@respan/instrumentation-pydantic-ai

Respan instrumentation for Pydantic AI-compatible TypeScript OpenTelemetry spans.

The repository does not currently document a stable TypeScript Pydantic AI SDK API to patch. This package therefore installs a conservative span processor instead: it normalizes Pydantic AI native OTel spans and Pydantic AI-scoped OpenInference spans into the canonical Respan span contract before export.

Installation

npm install @respan/respan @respan/instrumentation-pydantic-ai

Usage

import { Respan } from "@respan/respan";
import { PydanticAIInstrumentor } from "@respan/instrumentation-pydantic-ai";

const respan = new Respan({
  apiKey: process.env.RESPAN_API_KEY,
  baseURL: process.env.RESPAN_BASE_URL,
  appName: "pydantic-ai-typescript",
  instrumentations: [new PydanticAIInstrumentor()],
});

await respan.initialize();

The instrumentor expects spans to already be emitted by a Pydantic AI-compatible runtime or OpenInference bridge. It does not invent or monkey-patch unknown SDK entrypoints.

Mapped Span Shapes

• Native Pydantic AI-style GenAI spans such as gen_ai.operation.name, gen_ai.input.messages, gen_ai.output.messages, model_request_parameters, gen_ai.tool.definitions, and tool call attrs.
• Pydantic AI-scoped OpenInference spans with openinference.span.kind and indexed llm.input_messages.* / llm.output_messages.* attrs.

The package emits canonical Respan fields including:

• respan.entity.log_type
• traceloop.entity.name
• traceloop.entity.path
• traceloop.entity.input
• traceloop.entity.output
• llm.request.type
• llm.request.functions
• gen_ai.prompt.*
• gen_ai.completion.*
• gen_ai.usage.*

Raw Pydantic AI/OpenInference attrs and off-contract aliases are stripped before delegating to the active Respan span processor.