Core Concepts - Superwire

A Superwire workflow is a dependency graph built from declarations.

input {
    message: string
}

secrets {
    api_key: string
}

provider llm from openai {
    endpoint: "https://api.openai.com/v1"
    api_key: secrets.api_key
}

model fast from llm {
    id: "gpt-4.1-mini"
}

agent reply {
    model: model.fast
    instruction: "Reply to {{ input.message }}"
    output {
        message: string
    }
}

output {
    result: agent.reply
}

Main declarations

Declaration	Purpose
`input`	Public values supplied by the executor request.
`secrets`	Sensitive values supplied by the executor request.
`provider`	Configures access to an OpenAI-compatible backend through `openai`.
`model`	Defines a reusable model profile from a provider instance.
`schema`	Defines reusable structured types.
`mcp`	Defines an MCP server endpoint.
`tool`, `prompt`, `resource`	Import external MCP capabilities.
`dynamic`	Builds runtime values from expressions.
`agent`	Runs one model step.
`output`	Defines the final JSON response.

Runtime topology

provider instance -> model profile -> agent

A provider declaration does not select a model by itself. A model profile selects the provider-specific model ID. Agents reference named model profiles.

Design goals

Keep secrets outside .wire files.
Keep provider access separate from model selection.
Make dependencies explicit through references.
Require structured outputs so downstream steps can rely on stable shapes.
Let the executor schedule independent work concurrently.