@agentegrity/vercel-ai

Zero-config agentegrity adapter for the Vercel AI SDK (ai).

TypeScript-native — there is no Python equivalent.

Install

npm i @agentegrity/vercel-ai ai

Use

import { streamText } from "ai";
import { instrument, report } from "@agentegrity/vercel-ai";

const { textStream } = streamText({
  model: anthropic("claude-sonnet-4-5"),
  prompt: "Summarize the latest papers",
  experimental_telemetry: instrument(),
});

for await (const chunk of textStream) process.stdout.write(chunk);
console.log(await report());

Works with generateText, streamText, generateObject, streamObject, and the SDK's tool-calling path.

How it works

The Vercel AI SDK emits OpenTelemetry spans with well-known names (ai.generateText, ai.streamText, ai.toolCall, etc.). instrument() returns a minimal tracer that maps those spans to agentegrity events:

| AI SDK span | agentegrity start event | agentegrity end event | |---|---|---| | ai.generateText / ai.streamText | user_prompt_submit | stop | | ai.generateObject / ai.streamObject | user_prompt_submit | stop | | ai.toolCall | pre_tool_use | post_tool_use |

The tracer ships no @opentelemetry/api runtime dependency — it implements just enough of the Span/Tracer surface the AI SDK uses.

Pass { catchAll: true } to also emit events for unrecognized span names.

API

| Function | Returns | |---|---| | instrument(options?) | { isEnabled: true, tracer } for experimental_telemetry | | report() | Session summary snapshot | | reset() | Discard the module-global adapter | | registerExporter(exporter) | Subscribe an additional SessionExporter |

License

Apache-2.0