agenta 0.68.0__py3-none-any.whl → 0.72.4__py3-none-any.whl

agenta/sdk/litellm/litellm.py

@@ -166,24 +166,27 @@ def litellm_handler():
                 namespace="metrics.unit.costs",
             )
 
+            # Handle both dict and object attribute access for usage, and safely handle None
+            usage = getattr(response_obj, "usage", None)
+            if isinstance(usage, dict):
+                prompt_tokens = usage.get("prompt_tokens")
+                completion_tokens = usage.get("completion_tokens")
+                total_tokens = usage.get("total_tokens")
+            elif usage is not None:
+                prompt_tokens = getattr(usage, "prompt_tokens", None)
+                completion_tokens = getattr(usage, "completion_tokens", None)
+                total_tokens = getattr(usage, "total_tokens", None)
+            else:
+                prompt_tokens = completion_tokens = total_tokens = None
+
             span.set_attributes(
                 attributes=(
                     {
-                        "prompt": (
-                            float(response_obj.usage.prompt_tokens)
-                            if response_obj.usage.prompt_tokens
-                            else None
-                        ),
-                        "completion": (
-                            float(response_obj.usage.completion_tokens)
-                            if response_obj.usage.completion_tokens
-                            else None
-                        ),
-                        "total": (
-                            float(response_obj.usage.total_tokens)
-                            if response_obj.usage.total_tokens
-                            else None
-                        ),
+                        "prompt": float(prompt_tokens) if prompt_tokens else None,
+                        "completion": float(completion_tokens)
+                        if completion_tokens
+                        else None,
+                        "total": float(total_tokens) if total_tokens else None,
                     }
                 ),
                 namespace="metrics.unit.tokens",
@@ -300,24 +303,29 @@ def litellm_handler():
                 namespace="metrics.unit.costs",
             )
 
+            # Handle both dict and object attribute access for usage
+            usage = getattr(response_obj, "usage", None)
+            if usage is None:
+                prompt_tokens = None
+                completion_tokens = None
+                total_tokens = None
+            elif isinstance(usage, dict):
+                prompt_tokens = usage.get("prompt_tokens")
+                completion_tokens = usage.get("completion_tokens")
+                total_tokens = usage.get("total_tokens")
+            else:
+                prompt_tokens = getattr(usage, "prompt_tokens", None)
+                completion_tokens = getattr(usage, "completion_tokens", None)
+                total_tokens = getattr(usage, "total_tokens", None)
+
             span.set_attributes(
                 attributes=(
                     {
-                        "prompt": (
-                            float(response_obj.usage.prompt_tokens)
-                            if response_obj.usage.prompt_tokens
-                            else None
-                        ),
-                        "completion": (
-                            float(response_obj.usage.completion_tokens)
-                            if response_obj.usage.completion_tokens
-                            else None
-                        ),
-                        "total": (
-                            float(response_obj.usage.total_tokens)
-                            if response_obj.usage.total_tokens
-                            else None
-                        ),
+                        "prompt": float(prompt_tokens) if prompt_tokens else None,
+                        "completion": float(completion_tokens)
+                        if completion_tokens
+                        else None,
+                        "total": float(total_tokens) if total_tokens else None,
                     }
                 ),
                 namespace="metrics.unit.tokens",

agenta/sdk/middleware/auth.py

@@ -17,14 +17,12 @@ import agenta as ag
 
 log = get_module_logger(__name__)
 
-AGENTA_RUNTIME_PREFIX = getenv("AGENTA_RUNTIME_PREFIX", "")
-
 
 _CACHE_ENABLED = (
     getenv("AGENTA_SERVICE_MIDDLEWARE_CACHE_ENABLED", "true").lower() in TRUTHY
 )
 
-_ALWAYS_ALLOW_LIST = [f"{AGENTA_RUNTIME_PREFIX}/health"]
+_ALWAYS_ALLOW_LIST = ["/health"]
 
 _cache = TTLLRUCache()
 
