Skip to content

Natural Language Commands

DSPy Code supports natural language for all commands! You don't need to remember slash commands - just describe what you want to do.

đŸŽ¯ How It Works

Instead of remembering exact slash commands, you can either:

  • Use the interactive selector: /model
  • Or just describe what you want in natural language.

For example, instead of typing /connect ollama gpt-oss:120b, you can say:

connect to ollama gpt-oss:120b

DSPy Code automatically understands your intent and routes it to the appropriate command (for example, /model or /connect).

📋 Supported Natural Language Commands

Connection Commands

Instead of: /model (interactive selector)
You can say: - "help me pick a model" - "connect a model for me" - "set up a model" - "select a model to use"

Instead of: /connect ollama gpt-oss:120b
You can say: - "connect to ollama gpt-oss:120b" - "use model ollama gpt-oss:120b" - "switch to ollama gpt-oss:120b" - "set up model ollama gpt-oss:120b" - "configure model ollama gpt-oss:120b"

Instead of: /disconnect
You can say: - "disconnect from model" - "stop using model" - "close connection"

Instead of: /models
You can say: - "list available models" - "show me all models" - "what models are available"

Instead of: /status
You can say: - "status" - "what is the current status" - "show me the status" - "where am I"

File Operations

Instead of: /save my_program.py
You can say: - "save the code as my_program.py" - "save this as my_program.py" - "write to file my_program.py" - "store the code as my_program.py"

Instead of: /save-data training.jsonl
You can say: - "save the data as training.jsonl" - "save my examples as training.jsonl" - "export data to training.jsonl"

Validation and Execution

Instead of: /validate
You can say: - "validate the code" - "check the code for errors" - "verify the program" - "test the code syntax" - "is the code valid"

Instead of: /run
You can say: - "run the code" - "execute the program" - "test the execution" - "start the program"

Instead of: /test
You can say: - "test the code" - "run tests" - "execute tests"

Project Management

Instead of: /init
You can say: - "initialize the project" - "setup project" - "create new project" - "start new project" - "new project"

Instead of: /project
You can say: - "project info" - "show me the project" - "what is the project"

Optimization

Instead of: /optimize my_program.py
You can say: - "optimize the code" - "optimize my program" - "optimize my_program.py" - "improve the program" - "improve performance" - "use gepa" - "run gepa" - "run optimization" - "run GEPA optimization" - "make it better" - "enhance the code" - "optimize with GEPA" - "optimize my_module.py with training_data.jsonl"

Instead of: /optimize my_program.py training_data.jsonl
You can say: - "optimize my_program.py with training_data.jsonl" - "run GEPA on my_program.py using training_data.jsonl" - "optimize the program with training data" - "improve my code with examples from training_data.jsonl"

Instead of: /optimize-status
You can say: - "gepa status" - "optimization status" - "how is optimization going" - "check optimization progress" - "what's the optimization status"

Instead of: /optimize-cancel
You can say: - "cancel optimization" - "stop gepa" - "abort optimization" - "stop the optimization"

Evaluation

Instead of: /eval
You can say: - "evaluate the code" - "evaluate my program" - "evaluate the program" - "run evaluation" - "test performance" - "test the code" - "measure performance" - "assess performance" - "check performance"

Instead of: /eval my_program.py
You can say: - "evaluate my_program.py" - "test performance of my_program.py" - "run evaluation on my_program.py" - "evaluate the program in my_program.py"

Instead of: /eval my_program.py test_data.jsonl
You can say: - "evaluate my_program.py with test_data.jsonl" - "test my program using test_data.jsonl" - "run evaluation on my_program.py with test_data.jsonl" - "measure performance with test_data.jsonl"

Instead of: /eval metric=accuracy
You can say: - "evaluate with accuracy" - "test accuracy" - "measure accuracy" - "calculate accuracy" - "check accuracy" - "evaluate using accuracy metric" - "test performance with accuracy"

Instead of: /eval my_program.py test_data.jsonl metric=f1
You can say: - "evaluate my_program.py with test_data.jsonl using F1 score" - "test F1 score of my program with test data" - "measure F1 metric on my_program.py with test_data.jsonl" - "calculate F1 score for my program"

Data Generation

Instead of: /data sentiment 20
You can say: - "generate 20 examples for sentiment analysis" - "create training data for sentiment" - "make 20 examples for sentiment" - "produce data for sentiment analysis"

Explanation and Help

Instead of: /explain ChainOfThought
You can say: - "explain ChainOfThought" - "what is ChainOfThought" - "how does ChainOfThought work" - "tell me about ChainOfThought" - "describe ChainOfThought"

