agentic_otel 0.3.1

OpenTelemetry export for the agentic framework: agent, model and tool spans sent over OTLP to any backend, with the GenAI semantic conventions.

Changelog

0.3.1

• Updated package documentation for the published 0.3.x release line.

0.3.0

Initial release, at the framework's current version.

• OtlpHttpSpanExporter sends spans to any OTLP endpoint over HTTP with a JSON body — no protobuf dependency and no build step. A failed export is reported through onError and dropped, never thrown into the code that produced the span: telemetry that breaks a run is worse than no telemetry.
• BatchSpanExporter buffers spans so export costs one request rather than one per span, which on a phone is one TLS handshake rather than thirty. A full queue drops the oldest and counts them; a rejected batch is dropped rather than retried into a collector that is down.
• OtlpResource describes what produced the spans. Setting serviceName is the difference between a usable trace view and one where everything is unknown_service.
• GenAiAttributes, GenAiOperations, and the genAiChatAttributes, genAiToolAttributes and genAiAgentAttributes helpers: the OpenTelemetry GenAI semantic conventions, which are what make a backend render an LLM span as an LLM span rather than as a generic one.
• encodeSpans, encodeSpan and encodeAttributes are public, so an application with its own transport can use the encoding without the client.

Prompts and completions are deliberately not exported. The attribute keys are here; attaching content is a line the application writes, after deciding what redaction applies.