agenta 0.27.7__py3-none-any.whl → 0.27.7a0__py3-none-any.whl

agenta/sdk/decorators/routing.py

@@ -143,7 +143,9 @@ class entrypoint:
                 _MIDDLEWARES = False
 
         except:  # pylint: disable=bare-except
-            log.warning("Agenta SDK - failed to secure route: %s", route_path)
+            log.error("------------------------------------")
+            log.error("Agenta SDK - failed to secure route: %s", route_path)
+            log.error("------------------------------------")
         ### --- Update Middleware --- #
 
         DEFAULT_PATH = "generate"
@@ -355,7 +357,9 @@ class entrypoint:
         *args,
         **func_params,
     ):
-        log.info("Agenta SDK - handling route: %s", repr(self.route_path or "/"))
+        log.info("---------------------------")
+        log.info(f"Agenta SDK - running route: {repr(self.route_path or '/')}")
+        log.info("---------------------------")
 
         tracing_context.set(routing_context.get())
 
@@ -373,32 +377,36 @@ class entrypoint:
 
     async def handle_success(self, result: Any, inline_trace: bool):
         data = None
-        tree = None
+        trace = dict()
 
         with suppress():
             data = self.patch_result(result)
 
             if inline_trace:
-                tree = await self.fetch_inline_trace(inline_trace)
+                trace = await self.fetch_inline_trace(inline_trace)
 
         log.info(f"----------------------------------")
         log.info(f"Agenta SDK - exiting with success: 200")
         log.info(f"----------------------------------")
 
-        return BaseResponse(data=data, tree=tree)
+        return BaseResponse(data=data, trace=trace)
 
     def handle_failure(self, error: Exception):
-        log.warning("--------------------------------------------------")
-        log.warning("Agenta SDK - handling application exception below:")
-        log.warning("--------------------------------------------------")
-        log.warning(format_exc().strip("\n"))
-        log.warning("--------------------------------------------------")
+        log.error("--------------------------------------------------")
+        log.error("Agenta SDK - handling application exception below:")
+        log.error("--------------------------------------------------")
+        log.error(format_exc().strip("\n"))
+        log.error("--------------------------------------------------")
 
         status_code = error.status_code if hasattr(error, "status_code") else 500
         message = str(error)
         stacktrace = format_exception(error, value=error, tb=error.__traceback__)  # type: ignore
         detail = {"message": message, "stacktrace": stacktrace}
 
+        log.error(f"----------------------------------")
+        log.error(f"Agenta SDK - exiting with failure: {status_code}")
+        log.error(f"----------------------------------")
+
         raise HTTPException(status_code=status_code, detail=detail)
 
     def patch_result(self, result: Any):
@@ -675,7 +683,10 @@ class entrypoint:
 
         loop = get_event_loop()
 
-        with routing_context_manager(config=args_config_params):
+        with routing_context_manager(
+            config=args_config_params,
+            environment="terminal",
+        ):
             result = loop.run_until_complete(
                 self.execute_function(
                     func,
@@ -684,23 +695,30 @@ class entrypoint:
                 )
             )
 
+        SHOW_DETAILS = True
+        SHOW_DATA = False
+        SHOW_TRACE = False
+
         if result.trace:
             log.info("\n========= Result =========\n")
 
             log.info(f"trace_id: {result.trace['trace_id']}")
-            log.info(f"latency:  {result.trace.get('latency')}")
-            log.info(f"cost:     {result.trace.get('cost')}")
-            log.info(f"usage:   {list(result.trace.get('usage', {}).values())}")
-
-            log.info(" ")
-            log.info("data:")
-            log.info(dumps(result.data, indent=2))
-
-            log.info(" ")
-            log.info("trace:")
-            log.info("----------------")
-            log.info(dumps(result.trace.get("spans", []), indent=2))
-            log.info("----------------")
+            if SHOW_DETAILS:
+                log.info(f"latency:  {result.trace.get('latency')}")
+                log.info(f"cost:     {result.trace.get('cost')}")
+                log.info(f"usage:   {list(result.trace.get('usage', {}).values())}")
+
+            if SHOW_DATA:
+                log.info(" ")
+                log.info(f"data:")
+                log.info(dumps(result.data, indent=2))
+
+            if SHOW_TRACE:
+                log.info(" ")
+                log.info(f"trace:")
+                log.info(f"----------------")
+                log.info(dumps(result.trace.get("spans", []), indent=2))
+                log.info(f"----------------")
 
             log.info("\n==========================\n")
 

agenta/sdk/litellm/litellm.py

@@ -60,11 +60,7 @@ def litellm_handler():
             )
 
             if not self.span:
-                log.warning("Agenta SDK - litellm tracing failed")
-                return
-
-            if not self.span.is_recording():
-                log.error("Agenta SDK - litellm span not recording.")
+                log.error("LiteLLM callback error: span not found.")
                 return
 
             self.span.set_attributes(
@@ -90,11 +86,43 @@ def litellm_handler():
             end_time,
         ):
             if not self.span:
-                log.warning("Agenta SDK - litellm tracing failed")
+                log.error("LiteLLM callback error: span not found.")
                 return
 
