Introduction

bryel is observability for LLM apps and agents. The SDK is OpenTelemetry-native: it ships your model calls, tool calls, tokens, and cost to bryel as OpenInference traces — from any framework, not just one.

Vercel AI SDK

One processor + a per-call flag. No wrapping your model calls.

Any OpenTelemetry app

Already emit OpenInference spans? Add one processor and you’re done.

OpenAI · Anthropic · LangChain

Use an OpenInference instrumentor + the bryel processor.

User feedback

Record 👍/👎, scores, and corrections against a generation.

How it works

The SDK is, at its core, an OTLP exporter + a span processor. Any OpenTelemetry app can use it; adapters (like the Vercel one) add framework-aware mapping on top.

Packages

Package	Install for
`@bryel/sdk-core`	Any OpenTelemetry / OpenInference app
`@bryel/vercel`	The Vercel AI SDK
`@bryel/feedback`	Recording user feedback
`@bryel/semconv`	Conventions for custom mappers

Quickstart

⌘I