Add live web search to a Dify workflow

Dify is an open-source LLM app platform where you build Workflows and Chatflows by wiring nodes on a canvas — no framework code required. To give a Dify app live web results you have two paths: drop in an HTTP Request node that calls the Superhighway REST API with a free key (simplest), or import Superhighway's OpenAPI spec as a Custom Tool and hand all five tools to an Agent node. Dify's variable syntax is {{variable_name}} throughout.

Two paths

Get a free API key

Create a free-tier key — no credit card — from Settings → API Keys → Generate, or with one request:

curl -X POST https://superhighway.walls.sh/auth/guest-key

The response contains a key you pass as a bearer token on every request.

Path 1 — HTTP Request node (REST API + free key)

In a Dify Workflow or Chatflow, add an HTTP Request node. Configure it like this:

Here {{query}} is the input variable carrying the user's question (the Start node's input, or any upstream variable). Replace YOUR_FREE_KEY with the key from the guest-key call above.

The HTTP Request node exposes the full JSON response body as {{http_request.body}}. The body is a JSON object whose results array holds objects with title, url, and description. You can hand the raw body straight to an LLM, or parse it first in a Code node.

Pass results to an LLM node. Add an LLM node after the HTTP Request node and feed the body in as context in the system prompt:

Here are web search results:
{{http_request.body}}

Answer the user's question using only these results. Cite the URLs.

That's the whole loop: Start → HTTP Request → LLM → End. The model reads the live results and answers. If you'd rather give the LLM clean text instead of raw JSON, drop a Code node between them and parse the body:

def main(body: str) -> dict:
    import json
    data = json.loads(body)
    lines = []
    for r in data.get("results", []):
        lines.append(f"{r['title']}\n{r['url']}\n{r.get('description','')}")
    return {"context": "\n\n".join(lines)}

Wire {{http_request.body}} into the Code node's input and reference {{code.context}} in the LLM node's prompt.

Same pattern, other endpoints. Swap the URL to /news, /images, /scrape, or /research — they take the same bearer header and return JSON in the same shape, so any of them drops into an HTTP Request node the same way.

Path 2 — Custom Tool from the OpenAPI spec (Agent node, all five tools)

Dify lets you add Custom Tools from an OpenAPI 3.0 spec, and Superhighway publishes a live one at https://superhighway.walls.sh/openapi.json. Importing it registers all five endpoints as tools an Agent can call on its own.

Go to Settings → Tools → Custom → Create Custom Tool. In the Schema field, either paste the OpenAPI URL https://superhighway.walls.sh/openapi.json to import it, or paste the JSON directly. Dify parses it and lists the available actions — search, news, images, scrape, and research.

Under Authorization, set the auth method to API Key / Bearer and enter your free key as the token (header Authorization: Bearer YOUR_FREE_KEY). Save the tool.

Now add an Agent node to a Workflow (or use Agent mode in a Chatflow), pick a model, and under Tools add the Superhighway tools you imported. The agent decides when to call them — ask a time-sensitive question and it invokes search or research, reads the live results, and answers. You can export the finished workflow as a YAML DSL file to share or version it.

x402 pay-per-call. Superhighway also supports x402 — paying per call in USDC on Base with no API key. That flow needs a funded wallet to sign payments, which Dify's HTTP layer doesn't do natively; for now use the free-key path above, or run self-hosted Dify with a Code node that holds the wallet and signs requests.

Which tools are available

Full API reference: /openapi.json. Get a free API key or learn the x402 wallet flow.