-            if not self.span.is_recording():
-                return
+            try:
+                result = []
+                for choice in response_obj.choices:
+                    message = choice.message.__dict__
+                    result.append(message)
+
+                outputs = {"completion": result}
+                self.span.set_attributes(
+                    attributes={"outputs": outputs},
+                    namespace="data",
+                )
+
+            except Exception as e:
+                pass
+
+            self.span.set_attributes(
+                attributes={"total": kwargs.get("response_cost")},
+                namespace="metrics.unit.costs",
+            )
+
+            self.span.set_attributes(
+                attributes=(
+                    {
+                        "prompt": response_obj.usage.prompt_tokens,
+                        "completion": response_obj.usage.completion_tokens,
+                        "total": response_obj.usage.total_tokens,
+                    }
+                ),
+                namespace="metrics.unit.tokens",
+            )
+
+            self.span.set_status(status="OK")
+
+            self.span.end()
 
         def log_success_event(
             self,
@@ -103,14 +131,8 @@ def litellm_handler():
             start_time,
             end_time,
         ):
-            if kwargs.get("stream"):
-                return
-
             if not self.span:
-                log.warning("Agenta SDK - litellm tracing failed")
-                return
-
-            if not self.span.is_recording():
+                log.error("LiteLLM callback error: span not found.")
                 return
 
             try:
@@ -156,10 +178,7 @@ def litellm_handler():
             end_time,
         ):
             if not self.span:
-                log.warning("Agenta SDK - litellm tracing failed")
-                return
-
-            if not self.span.is_recording():
+                log.error("LiteLLM callback error: span not found.")
                 return
 
             self.span.record_exception(kwargs["exception"])
@@ -176,11 +195,43 @@ def litellm_handler():
             end_time,
         ):
             if not self.span:
-                log.warning("Agenta SDK - litellm tracing failed")
+                log.error("LiteLLM callback error: span not found.")
                 return
 
-            if not self.span.is_recording():
-                return
+            try:
+                result = []
+                for choice in response_obj.choices:
+                    message = choice.message.__dict__
+                    result.append(message)
+
+                outputs = {"completion": result}
+                self.span.set_attributes(
+                    attributes={"outputs": outputs},
+                    namespace="data",
+                )
+
+            except Exception as e:
+                pass
+
+            self.span.set_attributes(
+                attributes={"total": kwargs.get("response_cost")},
+                namespace="metrics.unit.costs",
+            )
+
+            self.span.set_attributes(
+                attributes=(
+                    {
+                        "prompt": response_obj.usage.prompt_tokens,
+                        "completion": response_obj.usage.completion_tokens,
+                        "total": response_obj.usage.total_tokens,
+                    }
+                ),
+                namespace="metrics.unit.tokens",
+            )
+
+            self.span.set_status(status="OK")
+
+            self.span.end()
 
         async def async_log_success_event(
             self,
@@ -190,10 +241,7 @@ def litellm_handler():
             end_time,
         ):
             if not self.span:
-                log.warning("Agenta SDK - litellm tracing failed")
-                return
-
-            if not self.span.is_recording():
+                log.error("LiteLLM callback error: span not found.")
                 return
 
             try:
@@ -239,10 +287,7 @@ def litellm_handler():
             end_time,
         ):
             if not self.span:
-                log.warning("Agenta SDK - litellm tracing failed")
-                return
-
-            if not self.span.is_recording():
+                log.error("LiteLLM callback error: span not found.")
                 return
 
             self.span.record_exception(kwargs["exception"])

agenta/sdk/middleware/auth.py

@@ -27,8 +27,6 @@ AGENTA_SDK_AUTH_CACHE = str(environ.get("AGENTA_SDK_AUTH_CACHE", True)).lower()
     "t",
 )
 