Instead of: /help
You can say: - "help" - "what can you do" - "show me help" - "list commands" - "what commands are available"

Instead of: /intro
You can say: - "introduction" - "guide" - "tutorial" - "getting started" - "how do I start"

History and Context

Instead of: /history
You can say: - "history" - "show me the conversation history" - "what did we talk about" - "previous messages"

Instead of: /clear
You can say: - "clear context" - "reset history" - "forget everything" - "start over" - "new conversation"

Sessions

Instead of: /sessions
You can say: - "list sessions" - "show me all sessions" - "what sessions do I have"

Instead of: /session save my-session
You can say: - "save session as my-session" - "save conversation as my-session"

Instead of: /session load my-session
You can say: - "load session my-session" - "restore session my-session" - "open session my-session"

Export/Import

Instead of: /export package.zip
You can say: - "export package to package.zip" - "package the project" - "create package"

Instead of: /import package.zip
You can say: - "import package from package.zip" - "load package package.zip" - "open package package.zip"

RAG and Indexing

Instead of: /refresh-index
You can say: - "refresh index" - "rebuild knowledge base" - "update index" - "reindex"

Instead of: /index-status
You can say: - "index status" - "knowledge base status" - "codebase index"

Feature Listings

Instead of: /predictors
You can say: - "list predictors" - "show me predictor types" - "what predictors are available"

Instead of: /adapters
You can say: - "list adapters" - "show me adapter types" - "what adapters"

Instead of: /retrievers
You can say: - "list retrievers" - "show me retriever types" - "what retrievers"

Instead of: /examples rag
You can say: - "show me examples" - "list templates" - "what examples are available"

MCP Commands

Instead of: /mcp-connect filesystem
You can say: - "connect to mcp server filesystem" - "link to mcp filesystem" - "use mcp server filesystem"

Instead of: /mcp-tools
You can say: - "list mcp tools" - "show me mcp tools" - "what mcp tools are available"

Instead of: /mcp-call filesystem read_file
You can say: - "call mcp tool filesystem read_file" - "use mcp tool read_file" - "invoke mcp tool read_file"

💡 Tips

  1. Be natural - Just describe what you want to do
  2. Include details - Mention filenames, model names, etc.
  3. Mix and match - You can still use slash commands if you prefer
  4. Both work - Natural language and slash commands work together

đŸŽ¯ Examples

Complete Workflow in Natural Language

You: initialize the project
DSPy Code: ✓ Project initialized

You: connect to ollama llama3.1:8b
DSPy Code: ✓ Connected to ollama llama3.1:8b

You: create a sentiment analyzer
DSPy Code: [Generates code...]

You: save the code as sentiment.py
DSPy Code: ✓ Saved to sentiment.py

You: validate the code
DSPy Code: ✓ Code is valid

You: run the code
DSPy Code: [Executes code...]

You: generate 20 examples for sentiment analysis
DSPy Code: [Generates training data...]

You: optimize the code
DSPy Code: [Runs GEPA optimization...]

You: evaluate my program with accuracy
DSPy Code: [Runs evaluation with accuracy metric...]

You: test performance of sentiment.py with test_data.jsonl
DSPy Code: [Evaluates program with test data...]

🔄 How It Works Internally

DSPy Code uses a hybrid intelligent routing system:

Step 1: Fast Pattern Matching (First Attempt)

  • Matches your natural language against known patterns
  • Fast and deterministic
  • Handles common cases instantly

Step 2: LLM Reasoning (Intelligent Fallback)

  • If pattern matching fails, uses your connected LLM (Ollama, OpenAI, etc.)
  • Understands context, variations, and intent
  • Handles edge cases and natural language variations
  • Uses conversation history and current state

Step 3: Argument Extraction

  • Automatically extracts parameters (filenames, model names, etc.)
  • Works with both pattern matching and LLM reasoning

Step 4: Command Execution

  • Routes to the appropriate slash command handler
  • Executes the command with extracted arguments

Example Flow

User: "I want to use the llama model that's running locally"
  ↓
Pattern Matching: ✗ No match
  ↓
LLM Reasoning: "User wants to connect to Ollama model"
  ↓
Extract: provider="ollama", model="llama3.1:8b" (from context)
  ↓
Route to: /connect ollama llama3.1:8b
  ↓
Execute command

Benefits

  • Fast - Common commands are instant (pattern matching)
  • Intelligent - Handles any natural language variation (LLM reasoning)
  • Context-Aware - Understands conversation history and current state
  • Flexible - Works even with unusual phrasings

You can use natural language for everything, or mix it with slash commands - whatever feels more natural to you! 🚀