@@ -64,7 +62,7 @@ class AuthHTTPMiddleware(BaseHTTPMiddleware):
 
     async def dispatch(self, request: Request, call_next: Callable):
         try:
-            if request.url.path in _ALWAYS_ALLOW_LIST:
+            if _strip_service_prefix(request.url.path) in _ALWAYS_ALLOW_LIST:
                 request.state.auth = {}
 
             else:
@@ -253,3 +251,20 @@ class AuthHTTPMiddleware(BaseHTTPMiddleware):
                 status_code=500,
                 content=f"Could not verify credentials: unexpected error - {str(exc)}. Please try again later or contact support if the issue persists.",
             ) from exc
+
+
+def _strip_service_prefix(path: str) -> str:
+    if not path.startswith("/services/"):
+        return path
+
+    parts = path.split("/", 3)
+    if len(parts) < 4:
+        return "/"
+
+    service_name = parts[2]
+    remainder = parts[3]
+
+    if not service_name or not remainder or remainder.startswith("/"):
+        return path
+
+    return f"/{remainder}"

agenta/sdk/middleware/config.py

@@ -224,6 +224,7 @@ class ConfigMiddleware(BaseHTTPMiddleware):
             baggage.get("ag.refs.variant.slug")
             # ALTERNATIVE
             or request.query_params.get("variant_slug")
+            or body.get("variant_slug")
             # LEGACY
             or baggage.get("variant_slug")
             or request.query_params.get("config")
@@ -234,6 +235,7 @@ class ConfigMiddleware(BaseHTTPMiddleware):
             baggage.get("ag.refs.variant.version")
             # ALTERNATIVE
             or request.query_params.get("variant_version")
+            or body.get("variant_version")
             # LEGACY
             or baggage.get("variant_version")
         )
@@ -244,7 +246,7 @@ class ConfigMiddleware(BaseHTTPMiddleware):
         return Reference(
             id=variant_id,
             slug=variant_slug,
-            version=variant_version,
+            version=str(variant_version) if variant_version is not None else None,
         )
 
     async def _parse_environment_ref(

agenta/sdk/middleware/otel.py

@@ -3,13 +3,12 @@ from typing import Callable
 from starlette.middleware.base import BaseHTTPMiddleware
 from fastapi import Request, FastAPI
 
-from opentelemetry.baggage.propagation import W3CBaggagePropagator
-from opentelemetry.trace.propagation.tracecontext import TraceContextTextMapPropagator
-
 from agenta.sdk.utils.exceptions import suppress
 from agenta.sdk.tracing.propagation import extract
-
+from agenta.sdk.utils.exceptions import suppress
 from agenta.sdk.utils.logging import get_module_logger
+from fastapi import FastAPI, Request
+from starlette.middleware.base import BaseHTTPMiddleware
 
 log = get_module_logger(__name__)
 

agenta/sdk/middleware/vault.py

@@ -21,11 +21,9 @@ import agenta as ag
 log = get_module_logger(__name__)
 
 
-AGENTA_RUNTIME_PREFIX = getenv("AGENTA_RUNTIME_PREFIX", "")
-
 _ALWAYS_ALLOW_LIST = [
-    f"{AGENTA_RUNTIME_PREFIX}/health",
-    f"{AGENTA_RUNTIME_PREFIX}/openapi.json",
+    "/health",
+    "/openapi.json",
 ]
 
 _PROVIDER_KINDS = [
@@ -116,7 +114,7 @@ class VaultMiddleware(BaseHTTPMiddleware):
         allow_secrets = True
 
         try:
-            if not request.url.path in _ALWAYS_ALLOW_LIST:
+            if _strip_service_prefix(request.url.path) not in _ALWAYS_ALLOW_LIST:
                 await self._allow_local_secrets(credentials)
 
             for provider_kind in _PROVIDER_KINDS:
@@ -331,3 +329,20 @@ class VaultMiddleware(BaseHTTPMiddleware):
                 status_code=500,
                 content=f"Could not verify credentials: unexpected error - {str(exc)}. Please try again later or contact support if the issue persists.",
             ) from exc
+
+
+def _strip_service_prefix(path: str) -> str:
+    if not path.startswith("/services/"):
+        return path
+
+    parts = path.split("/", 3)
+    if len(parts) < 4:
+        return "/"
+
+    service_name = parts[2]
+    remainder = parts[3]
+
+    if not service_name or not remainder or remainder.startswith("/"):
+        return path
+
+    return f"/{remainder}"