-AGENTA_SDK_AUTH_CACHE = False
-
 AGENTA_UNAUTHORIZED_EXECUTION_ALLOWED = str(
     environ.get("AGENTA_UNAUTHORIZED_EXECUTION_ALLOWED", False)
 ).lower() in ("true", "1", "t")
@@ -63,6 +61,7 @@ class AuthorizationMiddleware(BaseHTTPMiddleware):
         self,
         request: Request,
         call_next: Callable,
+        project_id: Optional[UUID] = None,
     ):
         if AGENTA_UNAUTHORIZED_EXECUTION_ALLOWED:
             return await call_next(request)
@@ -84,8 +83,6 @@ class AuthorizationMiddleware(BaseHTTPMiddleware):
                 "resource_id": self.resource_id,
             }
 
-            project_id = request.query_params.get("project_id")
-
             if project_id:
                 params["project_id"] = project_id
 
@@ -136,10 +133,10 @@ class AuthorizationMiddleware(BaseHTTPMiddleware):
             return await call_next(request)
 
         except:  # pylint: disable=bare-except
-            log.warning("------------------------------------------------------")
-            log.warning("Agenta SDK - handling auth middleware exception below:")
-            log.warning("------------------------------------------------------")
-            log.warning(format_exc().strip("\n"))
-            log.warning("------------------------------------------------------")
+            log.error("------------------------------------------------------")
+            log.error("Agenta SDK - handling auth middleware exception below:")
+            log.error("------------------------------------------------------")
+            log.error(format_exc().strip("\n"))
+            log.error("------------------------------------------------------")
 
             return Deny()

agenta/sdk/tracing/context.py

@@ -14,11 +14,11 @@ def tracing_context_manager():
     token = tracing_context.set(_tracing_context)
     try:
         yield
-    except:  # pylint: disable=bare-except
-        log.warning("----------------------------------------------")
-        log.warning("Agenta SDK - handling tracing exception below:")
-        log.warning("----------------------------------------------")
-        log.warning(format_exc().strip("\n"))
-        log.warning("----------------------------------------------")
+    except Exception as e:
+        log.error("----------------------------------------------")
+        log.error("Agenta SDK - handling tracing exception below:")
+        log.error("----------------------------------------------")
+        log.error(format_exc().strip("\n"))
+        log.error("----------------------------------------------")
     finally:
         tracing_context.reset(token)

agenta/sdk/tracing/inline.py

@@ -41,6 +41,7 @@ from uuid import UUID
 class TimeDTO(BaseModel):
     start: datetime
     end: datetime
+    span: int
 
 
 class StatusCode(Enum):
@@ -845,9 +846,12 @@ def parse_from_otel_span_dto(
         else None
     )
 
+    duration = (otel_span_dto.end_time - otel_span_dto.start_time).total_seconds()
+
     time = TimeDTO(
         start=otel_span_dto.start_time,
         end=otel_span_dto.end_time,
+        span=round(duration * 1_000_000),  # microseconds
     )
 
     status = StatusDTO(
@@ -859,13 +863,6 @@ def parse_from_otel_span_dto(
 
     data, metrics, meta, tags, refs = _parse_from_attributes(otel_span_dto)
 
-    duration = (otel_span_dto.end_time - otel_span_dto.start_time).total_seconds()
-
-    if metrics is None:
-        metrics = dict()
-
-    metrics["acc.duration.total"] = round(duration * 1_000, 3)  # milliseconds
-
     root_id = str(tree_id)
     if refs is not None:
         root_id = refs.get("scenario.id", root_id)
@@ -906,9 +903,9 @@ def parse_to_agenta_span_dto(
     if span_dto.data:
         span_dto.data = _unmarshal_attributes(span_dto.data)
 
-        if "outputs" in span_dto.data:
-            if "__default__" in span_dto.data["outputs"]:
-                span_dto.data["outputs"] = span_dto.data["outputs"]["__default__"]
+        # if "outputs" in span_dto.data:
+        #     if "__default__" in span_dto.data["outputs"]:
+        #         span_dto.data["outputs"] = span_dto.data["outputs"]["__default__"]
 
     # METRICS
     if span_dto.metrics:
@@ -937,17 +934,6 @@ def parse_to_agenta_span_dto(
             else:
                 parse_to_agenta_span_dto(v)
 
-    # MASK LINKS FOR NOW
-    span_dto.links = None
-    # ------------------
-
-    # MASK LIFECYCLE FOR NOW
-    # span_dto.lifecycle = None
-    if span_dto.lifecycle:
-        span_dto.lifecycle.updated_at = None
-        span_dto.lifecycle.updated_by_id = None
-    # ----------------------
-
     return span_dto
 
