Code Node

What the Code node does

The code node lets you run small pieces of JavaScript logic inside a blueprint. It is ideal for:

• Input validation and shaping
• Lightweight data processing between HTTP calls
• Branch-specific computations that are easier in code than templates

The node takes the current input as ctx.input, executes your function, and returns whatever main(ctx) resolves to as its output.

Required entrypoint: async function main(ctx)

Every Code node must define an async entrypoint:

async function main(ctx) {
  // ctx.input: unwrapped input value
  // ctx.state: shared execution state object
  // ctx.outputs: array of { nodeRef, items } results so far
  // ctx.lastOutputOf(ref): convenience helper to read another node's output
  // ctx.exists(value): true if value is not null/undefined
  // ctx.env(key): non-secret runtime config injected by the host

  return { ok: true };
}

If the code does not contain async function main(ctx), validation will fail with:

Missing required entrypoint: async function main(ctx)

At runtime, the engine wraps your code and calls main(ctx). Any thrown error will fail the node.

URL building pattern for http_request

Because templates are full-string only, Code nodes are the preferred place to build complex URLs:

1. Use a Code node to build a URL object:

async function main(ctx) {
  const input = ctx.input || {};
  const q = input.q || 'hello';

  return {
    url: `https://example.com/search?q=${encodeURIComponent(q)}`,
  };
}

2. In a downstream http_request node, reference that value with a single full-string template:

{
  "type": "http_request",
  "parameters": {
    "method": "GET",
    "url": "{{ (lastOutputOf('build_url') || {}).url }}",
    "timeout": 15000
  }
}

This avoids partial templates like "/user/repos?type={{ ... }}&sort={{ ... }}", which are rejected by the engine.

No direct HTTP or connector access

Code nodes cannot call connectors or the HostRuntime directly (there is no ctx.hostRuntime.http). To talk to external services:

• Build request data (URL, headers, body) in a Code node, then
• Use an http_request node to actually perform the call (absolute URLs only), or a connector_action node for connector-backed calls.

This keeps secrets and network configuration in the host, not inside user-authored code.