SynthLang 🚀

The Generative AI Pipeline DSL - Compose, evaluate, and deploy LLM pipelines with confidence.

pipeline CustomerSupport {
    // Multi-model routing with A/B testing
    router intent_classifier {
        strategy: ab_split(0.5)
        routes: [
            {name: "gpt4", target: gpt4_model},
            {name: "claude", target: claude_model}
        ]
    }
    
    // Built-in safety & compliance
    guardrail safety {
        toxicity_threshold: 0.1
        pii_detection: true
        bias_check: ["gender", "race"]
    }
    
    // Smart caching & evaluation
    cache responses {
        ttl: 3600
        strategy: semantic_similarity(0.95)
    }
}

🎯 Why SynthLang?

• 🔄 Multi-Model Orchestration: Route between models, A/B test, and optimize costs automatically
• 🛡️ Built-in Safety: Toxicity detection, bias testing, and PII protection out of the box
• 📊 Comprehensive Evaluation: Dataset versioning, statistical testing, and performance metrics
• ⚡ Production Ready: Caching, rate limiting, and monitoring built into the language
• 🎨 Developer Experience: VS Code extension with IntelliSense, debugging, and live metrics

🚀 Quick Start

Install the VS Code Extension (Free)

1. Open VS Code
2. Search for "SynthLang" in Extensions
3. Click Install

Install the CLI

# NPM
npm install -g synthlang

# Or using Cargo
cargo install synthlang

# Or download binary
curl -sSL https://get.synthlang.ai | sh

Your First Pipeline

Create hello.synth:

pipeline HelloWorld {
    prompt greeting {
        template: """
        Generate a friendly greeting for {{name}}.
        Make it warm and welcoming!
        """
    }
    
    model gpt {
        provider: "openai"
        model: "gpt-3.5-turbo"
        temperature: 0.7
    }
    
    edges: [
        input -> greeting -> gpt -> output
    ]
}

// Run with: synth run hello.synth --input '{"name": "Alice"}'

📦 Features

Free Tier

• ✅ VS Code Extension with syntax highlighting
• ✅ Local pipeline execution
• ✅ Basic evaluation metrics
• ✅ Community support

Pro Tier ($49/mo)

• ✅ Everything in Free
• ✅ Advanced caching strategies
• ✅ A/B testing & routing
• ✅ Bias & toxicity detection
• ✅ Dataset versioning
• ✅ Priority support

Team Tier ($199/mo)

• ✅ Everything in Pro
• ✅ Team collaboration
• ✅ Audit trails
• ✅ Custom guardrails
• ✅ SLA guarantees
• ✅ Dedicated support

Enterprise (Custom)

• ✅ Everything in Team
• ✅ On-premise deployment
• ✅ Custom integrations
• ✅ Compliance reporting
• ✅ 24/7 support

🏗️ Architecture

┌─────────────────────────────────────────┐
│           SynthLang Pipeline            │
├─────────────────────────────────────────┤
│  DSL Parser → Graph Builder → Executor  │
├─────────────────────────────────────────┤
│  Safety Layer (Toxicity, Bias, PII)     │
├─────────────────────────────────────────┤
│  Caching & Optimization Layer           │
├─────────────────────────────────────────┤
│  Model Providers (OpenAI, Anthropic...) │
└─────────────────────────────────────────┘

📊 Evaluation & Testing

eval CustomerSupportQuality {
    dataset: "support_test_v1"  // Versioned dataset
    
    metrics: {
        accuracy: true,
        toxicity: true,
        bias: true,
        latency_p95: 2000  // ms
    }
    
    comparison: {
        baseline: "gpt-3.5",
        candidates: ["CustomerSupport"],
        significance: 0.95
    }
}

🛡️ Safety First

Every pipeline automatically includes:

• Toxicity Detection: Perspective API integration
• Bias Testing: Multi-dimensional bias detection
• PII Protection: Automatic redaction
• Adversarial Testing: Jailbreak prevention
• Audit Logging: Complete traceability

🔧 Advanced Features

Multi-Stage Pipelines

pipeline MultiStage {
    // Stage 1: Classification
    model classifier { ... }
    
    // Stage 2: Specialized routing
    router by_intent {
        strategy: conditional
        routes: [
            {condition: "intent == 'technical'", target: tech_expert},
            {condition: "intent == 'billing'", target: billing_expert}
        ]
    }
}

Dataset Versioning

dataset CustomerQueries {
    version: "2.1.0"
    parent: "CustomerQueries@2.0.0"
    
    transformations: [
        {type: "filter", condition: "quality_score > 0.8"},
        {type: "augment", method: "paraphrase"}
    ]
}

A/B Testing

router experiment {
    strategy: ab_split(0.5)
    
    metrics: [
        "response_quality",
        "latency",
        "cost"
    ]
    
    auto_optimize: true  // Automatically shift traffic
}

📈 Metrics & Monitoring

Real-time metrics dashboard in VS Code:

• Request volume & latency
• Cost tracking
• Cache hit rates
• Error rates
• A/B test results

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

# Clone the repo
git clone https://github.com/synth-lang/synth
cd synth

# Install dependencies
cargo build --release

# Run tests
cargo test

# Run examples
synth run examples/

📚 Documentation

• Language Guide
• API Reference
• Examples
• Blog

🗺️ Roadmap

Phase 0 (Current) - Free Extension

• ✅ VS Code extension
• ✅ Basic pipeline runner
• ✅ Evaluation harness

Phase 1 (Q1 2024) - Pro Launch

• 🔄 Cloud execution
• 🔄 Advanced caching
• 🔄 Team features

Phase 2 (Q2 2024) - Enterprise

• 📋 Fine-tuning management
• 📋 Compliance suite
• 📋 Custom deployments

Phase 3 (Q3 2024) - Platform

• 📋 Model marketplace
• 📋 Component library
• 📋 Community hub

💬 Community

• Discord - Join our community
• Twitter - Follow for updates
• GitHub Discussions - Ask questions

📄 License

Licensed under Apache 2.0. See LICENSE for details.

🙏 Acknowledgments

Built with ❤️ by the SynthLang team.

Special thanks to our contributors and early adopters!

---

Ready to build safer, more reliable AI pipelines?

Get Started → | View Examples → | Join Discord →