 
@@ -959,8 +945,6 @@ def parse_to_agenta_span_dto(
 from litellm import cost_calculator
 from opentelemetry.sdk.trace import ReadableSpan
 
-from agenta.sdk.types import AgentaNodeDto, AgentaNodesResponse
-
 
 def parse_inline_trace(
     spans: Dict[str, ReadableSpan],
@@ -1008,19 +992,51 @@ def parse_inline_trace(
     ### services.observability.service.query() ###
     ##############################################
 
-    spans = [
-        loads(
-            span_dto.model_dump_json(
-                exclude_none=True,
-                exclude_defaults=True,
-            )
-        )
-        for span_dto in agenta_span_dtos
-    ]
-    inline_trace = AgentaNodesResponse(
-        version="1.0.0",
-        nodes=[AgentaNodeDto(**span) for span in spans],
-    ).model_dump(exclude_none=True, exclude_unset=True)
+    LEGACY = True
+    inline_trace = None
+
+    if LEGACY:
+        legacy_spans = [
+            _parse_to_legacy_span(span_dto) for span_dto in span_idx.values()
+        ]
+
+        root_span = agenta_span_dtos[0]
+
+        trace_id = root_span.root.id.hex
+        latency = root_span.time.span / 1_000_000
+        cost = root_span.metrics.get("acc", {}).get("costs", {}).get("total", 0.0)
+        tokens = {
+            "prompt_tokens": root_span.metrics.get("acc", {})
+            .get("tokens", {})
+            .get("prompt", 0),
+            "completion_tokens": root_span.metrics.get("acc", {})
+            .get("tokens", {})
+            .get("completion", 0),
+            "total_tokens": root_span.metrics.get("acc", {})
+            .get("tokens", {})
+            .get("total", 0),
+        }
+
+        spans = [
+            loads(span.model_dump_json(exclude_none=True)) for span in legacy_spans
+        ]
+
+        inline_trace = {
+            "trace_id": trace_id,
+            "latency": latency,
+            "cost": cost,
+            "usage": tokens,
+            "spans": spans,
+        }
+
+    else:
+        spans = [
+            loads(span_dto.model_dump_json(exclude_none=True))
+            for span_dto in agenta_span_dtos
+        ]
+
+        inline_trace = spans  # turn into Agenta Model ?
+
     return inline_trace
 
 
@@ -1104,6 +1120,98 @@ class LlmTokens(BaseModel):
     total_tokens: Optional[int] = 0
 
 
+class CreateSpan(BaseModel):
+    id: str
+    app_id: str
+    variant_id: Optional[str] = None
+    variant_name: Optional[str] = None
+    inputs: Optional[Dict[str, Optional[Any]]] = None
+    internals: Optional[Dict[str, Optional[Any]]] = None
+    outputs: Optional[Union[str, Dict[str, Optional[Any]], List[Any]]] = None
+    config: Optional[Dict[str, Optional[Any]]] = None
+    environment: Optional[str] = None
+    tags: Optional[List[str]] = None
+    token_consumption: Optional[int] = None
+    name: str
+    parent_span_id: Optional[str] = None
+    attributes: Optional[Dict[str, Optional[Any]]] = None
+    spankind: str
+    status: str
+    user: Optional[str] = None
+    start_time: datetime
+    end_time: datetime
+    tokens: Optional[LlmTokens] = None
+    cost: Optional[float] = None
+
+
+def _parse_to_legacy_span(span: SpanDTO) -> CreateSpan:
+    attributes = None
+    if span.otel:
+        attributes = span.otel.attributes or {}
+
+        if span.otel.events:
+            for event in span.otel.events:
+                if event.name == "exception":
+                    attributes.update(**event.attributes)
+
+    legacy_span = CreateSpan(
+        id=span.node.id.hex[:24],
+        spankind=span.node.type,
+        name=span.node.name,
+        #
+        status=span.status.code.name,
+        #
+        start_time=span.time.start,
+        end_time=span.time.end,
+        #
+        parent_span_id=span.parent.id.hex[:24] if span.parent else None,
+        #
+        inputs=span.data.get("inputs") if span.data else {},
+        internals=span.data.get("internals") if span.data else {},
+        outputs=span.data.get("outputs") if span.data else {},
+        #
+        environment=span.meta.get("environment") if span.meta else None,
+        config=span.meta.get("configuration") if span.meta else None,
+        #
+        tokens=(
+            LlmTokens(
+                prompt_tokens=span.metrics.get("acc", {})
+                .get("tokens", {})
+                .get("prompt", 0.0),
+                completion_tokens=span.metrics.get("acc", {})
+                .get("tokens", {})
+                .get("completion", 0.0),
+                total_tokens=span.metrics.get("acc", {})
+                .get("tokens", {})
+                .get("total", 0.0),
+            )
+            if span.metrics
+            else None
+        ),
+        cost=(
+            span.metrics.get("acc", {}).get("costs", {}).get("total", 0.0)
+            if span.metrics
+            else None
+        ),
+        #
+        app_id=(
+            span.refs.get("application", {}).get("id", "missing-app-id")
+            if span.refs
+            else "missing-app-id"
+        ),
+        #
+        attributes=attributes,
+        #
+        variant_id=None,
+        variant_name=None,
+        tags=None,
+        token_consumption=None,
+        user=None,
+    )
+
+    return legacy_span
+
+
 TYPES_WITH_COSTS = [
     "embedding",
     "query",
@@ -1121,15 +1229,13 @@ def calculate_costs(span_idx: Dict[str, SpanDTO]):
             and span.meta
             and span.metrics
         ):
-            model = span.meta.get("response.model")
-            prompt_tokens = span.metrics.get("unit.tokens.prompt", 0.0)
-            completion_tokens = span.metrics.get("unit.tokens.completion", 0.0)
-
             try:
                 costs = cost_calculator.cost_per_token(
-                    model=model,
-                    prompt_tokens=prompt_tokens,
-                    completion_tokens=completion_tokens,
+                    model=span.meta.get("response.model"),
+                    prompt_tokens=span.metrics.get("unit.tokens.prompt", 0.0),
+                    completion_tokens=span.metrics.get("unit.tokens.completion", 0.0),
+                    call_type=span.node.type.name.lower(),
+                    response_time_ms=span.time.span // 1_000,
                 )
 
