Superwire

A declarative DSL for building AI-powered workflows with type safety, composability, and runtime validation.

Overview

Superwire provides a domain-specific language for defining AI agent workflows that can be parsed, validated, and executed programmatically. Workflows define providers, schemas, inputs, agents, and outputs in a single .wire file.

Quick Start

Run with Docker

docker run --rm -p 13703:13703 rmilewski/superwire:latest

Open http://localhost:13703/playground to access the Playground UI.

Create a Workflow

Paste the following into the Playground editor and replace the provider values with your own:

provider openai from openai {
    endpoint: "https://api.openai.com/v1"
    api_key: "your-api-key"
}

model openai_model from openai {
    id: "gpt-4"
}

agent greeting {
    model: model.openai_model
    instruction: "Write a short welcome message."
    output {
        message: string
    }
}

output {
    greeting: agent.greeting.message
}

Core Concepts

Superwire workflows structure and orchestrate the execution of LLM agents. Define your agents, their models, and the data flow between them in a declarative .wire file—Superwire handles parsing, validation, and runtime execution.

Advantages:

• Type safety — Define inputs, outputs, and schemas with compile-time validation
• Composability — Reuse providers, models, and agents across workflows
• Runtime validation — Catch errors before they reach production
• Portable — Single .wire file contains everything needed to execute a workflow

For full documentation, visit https://docs.superwire.dev.

License

This project is licensed under the MIT License - see the LICENSE file for details.