Vercel AI SDK

Status: planned — the Vercel AI SDK is TypeScript, so it is out of scope for the Python radicas package on this page. Support arrives with the TypeScript SDK (@radicas/otel exists as a stub in radicas-integrations; the design roadmap targets the TS port after the Vercel AI SDK deep-dive).

What is known today (from the bench recipe)

• The AI SDK has first-class OTel: experimental_telemetry: { isEnabled: true } emits both gen_ai.* and ai.* attributes natively.
• Expected shim: light — the ai.* extras are noise and the span naming/tree differs from invoke_agent / generate_content / execute_tool; exact maps must come from a captured fixture.
• The same three-layer normalization applies once its registry card exists: the card will drive the collector OTTL and the (TS) client-side adapter alike.

What to do meanwhile

• Nothing Python-side — radicas does not detect or instrument TypeScript processes.
• You can already point the AI SDK at Radicas manually over plain OTLP env config (endpoint + Bearer header, as in the configuration OTEL_* table) using any OTel NodeSDK setup; spans will land raw and be caught by the server-side layers as mapping coverage grows.

// bench recipe sketch (providers/vercel-ai-sdk/typescript in radicas-integrations)
import { generateText } from "ai";
import { openai } from "@ai-sdk/openai";

await generateText({
  model: openai("gpt-4o-mini"),
  prompt: "...",
  experimental_telemetry: { isEnabled: true, functionId: "pricing_explanation" },
});

Track progress: the radicas-integrations repo (providers/vercel-ai-sdk/typescript) and this page will flip to experimental when the TS SDK lands.