struct-sdk 0.2.11__tar.gz → 0.2.12__tar.gz

{struct_sdk-0.2.11 → struct_sdk-0.2.12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: struct-sdk
-Version: 0.2.11
+Version: 0.2.12
 Summary: Struct agent observability SDK — auto-instruments AI agent frameworks with OpenTelemetry
 Project-URL: Homepage, https://struct.ai
 Project-URL: Documentation, https://struct.ai/docs

{struct_sdk-0.2.11 → struct_sdk-0.2.12}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "struct-sdk"
-version = "0.2.11"
+version = "0.2.12"
 description = "Struct agent observability SDK — auto-instruments AI agent frameworks with OpenTelemetry"
 readme = "README.md"
 requires-python = ">=3.10"

{struct_sdk-0.2.11 → struct_sdk-0.2.12}/src/struct_sdk/openai.py

@@ -333,6 +333,17 @@ def _set_response_attrs(span: trace.Span, sdk: StructSDK, response: Any, otel_lo
         # them and gets them re-added in struct_sdk.anthropic. Report as-is.
         span.set_attribute("gen_ai.usage.input_tokens", input_tokens)
         span.set_attribute("gen_ai.usage.output_tokens", output_tokens)
+        # Reasoning tokens are a SUBSET of ``output_tokens`` (chain-of-thought /
+        # extended thinking), not an additional bucket — surfaced here purely
+        # for observability, never for cost (``output_tokens`` already prices
+        # them at the output rate). ``gen_ai.usage.reasoning.output_tokens`` is
+        # the OTel GenAI semconv name; nested style matches ``cache_read`` below.
+        details_out = getattr(usage, "output_tokens_details", None)
+        reasoning_tokens = (
+            (getattr(details_out, "reasoning_tokens", 0) or 0) if details_out is not None else 0
+        )
+        if reasoning_tokens:
+            span.set_attribute("gen_ai.usage.reasoning.output_tokens", reasoning_tokens)
         if cache_read:
             # Nested-key naming (``cache_read.input_tokens``) matches the
             # Anthropic emitter — Struct's ingest views key on this exact
@@ -340,8 +351,11 @@ def _set_response_attrs(span: trace.Span, sdk: StructSDK, response: Any, otel_lo
             # ``gen_ai.usage.cache_read_input_tokens``: ingest compatibility
             # beats semconv purity here.
             span.set_attribute("gen_ai.usage.cache_read.input_tokens", cache_read)
-        # The Responses API usage block has no cache-write counter, so
-        # ``gen_ai.usage.cache_creation.input_tokens`` is never set here.
+        cache_write = (getattr(details, "cache_write_tokens", 0) or 0) if details is not None else 0
+        if cache_write:
+            # gpt-5.6 family charges for cache writes (1.25x input). Nested-key
+            # name matches the Anthropic emitter / Struct ingest views.
+            span.set_attribute("gen_ai.usage.cache_creation.input_tokens", cache_write)
 
     if getattr(response, "model", None):
         span.set_attribute("gen_ai.response.model", response.model)