Workflows - Superwire

A workflow file contains declarations. The executor validates the declarations, builds a dependency graph, runs the graph, and returns the output block as JSON.

Common declaration order

A readable workflow usually uses this order:

input
secrets
provider
model
schema
mcp and imports
dynamic
agent
output

The order is for readability. Dependencies are determined by references.

Small workflow

input {
    topic: 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 summarize {
    model: model.fast
    instruction: "Summarize {{ input.topic }} in one paragraph."

    output {
        summary: string
    }
}

output {
    summary: agent.summarize.summary
}

Final output

The final output block is an object expression. It determines the JSON shape returned by /execute.

output {
    summary: agent.summarize.summary
    raw: agent.summarize
}