agenta/sdk/middlewares/routing/otel.py

@@ -6,7 +6,7 @@ from fastapi import Request
 
 from agenta.sdk.utils.logging import get_module_logger
 from agenta.sdk.utils.exceptions import suppress
-from agenta.sdk.engines.tracing.propagation import extract
+from agenta.sdk.tracing.propagation import extract
 
 
 log = get_module_logger(__name__)

agenta/sdk/middlewares/running/vault.py

@@ -127,7 +127,7 @@ class VaultMiddleware:
         request: WorkflowServiceRequest,
         call_next: Callable[[WorkflowServiceRequest], Any],
     ):
-        api_url = f"{ag.DEFAULT_AGENTA_SINGLETON_INSTANCE.host}/api"
+        api_url = ag.DEFAULT_AGENTA_SINGLETON_INSTANCE.api_url
 
         with suppress():
             ctx = RunningContext.get()

agenta/sdk/models/shared.py

@@ -88,7 +88,7 @@ class Slug(BaseModel):
     def check_url_safety(cls, v):
         if v is not None:
             if not match(r"^[a-zA-Z0-9_-]+$", v):
-                raise ValueError("slug must be URL-safe.")
+                raise ValueError("'slug' must be URL-safe.")
         return v
 
 

agenta/sdk/tracing/exporters.py

@@ -122,6 +122,7 @@ class OTLPExporter(OTLPSpanExporter):
                     #     "[SPAN]  [EXPORT]",
                     #     trace_id=UUID(int=trace_id).hex,
                     #     span_id=UUID(int=span_id).hex[-16:],
+                    #     span_attributes=_span.attributes,
                     # )
 
                 serialized_spans.append(super().export(_spans))

agenta/sdk/tracing/processors.py

@@ -1,21 +1,21 @@
-from typing import Optional, Dict, List
 from threading import Lock
-from json import dumps
-from uuid import UUID
+from typing import Dict, List, Optional
 
+
+from agenta.sdk.contexts.tracing import TracingContext
+from agenta.sdk.utils.logging import get_module_logger
 from opentelemetry.baggage import get_all as get_baggage
 from opentelemetry.context import Context
 from opentelemetry.sdk.trace import Span, SpanProcessor
 from opentelemetry.sdk.trace.export import (
-    SpanExporter,
-    ReadableSpan,
     BatchSpanProcessor,
+    ReadableSpan,
+    SpanExporter,
 )
 from opentelemetry.trace import SpanContext
 
 from agenta.sdk.utils.logging import get_module_logger
-from agenta.sdk.tracing.conventions import Reference
-
+from agenta.sdk.models.tracing import BaseModel
 from agenta.sdk.contexts.tracing import TracingContext
 
 log = get_module_logger(__name__)
@@ -65,15 +65,38 @@ class TraceProcessor(SpanProcessor):
         # )
 
         for key in self.references.keys():
-            span.set_attribute(f"ag.refs.{key}", self.references[key])
+            ref = self.references[key]
+            if isinstance(ref, BaseModel):
+                try:
+                    ref = ref.model_dump(mode="json", exclude_none=True)
+                except Exception:  # pylint: disable=bare-except
+                    pass
+            if isinstance(ref, dict):
+                for field, value in ref.items():
+                    span.set_attribute(f"ag.refs.{key}.{field}", str(value))
 
         baggage = get_baggage(parent_context)
 
         for key in baggage.keys():
-            if key.startswith("ag.refs."):
-                _key = key.replace("ag.refs.", "")
-                if _key in [_.value for _ in Reference.__members__.values()]:
-                    span.set_attribute(key, baggage[key])
+            if key.startswith("ag."):
+                value = baggage[key]
+
+                if key.startswith("ag.refs."):
+                    ref = value
+                    if isinstance(value, BaseModel):
+                        try:
+                            ref = value.model_dump(mode="json", exclude_none=True)  # type: ignore
+                        except Exception:  # pylint: disable=bare-except
+                            pass
+                    if isinstance(ref, dict):
+                        for field, val in ref.items():
+                            span.set_attribute(f"{key}.{field}", str(val))
+                    elif isinstance(ref, (str, bool, int, float, bytes)):
+                        span.set_attribute(key, ref)
+                else:
+                    # Not a reference - only set if it's a valid attribute type
+                    if isinstance(value, (str, bool, int, float, bytes)):
+                        span.set_attribute(key, value)
 
         context = TracingContext.get()
 
@@ -105,10 +128,11 @@ class TraceProcessor(SpanProcessor):
         if not self.inline:
             if context.links:
                 for key, link in context.links.items():
-                    try:
-                        link = link.model_dump(mode="json", exclude_none=True)
-                    except:  # pylint: disable=bare-except
-                        pass
+                    if isinstance(link, BaseModel):
+                        try:
+                            link = link.model_dump(mode="json", exclude_none=True)
+                        except Exception:
+                            pass
                     if not isinstance(link, dict):
                         continue
                     if not link.get("trace_id") or not link.get("span_id"):
@@ -127,30 +151,14 @@ class TraceProcessor(SpanProcessor):
 
         if context.references:
             for key, ref in context.references.items():
-                try:
-                    ref = ref.model_dump(mode="json", exclude_none=True)
-                except:  # pylint: disable=bare-except
-                    pass
-                if not isinstance(ref, dict):
-                    continue
-                if not ref.get("id") and not ref.get("slug") and not ref.get("version"):
-                    continue
-
-                if ref.get("id"):
-                    span.set_attribute(
-                        f"ag.refs.{key}.id",
-                        str(ref.get("id")),
-                    )
-                if ref.get("slug"):
-                    span.set_attribute(
-                        f"ag.refs.{key}.slug",
-                        str(ref.get("slug")),
-                    )
-                if ref.get("version"):
-                    span.set_attribute(
-                        f"ag.refs.{key}.version",
-                        str(ref.get("version")),
-                    )
+                if isinstance(ref, BaseModel):
+                    try:
+                        ref = ref.model_dump(mode="json", exclude_none=True)
+                    except Exception:
+                        pass
+                if isinstance(ref, dict):
+                    for field, value in ref.items():
+                        span.set_attribute(f"ag.refs.{key}.{field}", str(value))
 
         trace_id = span.context.trace_id
         span_id = span.context.span_id

agenta/sdk/tracing/propagation.py

@@ -1,14 +1,10 @@
-from typing import Tuple, Optional, Dict, Any
+from typing import Any, Dict, Optional, Tuple
 
-from opentelemetry.trace import Span, set_span_in_context, get_current_span
-from opentelemetry.baggage.propagation import W3CBaggagePropagator
-from opentelemetry.trace.propagation.tracecontext import TraceContextTextMapPropagator
+from agenta.sdk.contexts.tracing import TracingContext
 from opentelemetry.baggage import set_baggage
+from opentelemetry.baggage.propagation import W3CBaggagePropagator
 from opentelemetry.context import get_current
-
-from agenta.sdk.contexts.tracing import TracingContext
-
-import agenta as ag
+from opentelemetry.trace.propagation.tracecontext import TraceContextTextMapPropagator
 
 
 def extract(
@@ -47,11 +43,12 @@ def extract(
     baggage = {}
 
     try:
-        _carrier = {
-            "baggage": headers.get("Baggage")  # Uppercase
+        raw_baggage = (
+            headers.get("Baggage")  # Uppercase
             or headers.get("baggage")  # Lowercase
-            or "",
-        }
+            or ""
+        )
+        _carrier = {"baggage": raw_baggage}
 
         _context = W3CBaggagePropagator().extract(_carrier)
 

agenta/sdk/tracing/tracing.py

@@ -31,6 +31,7 @@ from agenta.sdk.tracing.conventions import Reference, is_valid_attribute_key
 from agenta.sdk.tracing.propagation import extract, inject
 from agenta.sdk.utils.cache import TTLLRUCache
 
