@mayurajs/observability-langsmith

LangSmith for Mayura's traces (mayura/exporter-otlp), through LangSmith's OpenTelemetry endpoint.

npm install mayura @mayurajs/observability-langsmith

import { agentRunTraceSpans } from 'mayura/exporter-otlp';
import { langSmithTraceExporter } from '@mayurajs/observability-langsmith';

const langsmith = langSmithTraceExporter({ apiKey, project: 'support-agent', region: 'us', serviceName: 'support-agent' });
await langsmith.sink(agentRunTraceSpans(observer.inspect(handle.id)?.recent ?? []), { signal: AbortSignal.timeout(10_000) });

• Regions us (the default), eu, apac and aws-us, or baseUrl for a self-hosted LangSmith's API URL.
• project names the project the traces go to; LangSmith's default project when left out.
• Each span carries its LangSmith run type (langsmith.span.kind: llm for model calls, tool for tool calls, chain for runs) and the provider as gen_ai.system, beside its GenAI attributes, so runs show as chains with their LLM and tool runs, models and token usage. langSmithAttributes is exported to reuse the mapping.
• Spans are metadata only, so traces carry no prompts or outputs.
• The sink also works as the sink of createWorkflowTraceExport, for durable workflow runs.
• The key is passed in; nothing is read from the environment. Every other option is the OTLP trace exporter's.

See the observability guide. Apache-2.0.