@agentmark-ai/otel

Group AgentMark traces by session, user, tags, and custom metadata using OpenTelemetry context.

It pairs a small wrapper with a span processor:

• withAgentMark(grouping, fn) — stashes grouping in the active OpenTelemetry context.
• AgentMarkSpanProcessor — stamps that grouping onto every span started in the scope (using the attribute keys AgentMark reads) and batches them to AgentMark's OTLP endpoint.

It is framework-agnostic: any OpenTelemetry span created in the wrapped scope — Vercel AI SDK calls, HTTP requests, DB queries, custom spans — is grouped. It is especially useful with the Vercel AI SDK v7, where telemetry.metadata no longer reaches spans, so session/user/metadata grouping can't ride it.

Install

npm install @agentmark-ai/otel @opentelemetry/sdk-trace-node

Usage

import { NodeTracerProvider } from '@opentelemetry/sdk-trace-node';
import { AgentMarkSpanProcessor, withAgentMark } from '@agentmark-ai/otel';

// Once, at startup:
const provider = new NodeTracerProvider({
  spanProcessors: [
    new AgentMarkSpanProcessor({
      apiKey: process.env.AGENTMARK_API_KEY!, // raw key, no "Bearer" prefix
      appId: process.env.AGENTMARK_APP_ID!,
    }),
  ],
});
provider.register();

// Per request — wrap your work:
await withAgentMark(
  { sessionId, userId, tags: ['prod'], metadata: { feature: 'chat' } },
  () => generateText({ model, prompt }), // or any traced work
);

withAgentMark requires an OpenTelemetry context manager to be registered (both NodeTracerProvider#register() and @vercel/otel do this) so the grouping propagates across await.

License

AGPL-3.0-or-later