+import agenta as ag
 
 log = get_module_logger(__name__)
 
@@ -215,6 +216,42 @@ class Tracing(metaclass=Singleton):
                         namespace="metrics",
                     )
 
+    def store_session(
+        self,
+        session_id: Optional[str] = None,
+        span: Optional[Span] = None,
+    ):
+        """Set session attributes on the current span.
+
+        Args:
+            session_id: Unique identifier for the session
+            span: Optional span to set attributes on (defaults to current span)
+        """
+        with suppress():
+            if span is None:
+                span = self.get_current_span()
+
+            if session_id:
+                span.set_attribute("id", session_id, namespace="session")
+
+    def store_user(
+        self,
+        user_id: Optional[str] = None,
+        span: Optional[Span] = None,
+    ):
+        """Set user attributes on the current span.
+
+        Args:
+            user_id: Unique identifier for the user
+            span: Optional span to set attributes on (defaults to current span)
+        """
+        with suppress():
+            if span is None:
+                span = self.get_current_span()
+
+            if user_id:
+                span.set_attribute("id", user_id, namespace="user")
+
     def is_inline_trace_ready(
         self,
         trace_id: Optional[int] = None,
@@ -314,6 +351,58 @@ class Tracing(metaclass=Singleton):
 
         return None
 
+    def get_trace_url(
+        self,
+        trace_id: Optional[str] = None,
+    ) -> str:
+        """
+        Build a URL to view a trace in the Agenta UI.
+
+        Automatically extracts the trace ID from the current tracing context
+        if not explicitly provided.
+
+        Args:
+            trace_id: Optional trace ID (hex string format). If not provided,
+                      it will be automatically extracted from the current trace context.
+
+        Returns:
+            The full URL to view the trace in the observability dashboard
+
+        Raises:
+            RuntimeError: If the SDK is not initialized, no active trace context exists,
+                          or scope info cannot be fetched
+        """
+        if trace_id is None:
+            span_ctx = self.get_span_context()
+            if span_ctx is None or not span_ctx.is_valid:
+                raise RuntimeError(
+                    "No active trace context found. "
+                    "Make sure you call this within an instrumented function or span."
+                )
+
+            trace_id = f"{span_ctx.trace_id:032x}"
+
+        if not ag or not ag.DEFAULT_AGENTA_SINGLETON_INSTANCE:
+            raise RuntimeError(
+                "Agenta SDK is not initialized. Please call ag.init() first."
+            )
+
+        api_url = ag.DEFAULT_AGENTA_SINGLETON_INSTANCE.api_url
+        web_url = api_url.replace("/api", "") if api_url else None
+
+        (organization_id, workspace_id, project_id) = (
+            ag.DEFAULT_AGENTA_SINGLETON_INSTANCE.resolve_scopes()
+        )
+
+        if not web_url or not workspace_id or not project_id:
+            raise RuntimeError(
+                "Could not determine workspace/project context. Please call ag.init() first."
+            )
+
+        return (
+            f"{web_url}/w/{workspace_id}/p/{project_id}/observability?trace={trace_id}"
+        )
+
 
 def get_tracer(
     tracing: Tracing,

agenta/sdk/types.py

@@ -8,7 +8,7 @@ from pydantic import BaseModel, Field, model_validator, AliasChoices
 from starlette.responses import StreamingResponse
 
 
-from agenta.sdk.assets import supported_llm_models
+from agenta.sdk.assets import supported_llm_models, model_metadata
 from agenta.client.backend.types import AgentaNodesResponse, AgentaNodeDto
 
