Skip to main content

Your First Prism Program

Let's write a simple program that demonstrates Prism's unique features: confidence values and uncertainty-aware control flow.

Hello, Uncertain World!​

Create a file called hello.prism:

// Basic confidence assignment
greeting = "Hello, World!" ~> 0.95
console.log(greeting)

// Working with uncertainty
temperature = 72.5 ~> 0.9
if (temperature > 70) {
  console.log("It's warm today!")
}

Running Your Program​

Using the CLI​

prism run hello.prism

Using TypeScript/JavaScript​

import { parse, createRuntime } from '@prism-lang/core';
import { readFileSync } from 'fs';

const code = readFileSync('hello.prism', 'utf-8');
const ast = parse(code);
const runtime = createRuntime();

const result = await runtime.execute(ast);
console.log(result);

A More Complex Example​

Let's build a simple AI-powered decision maker:

// AI Analysis with Confidence
analysis = llm("Should we deploy this code to production?")
confidence = <~ analysis  // Extract confidence from LLM response

// Multi-level decision making
uncertain if (analysis ~> 0.8) {
  high {
    console.log("✅ High confidence - deploying to production")
    deploy_status = "approved"
  }
  medium {
    console.log("⚠️ Medium confidence - requesting human review")
    deploy_status = "review_required"
  }
  low {
    console.log("❌ Low confidence - deployment blocked")
    deploy_status = "blocked"
  }
}

// Confidence combination
security_check = llm("Any security vulnerabilities?") ~> 0.85
performance_check = llm("Performance impact acceptable?") ~> 0.9

// Parallel confidence selection (picks highest confidence)
final_decision = security_check ~||> performance_check

console.log("Final deployment confidence: " + (<~ final_decision))

Understanding the Output​

When you run Prism programs, confidence values are displayed alongside results:

"Hello, World!" (95.0% confidence)
It's warm today!
⚠️ Medium confidence - requesting human review
Final deployment confidence: 90.0%

Key Concepts Demonstrated​

  1. Confidence Assignment (~>): Attach confidence to any value
  2. Confidence Extraction (<~): Get the confidence level of a value
  3. Uncertain Control Flow: Make decisions based on confidence thresholds
  4. Parallel Selection (~||>): Choose the option with highest confidence
  5. LLM Integration: Native support for AI/LLM calls

Exercises​

Try modifying the program to:

  1. Add a third check for code quality
  2. Use different confidence thresholds
  3. Implement a confidence budget (total confidence must exceed a threshold)
  4. Create a weighted average of multiple confidence values

Next Steps​

Now that you've written your first Prism program, explore: