agenta 0.27.4a1__py3-none-any.whl → 0.27.5a1__py3-none-any.whl

agenta/client/backend/variants/client.py

@@ -1104,14 +1104,14 @@ class VariantsClient:
     def configs_delete(
         self,
         *,
-        variant_ref: ReferenceRequestModel,
+        variant_ref: typing.Optional[ReferenceRequestModel] = OMIT,
         application_ref: typing.Optional[ReferenceRequestModel] = OMIT,
         request_options: typing.Optional[RequestOptions] = None,
     ) -> int:
         """
         Parameters
         ----------
-        variant_ref : ReferenceRequestModel
+        variant_ref : typing.Optional[ReferenceRequestModel]
 
         application_ref : typing.Optional[ReferenceRequestModel]
 
@@ -1125,15 +1125,13 @@ class VariantsClient:
 
         Examples
         --------
-        from agenta import AgentaApi, ReferenceRequestModel
+        from agenta import AgentaApi
 
         client = AgentaApi(
             api_key="YOUR_API_KEY",
             base_url="https://yourhost.com/path/to/api",
         )
-        client.variants.configs_delete(
-            variant_ref=ReferenceRequestModel(),
-        )
+        client.variants.configs_delete()
         """
         _response = self._client_wrapper.httpx_client.request(
             "variants/configs/delete",
@@ -1246,14 +1244,14 @@ class VariantsClient:
     def configs_history(
         self,
         *,
-        variant_ref: ReferenceRequestModel,
+        variant_ref: typing.Optional[ReferenceRequestModel] = OMIT,
         application_ref: typing.Optional[ReferenceRequestModel] = OMIT,
         request_options: typing.Optional[RequestOptions] = None,
     ) -> typing.List[ConfigResponseModel]:
         """
         Parameters
         ----------
-        variant_ref : ReferenceRequestModel
+        variant_ref : typing.Optional[ReferenceRequestModel]
 
         application_ref : typing.Optional[ReferenceRequestModel]
 
@@ -1267,15 +1265,13 @@ class VariantsClient:
 
         Examples
         --------
-        from agenta import AgentaApi, ReferenceRequestModel
+        from agenta import AgentaApi
 
         client = AgentaApi(
             api_key="YOUR_API_KEY",
             base_url="https://yourhost.com/path/to/api",
         )
-        client.variants.configs_history(
-            variant_ref=ReferenceRequestModel(),
-        )
+        client.variants.configs_history()
         """
         _response = self._client_wrapper.httpx_client.request(
             "variants/configs/history",
@@ -2508,14 +2504,14 @@ class AsyncVariantsClient:
     async def configs_delete(
         self,
         *,
-        variant_ref: ReferenceRequestModel,
+        variant_ref: typing.Optional[ReferenceRequestModel] = OMIT,
         application_ref: typing.Optional[ReferenceRequestModel] = OMIT,
         request_options: typing.Optional[RequestOptions] = None,
     ) -> int:
         """
         Parameters
         ----------
-        variant_ref : ReferenceRequestModel
+        variant_ref : typing.Optional[ReferenceRequestModel]
 
         application_ref : typing.Optional[ReferenceRequestModel]
 
@@ -2531,7 +2527,7 @@ class AsyncVariantsClient:
         --------
         import asyncio
 
-        from agenta import AsyncAgentaApi, ReferenceRequestModel
+        from agenta import AsyncAgentaApi
 
         client = AsyncAgentaApi(
             api_key="YOUR_API_KEY",
@@ -2540,9 +2536,7 @@ class AsyncVariantsClient:
 
 
         async def main() -> None:
-            await client.variants.configs_delete(
-                variant_ref=ReferenceRequestModel(),
-            )
+            await client.variants.configs_delete()
 
 
         asyncio.run(main())
@@ -2666,14 +2660,14 @@ class AsyncVariantsClient:
     async def configs_history(
         self,
         *,
-        variant_ref: ReferenceRequestModel,
+        variant_ref: typing.Optional[ReferenceRequestModel] = OMIT,
         application_ref: typing.Optional[ReferenceRequestModel] = OMIT,
         request_options: typing.Optional[RequestOptions] = None,
     ) -> typing.List[ConfigResponseModel]:
         """
         Parameters
         ----------
-        variant_ref : ReferenceRequestModel
+        variant_ref : typing.Optional[ReferenceRequestModel]
 
         application_ref : typing.Optional[ReferenceRequestModel]
 
@@ -2689,7 +2683,7 @@ class AsyncVariantsClient:
         --------
         import asyncio
 
-        from agenta import AsyncAgentaApi, ReferenceRequestModel
+        from agenta import AsyncAgentaApi
 
         client = AsyncAgentaApi(
             api_key="YOUR_API_KEY",
@@ -2698,9 +2692,7 @@ class AsyncVariantsClient:
 
 
         async def main() -> None:
-            await client.variants.configs_history(
-                variant_ref=ReferenceRequestModel(),
-            )
+            await client.variants.configs_history()
 
 
         asyncio.run(main())

agenta/sdk/__init__.py

@@ -1,4 +1,4 @@
-from typing import Optional
+from typing import Optional, Callable, Any
 
 from .utils.preinit import PreInitObject  # always the first import!
 
@@ -17,8 +17,6 @@ from .types import (
     FileInputURL,
     BinaryParam,
     Prompt,
-    AgentaNodeDto,
-    AgentaNodesResponse,
 )
 
 from .tracing import Tracing, get_tracer
@@ -45,17 +43,21 @@ tracer = get_tracer(tracing)
 
 def init(
     host: Optional[str] = None,
-    app_id: Optional[str] = None,
     api_key: Optional[str] = None,
     config_fname: Optional[str] = None,
+    redact: Optional[Callable[..., Any]] = None,
+    redact_on_error: Optional[bool] = True,
+    # DEPRECATING
+    app_id: Optional[str] = None,
 ):
-    global api, async_api, tracing, tracer
+    global api, async_api, tracing, tracer  # pylint: disable=global-statement
 
     _init(
         host=host,
         api_key=api_key,
         config_fname=config_fname,
-        # DEPRECATING
+        redact=redact,
+        redact_on_error=redact_on_error,
         app_id=app_id,
     )
 

agenta/sdk/agenta_init.py

@@ -1,7 +1,7 @@
 import logging
 import toml
 from os import getenv
-from typing import Optional
+from typing import Optional, Callable, Any
 from importlib.metadata import version
 
 from agenta.sdk.utils.logging import log
@@ -36,6 +36,8 @@ class AgentaSingleton:
         host: Optional[str] = None,
         api_key: Optional[str] = None,
         config_fname: Optional[str] = None,
+        redact: Optional[Callable[..., Any]] = None,
+        redact_on_error: Optional[bool] = True,
         # DEPRECATING
         app_id: Optional[str] = None,
     ) -> None:
@@ -91,6 +93,8 @@ class AgentaSingleton:
 
         self.tracing = Tracing(
             url=f"{self.host}/api/observability/v1/otlp/traces",  # type: ignore
+            redact=redact,
+            redact_on_error=redact_on_error,
         )
 
         self.tracing.configure(
@@ -258,7 +262,9 @@ def init(
     host: Optional[str] = None,
     api_key: Optional[str] = None,
     config_fname: Optional[str] = None,
-    # DEPRECATED
+    redact: Optional[Callable[..., Any]] = None,
+    redact_on_error: Optional[bool] = True,
+    # DEPRECATING
     app_id: Optional[str] = None,
 ):
     """Main function to initialize the agenta sdk.
@@ -289,7 +295,8 @@ def init(
         host=host,
         api_key=api_key,
         config_fname=config_fname,
-        # DEPRECATED
+        redact=redact,
+        redact_on_error=redact_on_error,
         app_id=app_id,
     )
 

agenta/sdk/decorators/routing.py

@@ -386,7 +386,7 @@ class entrypoint:
         log.info(f"Agenta SDK - exiting with success: 200")
         log.info(f"----------------------------------")
 
-        return BaseResponse(data=data, tree=trace)
+        return BaseResponse(data=data, trace=trace)
 
     def handle_failure(self, error: Exception):
         log.error("--------------------------------------------------")

agenta/sdk/decorators/tracing.py

@@ -19,6 +19,8 @@ class instrument:  # pylint: disable=invalid-name
         config: Optional[Dict[str, Any]] = None,
         ignore_inputs: Optional[bool] = None,
         ignore_outputs: Optional[bool] = None,
+        redact: Optional[Callable[..., Any]] = None,
+        redact_on_error: Optional[bool] = True,
         max_depth: Optional[int] = 2,
         # DEPRECATING
         kind: str = "task",
@@ -29,6 +31,8 @@ class instrument:  # pylint: disable=invalid-name
         self.config = config
         self.ignore_inputs = ignore_inputs
         self.ignore_outputs = ignore_outputs
+        self.redact = redact
+        self.redact_on_error = redact_on_error
         self.max_depth = max_depth
 
     def __call__(self, func: Callable[..., Any]):
@@ -109,12 +113,10 @@ class instrument:  # pylint: disable=invalid-name
                 )
 
             _inputs = self._redact(
-                self._parse(
-                    func,
-                    *args,
-                    **kwargs,
-                ),
-                self.ignore_inputs,
+                name=span.name,
+                field="inputs",
+                io=self._parse(func, *args, **kwargs),
+                ignore=self.ignore_inputs,
             )
             span.set_attributes(
                 attributes={"inputs": _inputs},
@@ -153,7 +155,12 @@ class instrument:  # pylint: disable=invalid-name
                 namespace="metrics.unit.tokens",
             )
 
-            _outputs = self._redact(self._patch(result), self.ignore_outputs)
+            _outputs = self._redact(
+                name=span.name,
+                field="outputs",
+                io=self._patch(result),
+                ignore=self.ignore_outputs,
+            )
             span.set_attributes(
                 attributes={"outputs": _outputs},
                 namespace="data",
@@ -192,6 +199,9 @@ class instrument:  # pylint: disable=invalid-name
 
     def _redact(
         self,
+        *,
+        name: str,
+        field: str,
         io: Dict[str, Any],
         ignore: Union[List[str], bool] = False,
     ) -> Dict[str, Any]:
@@ -220,6 +230,20 @@ class instrument:  # pylint: disable=invalid-name
             )
         }
 
+        if self.redact is not None:
+            try:
+                io = self.redact(name, field, io)
+            except:  # pylint: disable=bare-except
+                if self.redact_on_error:
+                    io = {}
+
+        if ag.tracing.redact is not None:
+            try:
+                io = ag.tracing.redact(name, field, io)
+            except:  # pylint: disable=bare-except
+                if ag.tracing.redact_on_error:
+                    io = {}
+
         return io
 
     def _patch(

agenta/sdk/litellm/litellm.py

@@ -89,16 +89,20 @@ def litellm_handler():
                 log.error("LiteLLM callback error: span not found.")
                 return
 
-            result = kwargs.get("complete_streaming_response")
+            try:
+                result = []
+                for choice in response_obj.choices:
+                    message = choice.message.__dict__
+                    result.append(message)
 
-            outputs = (
-                {"__default__": result} if not isinstance(result, dict) else result
-            )
+                outputs = {"completion": result}
+                self.span.set_attributes(
+                    attributes={"outputs": outputs},
+                    namespace="data",
+                )
 
-            self.span.set_attributes(
-                attributes={"outputs": outputs},
-                namespace="data",
-            )
+            except Exception as e:
+                pass
 
             self.span.set_attributes(
                 attributes={"total": kwargs.get("response_cost")},
@@ -194,16 +198,20 @@ def litellm_handler():
                 log.error("LiteLLM callback error: span not found.")
                 return
 
-            result = kwargs.get("complete_streaming_response")
+            try:
+                result = []
+                for choice in response_obj.choices:
+                    message = choice.message.__dict__
+                    result.append(message)
 
-            outputs = (
-                {"__default__": result} if not isinstance(result, dict) else result
-            )
+                outputs = {"completion": result}
+                self.span.set_attributes(
+                    attributes={"outputs": outputs},
+                    namespace="data",
+                )
 
