ensure usage is included

autogpt_platform/backend/backend/copilot/otlp_trace.py

@@ -119,6 +119,10 @@ def build_otlp_payload(
     prompt_tokens: int | None = None,
     completion_tokens: int | None = None,
     total_tokens: int | None = None,
+    total_cost_usd: float | None = None,
+    cache_creation_input_tokens: int | None = None,
+    cache_read_input_tokens: int | None = None,
+    reasoning_tokens: int | None = None,
     user_id: str | None = None,
     session_id: str | None = None,
     tool_calls: list[dict[str, Any]] | None = None,
@@ -166,17 +170,14 @@ def build_otlp_payload(
         _kv("gen_ai.usage.input_tokens", prompt_tokens),
         _kv("gen_ai.usage.output_tokens", completion_tokens),
         _kv("gen_ai.usage.total_tokens", total_tokens),
-        _kv(
-            "gen_ai.usage.input_tokens.cached", 0 if prompt_tokens is not None else None
-        ),
-        _kv(
-            "gen_ai.usage.output_tokens.reasoning",
-            0 if completion_tokens is not None else None,
-        ),
+        _kv("gen_ai.usage.input_tokens.cached", cache_read_input_tokens),
+        _kv("gen_ai.usage.input_tokens.cache_creation", cache_creation_input_tokens),
+        _kv("gen_ai.usage.output_tokens.reasoning", reasoning_tokens),
         _kv("user.id", user_id),
         _kv("session.id", session_id),
         _kv("trace.metadata.openrouter.source", "openrouter"),
         _kv("trace.metadata.openrouter.user_id", user_id),
+        _kv("gen_ai.usage.total_cost", total_cost_usd),
         _kv("trace.metadata.openrouter.provider_name", provider_name),
         _kv("trace.metadata.openrouter.provider_slug", provider_slug),
         _kv("trace.metadata.openrouter.finish_reason", finish_reason),
@@ -184,40 +185,6 @@ def build_otlp_payload(
         if kv is not None:
             attrs.append(kv)
 
-    # Keep keyset aligned with historical OpenRouter traces even when values
-    # are unavailable from this path.
-    attrs.extend(
-        [
-            {"key": "gen_ai.usage.input_cost", "value": {"doubleValue": None}},
-            {"key": "gen_ai.usage.output_cost", "value": {"doubleValue": None}},
-            {"key": "gen_ai.usage.total_cost", "value": {"doubleValue": None}},
-            {
-                "key": "trace.metadata.openrouter.api_key_name",
-                "value": {"stringValue": None},
-            },
-            {
-                "key": "trace.metadata.openrouter.entity_id",
-                "value": {"stringValue": None},
-            },
-            {
-                "key": "trace.metadata.openrouter.creator_user_id",
-                "value": {"stringValue": None},
-            },
-            {
-                "key": "trace.metadata.openrouter.organization_id",
-                "value": {"stringValue": None},
-            },
-            {
-                "key": "trace.metadata.openrouter.input_unit_price",
-                "value": {"doubleValue": None},
-            },
-            {
-                "key": "trace.metadata.openrouter.output_unit_price",
-                "value": {"doubleValue": None},
-            },
-        ]
-    )
-
     if prompt_payload is not None:
         attrs.append({"key": "trace.input", "value": {"stringValue": prompt_payload}})
         attrs.append({"key": "span.input", "value": {"stringValue": prompt_payload}})
@@ -295,6 +262,10 @@ def emit_trace(
     prompt_tokens: int | None = None,
     completion_tokens: int | None = None,
     total_tokens: int | None = None,
+    total_cost_usd: float | None = None,
+    cache_creation_input_tokens: int | None = None,
+    cache_read_input_tokens: int | None = None,
+    reasoning_tokens: int | None = None,
     user_id: str | None = None,
     session_id: str | None = None,
     tool_calls: list[dict[str, Any]] | None = None,
@@ -319,6 +290,10 @@ def emit_trace(
         prompt_tokens=prompt_tokens,
         completion_tokens=completion_tokens,
         total_tokens=total_tokens,
+        total_cost_usd=total_cost_usd,
+        cache_creation_input_tokens=cache_creation_input_tokens,
+        cache_read_input_tokens=cache_read_input_tokens,
+        reasoning_tokens=reasoning_tokens,
         user_id=user_id,
         session_id=session_id,
         tool_calls=tool_calls,

