@bryel/sdk-core

Framework-agnostic core for bryel — LLM & agent tracing. It's an OpenTelemetry OTLP exporter + span processor, so you can ship traces to bryel from any OpenTelemetry-instrumented app. On the Vercel AI SDK? Use @bryel/vercel instead.

npm i @bryel/sdk-core @opentelemetry/sdk-trace-base

Any OpenTelemetry / OpenInference app

If your app already emits OpenInference spans (e.g. via an OpenInference instrumentor for OpenAI, Anthropic, or LangChain — or your own instrumentation), add the bryel processor:

import { bryelSpanProcessor } from "@bryel/sdk-core";

// with @vercel/otel
registerOTel({ spanProcessors: [bryelSpanProcessor({ apiKey: process.env.BRYEL_KEY })] });

// or a NodeSDK
new NodeSDK({ spanProcessors: [bryelSpanProcessor({ apiKey: process.env.BRYEL_KEY })] }).start();

That's it — spans batch and POST to the bryel ingest with Authorization: Bearer <key>. The receiver stamps tenant/project from the key; the SDK never sends them.

Custom mapping

If your spans aren't OpenInference yet, supply a SpanMapper (this is exactly what @bryel/vercel does for the AI SDK):

import { buildSpanProcessor, type SpanMapper } from "@bryel/sdk-core";
const processor = buildSpanProcessor(myMapper, { apiKey: process.env.BRYEL_KEY });

Options

{ apiKey, endpoint?, headers? } — endpoint defaults to the bryel SaaS ingest (https://ingest.bryel.ai/v1/traces); point it at your own ingest for self-hosting.

MIT © bryel