-            self.span.set_attributes(
-                attributes={"outputs": outputs},
-                namespace="data",
-            )
+            except Exception as e:
+                pass
 
             self.span.set_attributes(
                 attributes={"total": kwargs.get("response_cost")},
@@ -236,17 +244,20 @@ def litellm_handler():
                 log.error("LiteLLM callback error: span not found.")
                 return
 
-            # result = kwargs.get("complete_streaming_response")
-            result = response_obj.choices[0].message.content
+            try:
+                result = []
+                for choice in response_obj.choices:
+                    message = choice.message.__dict__
+                    result.append(message)
 
-            outputs = (
-                {"__default__": result} if not isinstance(result, dict) else result
-            )
+                outputs = {"completion": result}
+                self.span.set_attributes(
+                    attributes={"outputs": outputs},
+                    namespace="data",
+                )
 
-            self.span.set_attributes(
-                attributes={"outputs": outputs},
-                namespace="data",
-            )
+            except Exception as e:
+                pass
 
             self.span.set_attributes(
                 attributes={"total": kwargs.get("response_cost")},

agenta/sdk/middleware/auth.py

@@ -21,6 +21,12 @@ AGENTA_SDK_AUTH_CACHE_TTL = environ.get(
     15 * 60,  # 15 minutes
 )
 
+AGENTA_SDK_AUTH_CACHE = str(environ.get("AGENTA_SDK_AUTH_CACHE", True)).lower() in (
+    "true",
+    "1",
+    "t",
+)
+
 AGENTA_UNAUTHORIZED_EXECUTION_ALLOWED = str(
     environ.get("AGENTA_UNAUTHORIZED_EXECUTION_ALLOWED", False)
 ).lower() in ("true", "1", "t")
@@ -89,9 +95,11 @@ class AuthorizationMiddleware(BaseHTTPMiddleware):
                 sort_keys=True,
             )
 
-            cached_policy = cache.get(_hash)
+            policy = None
+            if AGENTA_SDK_AUTH_CACHE:
+                policy = cache.get(_hash)
 
-            if not cached_policy:
+            if not policy:
                 async with httpx.AsyncClient() as client:
                     response = await client.get(
                         f"{self.host}/api/permissions/verify",
@@ -110,19 +118,17 @@ class AuthorizationMiddleware(BaseHTTPMiddleware):
                         cache.put(_hash, {"effect": "deny"})
                         return Deny()
 
-                    cached_policy = {
+                    policy = {
                         "effect": "allow",
                         "credentials": auth.get("credentials"),
                     }
 
-                    cache.put(_hash, cached_policy)
+                    cache.put(_hash, policy)
 
-            if cached_policy.get("effect") == "deny":
+            if policy.get("effect") == "deny":
                 return Deny()
 
-            request.state.credentials = cached_policy.get("credentials")
-
-            print(f"credentials: {request.state.credentials}")
+            request.state.credentials = policy.get("credentials")
 
             return await call_next(request)
 

agenta/sdk/tracing/inline.py

@@ -903,9 +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:
@@ -934,17 +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
 
 
@@ -956,8 +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],
@@ -1005,19 +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
 
 
@@ -1101,6 +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",

agenta/sdk/tracing/tracing.py

@@ -1,4 +1,4 @@
-from typing import Optional, Any, Dict
+from typing import Optional, Any, Dict, Callable
 from enum import Enum
 
 from httpx import get as check
@@ -32,6 +32,8 @@ class Tracing(metaclass=Singleton):
     def __init__(
         self,
         url: str,
+        redact: Optional[Callable[..., Any]] = None,
+        redact_on_error: Optional[bool] = True,
     ) -> None:
         # ENDPOINT (OTLP)
         self.otlp_url = url
@@ -49,6 +51,10 @@ class Tracing(metaclass=Singleton):
         # INLINE SPANS for INLINE TRACES (INLINE PROCESSOR)
         self.inline_spans: Dict[str, Any] = dict()
 
+        # REDACT
+        self.redact = redact
+        self.redact_on_error = redact_on_error
+
     # PUBLIC
 
     def configure(