Model APIs and requests

Methods to understand

Keep API keys on the server and define the model, input, timeout, and output limits. The client sends only task-relevant material. Implement a minimal request first, then add error handling.

Apply the method to a scenario

Goal: complete a clearly scoped work task using AI.
Input: public, task-relevant material.
Acceptance: verifiable results and explicitly marked unknowns.

Hands-on workshop

Make a real server-side model request.

1. Bind AI in your own Worker configuration.
2. Call the model through server-side env.AI; keep keys out of the browser.
3. Check success, invalid input, and service failure and record actual responses.

const result = await env.AI.run(
  "@cf/meta/llama-3.3-70b-instruct-fp8-fast",
  { messages: [
    { role: "system", content: "Answer only from supplied material; state uncertainty when evidence is missing." },
    { role: "user", content: "Material: the course contains one retrieval evaluation task. Question: what task does it contain?" }
  ], max_tokens: 300 }
);
if (!result.response) throw new Error("Empty model result");
return Response.json({ answer: result.response });

How to verify

The output must answer the question from the supplied material. Also test material without an answer and check whether the model admits uncertainty.

Practice task

Choose a case related to your work and practice "Model APIs and requests". Submit inputs, steps, results, and verification records.

• Explain the task goal and inputs
• Provide actual results or runnable deliverables
• Identify errors, limits, and verification evidence