Skip to main content

Vertex AI Generate Content Agent

This integration sends your input text to Google Vertex AI and returns the generated response text.

Required secrets​

SecretWhat it is
GOOGLE_VERTEX_AI_API_KEYAPI key for Vertex AI generate-content requests.

Where to get the credentials​

Provision an API key for Vertex AI in your Google Cloud project, then store it as a project secret named GOOGLE_VERTEX_AI_API_KEY.

Turn support​

Single-turn. Each evaluation case is sent as an independent request.

Limitations and notes​

  • The model is set in the copied agent code. Edit the agent if you need a different one.
  • Calls are billed to your own Google Cloud project, subject to its quotas and rate limits.

How the agent works​

The copied agent is a single JavaScript function. This walkthrough shows the pattern so you can adapt it for a custom agent of your own.

The entry point is process, which receives input (the evaluation case text) and returns the result:

export async function process(input) {
 try {
  const apiKey = String(env.GOOGLE_VERTEX_AI_API_KEY ?? "").trim();
  const { text } = parseInput(input);
  // ...

Secrets are read from the env object by name. Here the Vertex AI API key is read from env.GOOGLE_VERTEX_AI_API_KEY (the secret listed in Required secrets above).

Input is normalized so it accepts either plain text or a JSON object. A JSON object can carry the prompt under text, prompt, or input; anything that is not parseable JSON is used verbatim:

const parseInput = (input) => {
 const raw = String(input ?? "").trim();
 if (!raw) {
  return { text: "" };
 }

 try {
  const parsed = JSON.parse(raw);
  if (typeof parsed === "string") {
   return { text: parsed.trim() };
  }
  if (parsed && typeof parsed === "object" && !Array.isArray(parsed)) {
   return {
    text: String(parsed.text ?? parsed.prompt ?? parsed.input ?? "").trim(),
   };
  }
 } catch {}

 return { text: raw };
};

The request is a POST with the API key passed as a query parameter, a JSON content type, and a contents array carrying the user text:

const response = await fetch(
 `${ENDPOINT}?key=${encodeURIComponent(apiKey)}`,
 {
  method: "POST",
  headers: {
   "content-type": "application/json",
  },
  body: JSON.stringify({
   contents: [
    {
     role: "user",
     parts: [{ text }],
    },
   ],
  }),
 },
);

The response body is read as text, then parsed as JSON. The assistant text is pulled from the first candidate that has non-empty content.parts text:

const extractResponseText = (payload) => {
 const candidates = Array.isArray(payload?.candidates)
  ? payload.candidates
  : [];

 for (const candidate of candidates) {
  const parts = Array.isArray(candidate?.content?.parts)
   ? candidate.content.parts
   : [];
  const text = parts
   .map(extractTextFromPart)
   .filter(Boolean)
   .join("\n")
   .trim();
  if (text) {
   return text;
  }
 }

 return "";
};

Errors are returned, not thrown. The function returns a { ok: false, error } object for missing input, failed requests, and empty responses, and a { ok: true, text } object on success:

return {
 ok: true,
 text: outputText,
 model_version:
  typeof payload?.modelVersion === "string" ? payload.modelVersion : null,
};

Adapting it for your own agent​

To point the same skeleton at a different service, change:

  • the ENDPOINT URL,
  • how the request authenticates (here the key is a query parameter; many APIs use an authorization header instead),
  • the request body shape your service expects,
  • the response field you extract the text from in extractResponseText, and
  • the secret key you read from env (and declare it in the required-secrets table).

Set it up​

  1. Copy this integration from the registry — see Registry integrations.
  2. Create the project secret above — see Project secrets.
  3. Preview the agent to confirm it works.