Command Line Interface (CLI)

The Prism CLI provides a complete environment for developing and running Prism programs.

Installation

# Install globally with npm
npm install -g @prism-lang/cli

# Or with yarn
yarn global add @prism-lang/cli

# Or with pnpm
pnpm add -g @prism-lang/cli

Commands

prism run [--watch] <file>

Execute a Prism program from a file.

# Run a Prism file
prism run script.prism

# Enable hot reload (invalidates the changed module + dependents)
prism run --watch script.prism

# Example output
🚀 Running script.prism...

Result: Hello from Prism! (~0.85)

Features:

• Executes .prism files
• Optional --watch flag for hot reload (uses the runtime module reloader so imports are refreshed in-place)
• Shows confidence values in output
• Supports all Prism language features
• Configures LLM providers from environment

prism eval <code>

Evaluate Prism expressions directly from the command line.

# Simple expression
prism eval "2 + 2"
# Output: 4

# With confidence
prism eval "42 ~> 0.9"
# Output: 42 (~0.9)

# Multi-word expressions
prism eval "x = 10; x * 2"
# Output: 20

# LLM calls
prism eval 'llm("Hello AI!")'
# Output: Hello! How can I help you today? (~0.85)

prism llm [options] <prompt>

Send a one-off LLM prompt directly from the terminal. Add --stream to see tokens as they arrive.

# Simple completion
prism llm "Generate three product names"

# Stream output with provider override
prism llm --provider claude --stream "Summarize today's status update"

Useful flags:

• --provider, -p <name> – choose a configured provider (default environment selection)
• --stream / --no-stream – toggle live streaming (Ctrl+C to cancel)
• --temperature <value> – adjust sampling temperature
• --maxTokens <value> / --topP <value> – advanced sampling controls
• --model <id> – override the model for this invocation
• --timeout <ms> – abort the request after the given time budget
• --include-reasoning / --no-include-reasoning – ask providers to attach reasoning traces
• --structured-output / --no-structured-output – force JSON-style output or plain text (streaming requires --no-structured-output)

prism or prism repl

Launch the interactive REPL (Read-Eval-Print Loop).

prism
# Welcome to Prism REPL v1.0.23
# Type expressions and see results with confidence tracking
# Commands: :help, :vars, :clear, :exit
# 
# prism> 

REPL Features:

• Interactive expression evaluation
• Variable persistence across commands
• Command history
• Special commands (:help, :vars, etc.)
• Multi-line input support

prism --version or prism -v

Display the installed version.

prism --version
# Prism v1.0.23

prism --help or prism -h

Show help information and available commands.

Configuration

LLM Provider Setup

The CLI automatically configures LLM providers from environment variables:

# Claude (Anthropic)
export CLAUDE_API_KEY=your-api-key

# Google Gemini
export GEMINI_API_KEY=your-api-key

# Default provider (optional)
export PRISM_DEFAULT_LLM=claude  # or 'gemini'

Environment Files

The CLI loads .env files automatically if present:

# .env file in your project
CLAUDE_API_KEY=sk-ant-...
GEMINI_API_KEY=AIza...
PRISM_DEFAULT_LLM=claude

Examples

Basic Script

// analyze.prism
let data = "User feedback: The product is amazing but shipping was slow"

let sentiment = llm("Analyze sentiment: ${data}") ~> 0.9
let entities = llm("Extract entities: ${data}") ~> 0.85

console.log("Sentiment:", sentiment)
console.log("Entities:", entities)

uncertain if (sentiment) {
  high { 
    console.log("✅ Positive feedback detected")
  }
  medium { 
    console.log("⚠️ Mixed feedback detected")
  }
  low { 
    console.log("❌ Negative feedback detected")
  }
}

Run with: prism run analyze.prism

Interactive Development

$ prism
Welcome to Prism REPL v1.0.23

prism> x = 10 ~> 0.95
10 (~0.95) (confident)

prism> y = 20 ~> 0.90
20 (~0.9) (confident)

prism> x + y
30 (~0.9) (confident)

prism> :vars
Session Variables:
  x: 10 (~95.0%)
  y: 20 (~90.0%)

prism> analysis = llm("What is ${x} + ${y}?")
The sum of 10 + 20 equals 30. (confident)

prism> :exit
Goodbye! 👋

Pipeline Processing

# Process data through multiple transformations
prism eval 'data = "raw text" |> clean |> analyze |> summarize'

# With confidence tracking
prism eval 'result = input ~> 0.8 |> process ~> 0.9 |> validate ~> 0.95'

Error Handling

The CLI provides clear error messages:

# Syntax errors
prism eval "x = "
❌ Error: Expected expression after '='

# Runtime errors
prism run nonexistent.prism
❌ Error: File not found: nonexistent.prism

# LLM configuration
prism eval 'llm("test")'
⚠️  Only mock LLM provider available. Set CLAUDE_API_KEY or GEMINI_API_KEY for real AI integration.

Tips

1. Use the REPL for experimentation - Test expressions interactively before adding to scripts
2. Check confidence values - The CLI always shows confidence in output
3. Set up LLM providers - Export API keys for real AI capabilities
4. Use .env files - Keep API keys secure and out of scripts
5. Leverage multi-line input - The CLI handles complex expressions naturally

Troubleshooting

LLM Provider Not Working

# Check available providers
prism eval ":llm"

# Verify environment variables
echo $CLAUDE_API_KEY
echo $GEMINI_API_KEY

Command Not Found

# Verify installation
npm list -g @prism-lang/cli

# Reinstall if needed
npm install -g @prism-lang/cli

File Encoding Issues

Ensure .prism files are saved with UTF-8 encoding for proper string handling.

Watch Mode

Use --watch (or -w) to keep the same runtime alive while editing files. When a watched file changes, Prism invalidates that module, re-runs it (plus any dependents), and logs the results without restarting the CLI. This is ideal for multi-module projects or long-lived runtimes that keep LLM providers in memory.