 
@@ -23,7 +23,11 @@ def MCField(  # pylint: disable=invalid-name
 ) -> Field:
     # Pydantic 2.12+ no longer allows post-creation mutation of field properties
     if isinstance(choices, dict):
-        json_extra = {"choices": choices, "x-parameter": "grouped_choice"}
+        json_extra = {
+            "choices": choices,
+            "x-parameter": "grouped_choice",
+            "x-model-metadata": model_metadata,
+        }
     elif isinstance(choices, list):
         json_extra = {"choices": choices, "x-parameter": "choice"}
     else:

agenta/sdk/workflows/runners/daytona.py

@@ -88,7 +88,6 @@ class DaytonaRunner(CodeRunner):
                 target=target,
             )
             self.daytona = Daytona(config)
-            # log.debug("Daytona client initialized")
 
         except Exception as e:
             raise RuntimeError(f"Failed to initialize Daytona client: {e}")
@@ -107,8 +106,6 @@ class DaytonaRunner(CodeRunner):
                     "Set it to the Daytona sandbox ID or snapshot name you want to use."
                 )
 
-            # log.debug(f"Creating sandbox from snapshot: {snapshot_id}")
-
             from daytona import CreateSandboxFromSnapshotParams
 
             sandbox = self.daytona.create(
@@ -118,9 +115,6 @@ class DaytonaRunner(CodeRunner):
                 )
             )
 
-            # log.debug(
-            #     f"Sandbox created: {sandbox.id if hasattr(sandbox, 'id') else sandbox}"
-            # )
             return sandbox
 
         except Exception as e:

agenta/sdk/workflows/runners/registry.py

@@ -1,7 +1,17 @@
 import os
+from typing import TYPE_CHECKING
+
 from agenta.sdk.workflows.runners.base import CodeRunner
 from agenta.sdk.workflows.runners.local import LocalRunner
-from agenta.sdk.workflows.runners.daytona import DaytonaRunner
+
+if TYPE_CHECKING:
+    from agenta.sdk.workflows.runners.daytona import DaytonaRunner
+
+
+def _get_daytona_runner() -> "DaytonaRunner":
+    from agenta.sdk.workflows.runners.daytona import DaytonaRunner
+
+    return DaytonaRunner()
 
 
 def get_runner() -> CodeRunner:
@@ -21,7 +31,14 @@ def get_runner() -> CodeRunner:
     runner_type = os.getenv("AGENTA_SERVICES_SANDBOX_RUNNER", "local").lower()
 
     if runner_type == "daytona":
-        return DaytonaRunner()
+        try:
+            return _get_daytona_runner()
+        except ImportError as exc:
+            raise ValueError(
+                "Daytona runner requires the 'daytona' package. "
+                "Install optional dependencies or set "
+                "AGENTA_SERVICES_SANDBOX_RUNNER=local."
+            ) from exc
     elif runner_type == "local":
         return LocalRunner()
     else:

{agenta-0.68.0.dist-info → agenta-0.72.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agenta
-Version: 0.68.0
+Version: 0.72.4
 Summary: The SDK for agenta is an open-source LLMOps platform.
 Keywords: LLMOps,LLM,evaluation,prompt engineering
 Author: Mahmoud Mabrouk
@@ -16,7 +16,7 @@ Classifier: Programming Language :: Python :: 3.14
 Classifier: Programming Language :: Python :: 3.9
 Classifier: Topic :: Software Development :: Libraries
 Requires-Dist: daytona (>=0.121.0,<0.122.0)
-Requires-Dist: fastapi (>=0.122,<0.123)
+Requires-Dist: fastapi (>=0.125)
 Requires-Dist: httpx (>=0.28,<0.29)
 Requires-Dist: importlib-metadata (>=8,<9)
 Requires-Dist: jinja2 (>=3,<4)
@@ -26,6 +26,7 @@ Requires-Dist: opentelemetry-api (>=1,<2)
 Requires-Dist: opentelemetry-exporter-otlp-proto-http (>=1,<2)
 Requires-Dist: opentelemetry-instrumentation (>=0.59b0,<0.60)
 Requires-Dist: opentelemetry-sdk (>=1,<2)
+Requires-Dist: orjson (>=3,<4)
 Requires-Dist: pydantic (>=2,<3)
 Requires-Dist: python-dotenv (>=1,<2)
 Requires-Dist: python-jsonpath (>=2,<3)