@virajmishra1/bench-vercel-ai

Auto-instrument the Vercel AI SDK with Bench.

npm install @virajmishra1/bench-vercel-ai @virajmishra1/bench-sdk ai

Use it

Two flavors. Pick whichever matches your style.

A — One-line wrappers

import { anthropic } from "@ai-sdk/anthropic";
import { observe } from "@virajmishra1/bench-sdk";
import { wrappedGenerateText } from "@virajmishra1/bench-vercel-ai";

const bench = observe({ apiKey: process.env.BENCH_KEY, agent: "my-agent" });

const result = await wrappedGenerateText(bench, "summarize", {
  model: anthropic("claude-sonnet-4-6"),
  prompt: "summarize: ...",
});

B — Track any call

import { generateText } from "ai";
import { track } from "@virajmishra1/bench-vercel-ai";

const result = await track(bench, "summarize", () =>
  generateText({ model: anthropic("claude-sonnet-4-6"), prompt: "..." }),
  { input: { prompt: "..." }, pricePerMTok: { input: 3, output: 15 } }
);

What gets recorded

• Task name, input (prompt / messages, truncated)
• Output finish reason
• Token usage from the SDK's result.usage
• Optional cost (pass pricePerMTok)
• Duration, status

License

MIT