcp-add filesystem --transport stdio --command npx --args -y "@modelcontextprotocol/server-filesystem" /path/to/directory
/mcp-add github --transport sse --url https://api.github.com/mcp

/mcp-connect - Connect to MCP Server

Connect to a configured MCP server.

Usage: 

/mcp-connect <server-name>

Examples: 

/mcp-connect filesystem
/mcp-connect github

/mcp-tools - List MCP Tools

List available tools from connected MCP servers.

Usage: 

/mcp-tools [server-name]

Examples: 

/mcp-tools              # List all tools from all servers
/mcp-tools filesystem   # List tools from specific server

/mcp-resources - List MCP Resources

List available resources from connected MCP servers.

Usage: 

/mcp-resources [server-name]

/mcp-prompts - List MCP Prompts

List available prompts from connected MCP servers.

Usage: 

/mcp-prompts [server-name]

RAG & Performance

/refresh-index - Rebuild Codebase Index

Rebuild the RAG codebase index for better code generation context.

Usage: 

/refresh-index [--force]

Options: - --force - Force rebuild even if cache is fresh

Examples: 

/refresh-index           # Rebuild if stale
/refresh-index --force   # Force rebuild

What it does: - Re-indexes installed packages (dspy, gepa, mcp) - Updates project code index - Improves code generation quality

/index-status - Show RAG Index Status

Display current RAG index statistics and status.

Usage: 

/index-status

Shows: - Total indexed elements - Number of codebases - Cache size - Index age - Per-codebase statistics

/fast-mode - Toggle Fast Mode

Enable or disable fast mode for faster responses (disables RAG context building).

Usage: 

/fast-mode [on|off]

Options: - on - Enable fast mode (faster responses, lower code quality) - off - Disable fast mode (slower responses, better code quality) - (no args) - Show current status

Examples: 

/fast-mode on    # Enable fast mode
/fast-mode off   # Disable fast mode
/fast-mode       # Show current status

Trade-offs: - ✅ Faster responses (0.5-1s vs 2-5s) - ⚠ī¸ Lower code generation quality (no real DSPy examples) - ✅ Still uses templates and reference docs

/disable-rag - Disable RAG Completely

Disable RAG indexing for maximum speed.

Usage: 

/disable-rag

What it does: - Disables RAG index loading - Disables all RAG searches - Fastest mode (0.3-0.8s responses) - Code generation uses templates only

Note: Code generation quality will be significantly reduced. Use only if speed is critical.

/enable-rag - Re-enable RAG

Re-enable RAG indexing for better code quality.

Usage: 

/enable-rag

What it does: - Re-enables RAG indexing - Restores code generation quality - Slower responses (2-5s) but better code

Note: Run /init to build index if not already built.

Session Management

/status - Show Session Status

Display current session information.

Usage: 

/status

Shows: - Connected models - Active MCP servers - Last generated code - Project configuration

/history - View Command History

View your command history in the current session.

Usage: 

/history

/clear - Clear Screen

Clear the terminal screen.

Usage: 

/clear

Help & Information

/help - Show Help

Display help information for all commands.

Usage: 

/help [command]

Examples: 

/help              # Show all commands
/help optimize     # Show help for /optimize

/intro - Introduction Guide

Show comprehensive introduction to DSPy Code.

Usage: 

/intro

Includes: - What is DSPy Code - Key features - Getting started guide - Tips and best practices

/version - Show Version

Display DSPy Code and DSPy versions.

Usage: 

/version

Exit

/exit - Exit Interactive Mode

Exit the DSPy Code interactive session.

Usage: 

/exit

Note: Your session state is automatically saved.

Natural Language

You can also use natural language instead of slash commands:

Examples: 

Build a sentiment analyzer module
Create a signature for email classification
Generate 20 training examples for sentiment analysis
Optimize my_program.py with training_data.jsonl

DSPy Code will understand your intent and execute the appropriate action.

Command Aliases

Some commands have shorter aliases:

  • /h → /help
  • /q → /exit
  • /v → /validate
  • /s → /save

Tips

  1. Tab Completion: Use tab to autocomplete command names
  2. Command History: Use up/down arrows to navigate history
  3. Natural Language: Describe what you want instead of remembering commands
  4. Help: Use /help anytime to see available commands
  5. Status: Use /status to see what's currently loaded

For more details, see the User Guide