autogpt_platform/backend/backend/copilot/sdk/service.py

@@ -620,6 +620,8 @@ async def stream_chat_completion_sdk(
             # are always bound when the OTLP trace section runs.
             assistant_response = ChatMessage(role="assistant", content="")
             trace_tool_calls: list[dict[str, Any]] = []
+            trace_usage: dict[str, Any] = {}
+            trace_cost_usd: float | None = None
 
             sdk_options_kwargs: dict[str, Any] = {
                 "system_prompt": system_prompt,
@@ -785,8 +787,13 @@ async def stream_chat_completion_sdk(
                                     - len(adapter.resolved_tool_calls),
                                 )
 
-                        # Log ResultMessage details for debugging
+                        # Extract usage and cost from ResultMessage for OTLP trace
                         if isinstance(sdk_msg, ResultMessage):
+                            if sdk_msg.usage:
+                                trace_usage = sdk_msg.usage
+                            if sdk_msg.total_cost_usd is not None:
+                                trace_cost_usd = sdk_msg.total_cost_usd
+
                             logger.info(
                                 "[SDK] [%s] Received: ResultMessage %s "
                                 "(unresolved=%d, current=%d, resolved=%d)",
@@ -1041,11 +1048,21 @@ async def stream_chat_completion_sdk(
             dict(m) for m in session.to_openai_messages()
         ]
 
+        _input = trace_usage.get("input_tokens")
+        _output = trace_usage.get("output_tokens")
+        _total = (_input or 0) + (_output or 0) if _input or _output else None
+
         emit_trace(
             model=sdk_model or config.model,
             messages=trace_messages,
             assistant_content=assistant_response.content or None,
             finish_reason=trace_finish_reason,
+            prompt_tokens=_input,
+            completion_tokens=_output,
+            total_tokens=_total,
+            total_cost_usd=trace_cost_usd,
+            cache_creation_input_tokens=trace_usage.get("cache_creation_input_tokens"),
+            cache_read_input_tokens=trace_usage.get("cache_read_input_tokens"),
             user_id=user_id,
             session_id=session_id,
             tool_calls=trace_tool_calls_payload,

autogpt_platform/backend/backend/copilot/service.py

@@ -1043,11 +1043,7 @@ async def _stream_chat_chunks(
 
                 # Accumulators for OTLP trace export
                 _trace_text_parts: list[str] = []
-                _trace_usage: dict[str, int | None] = {
-                    "prompt": None,
-                    "completion": None,
-                    "total": None,
-                }
+                _trace_usage: dict[str, Any] = {}
 
                 # Process the stream
                 chunk: ChatCompletionChunk
@@ -1057,6 +1053,12 @@ async def _stream_chat_chunks(
                         _trace_usage["prompt"] = chunk.usage.prompt_tokens
                         _trace_usage["completion"] = chunk.usage.completion_tokens
                         _trace_usage["total"] = chunk.usage.total_tokens
+                        if chunk.usage.prompt_tokens_details:
+                            d = chunk.usage.prompt_tokens_details
+                            _trace_usage["cached"] = d.cached_tokens
+                        if chunk.usage.completion_tokens_details:
+                            d = chunk.usage.completion_tokens_details
+                            _trace_usage["reasoning"] = d.reasoning_tokens
                         yield StreamUsage(
                             promptTokens=chunk.usage.prompt_tokens,
                             completionTokens=chunk.usage.completion_tokens,
@@ -1190,9 +1192,11 @@ async def _stream_chat_chunks(
                     finish_reason=(
                         "tool_calls" if tool_calls else (finish_reason or "stop")
                     ),
-                    prompt_tokens=_trace_usage["prompt"],
-                    completion_tokens=_trace_usage["completion"],
-                    total_tokens=_trace_usage["total"],
+                    prompt_tokens=_trace_usage.get("prompt"),
+                    completion_tokens=_trace_usage.get("completion"),
+                    total_tokens=_trace_usage.get("total"),
+                    cache_read_input_tokens=_trace_usage.get("cached"),
+                    reasoning_tokens=_trace_usage.get("reasoning"),
                     user_id=session.user_id,
                     session_id=session.session_id,
                     tool_calls=tool_calls or None,