                 if not costs:
@@ -1142,8 +1248,5 @@ def calculate_costs(span_idx: Dict[str, SpanDTO]):
                 span.metrics["unit.costs.completion"] = completion_cost
                 span.metrics["unit.costs.total"] = total_cost
 
-            except:  # pylint: disable=bare-except
-                print("Failed to calculate costs:")
-                print(
-                    f"model={model}, prompt_tokens={prompt_tokens}, completion_tokens={completion_tokens}"
-                )
+            except:
+                pass

agenta/sdk/tracing/processors.py

@@ -91,7 +91,9 @@ class TraceProcessor(BatchSpanProcessor):
         ret = super().force_flush(timeout_millis)
 
         if not ret:
-            log.warning("Agenta SDK - skipping export due to timeout.")
+            log.error("--------------------------------------------")
+            log.error("Agenta SDK - skipping export due to timeout.")
+            log.error("--------------------------------------------")
 
     def is_ready(
         self,

agenta/sdk/tracing/tracing.py

@@ -90,6 +90,7 @@ class Tracing(metaclass=Singleton):
                 self.otlp_url,
             )
             log.info("--------------------------------------------")
+
             check(
                 self.otlp_url,
                 headers=self.headers,
@@ -105,10 +106,13 @@ class Tracing(metaclass=Singleton):
             )
 
             self.tracer_provider.add_span_processor(_otlp)
+
             log.info("Success: traces will be exported.")
             log.info("--------------------------------------------")
+
         except:  # pylint: disable=bare-except
-            log.warning("Agenta SDK - traces will not be exported.")
+            log.warning("Failure: traces will not be exported.")
+            log.warning("--------------------------------------------")
 
         # GLOBAL TRACER PROVIDER -- INSTRUMENTATION LIBRARIES
         set_tracer_provider(self.tracer_provider)

agenta/sdk/types.py

@@ -4,9 +4,6 @@ from typing import Dict, List, Optional, Any, Union
 
 from pydantic import ConfigDict, BaseModel, HttpUrl
 
-from agenta.client.backend.types.agenta_node_dto import AgentaNodeDto
-from agenta.client.backend.types.agenta_nodes_response import AgentaNodesResponse
-
 
 @dataclass
 class MultipleChoice:
@@ -26,9 +23,9 @@ class LLMTokenUsage(BaseModel):
 
 
 class BaseResponse(BaseModel):
-    version: Optional[str] = "3.0"
+    version: Optional[str] = "2.0"
     data: Optional[Union[str, Dict[str, Any]]] = None
-    tree: Optional[AgentaNodesResponse] = None
+    trace: Optional[Dict[str, Any]] = None
 
 
 class DictInput(dict):