diff --git a/agenta-cli/agenta/__init__.py b/agenta-cli/agenta/__init__.py
index 49ab103b87..7108ba4cd9 100644
--- a/agenta-cli/agenta/__init__.py
+++ b/agenta-cli/agenta/__init__.py
@@ -14,7 +14,8 @@
 )
 from .sdk.tracing.decorators import span
 from .sdk.utils.preinit import PreInitObject
-from .sdk.agenta_init import Config, init, tracing
+from .sdk.agenta_init import Config, init, llm_tracing
 from .sdk.utils.helper.openai_cost import calculate_token_usage
 
+
 config = PreInitObject("agenta.config", Config)
diff --git a/agenta-cli/agenta/sdk/__init__.py b/agenta-cli/agenta/sdk/__init__.py
index 1d8e6a69ba..252feff591 100644
--- a/agenta-cli/agenta/sdk/__init__.py
+++ b/agenta-cli/agenta/sdk/__init__.py
@@ -15,7 +15,8 @@
     BinaryParam,
 )
 from .tracing.decorators import span
-from .agenta_init import Config, init, tracing
+from .agenta_init import Config, init, llm_tracing
 from .utils.helper.openai_cost import calculate_token_usage
 
+
 config = PreInitObject("agenta.config", Config)
diff --git a/agenta-cli/agenta/sdk/agenta_decorator.py b/agenta-cli/agenta/sdk/agenta_decorator.py
index eb5fa0f4b4..b5c08dfec5 100644
--- a/agenta-cli/agenta/sdk/agenta_decorator.py
+++ b/agenta-cli/agenta/sdk/agenta_decorator.py
@@ -73,12 +73,15 @@ def entrypoint(func: Callable[..., Any]) -> Callable[..., Any]:
     config_params = agenta.config.all()
     ingestible_files = extract_ingestible_files(func_signature)
 
+    # Initialize tracing
+    tracing = agenta.llm_tracing()
+
     @functools.wraps(func)
     async def wrapper(*args, **kwargs) -> Any:
         func_params, api_config_params = split_kwargs(kwargs, config_params)
 
         # Start tracing
-        agenta.tracing.trace(
+        tracing.trace(
             trace_name=func.__name__,
             inputs=func_params,
             config=api_config_params,
diff --git a/agenta-cli/agenta/sdk/agenta_init.py b/agenta-cli/agenta/sdk/agenta_init.py
index 0d707ea594..5b1302bab9 100644
--- a/agenta-cli/agenta/sdk/agenta_init.py
+++ b/agenta-cli/agenta/sdk/agenta_init.py
@@ -240,6 +240,3 @@ def llm_tracing(max_workers: Optional[int] = None) -> Tracing:
         api_key=singleton.api_key,
         max_workers=max_workers,
     )
-
-
-tracing = llm_tracing()
diff --git a/agenta-cli/agenta/sdk/tracing/decorators.py b/agenta-cli/agenta/sdk/tracing/decorators.py
index 84dfb607bf..af799d5876 100644
--- a/agenta-cli/agenta/sdk/tracing/decorators.py
+++ b/agenta-cli/agenta/sdk/tracing/decorators.py
@@ -9,15 +9,17 @@
 def span(type: str):
     """Decorator to automatically start and end spans."""
 
+    tracing = ag.llm_tracing()
+
     def decorator(func):
         @wraps(func)
         async def wrapper(*args, **kwargs):
             result = None
-            span = ag.tracing.start_span(
+            span = tracing.start_span(
                 func.__name__,
                 input=kwargs,
                 type=type,
-                trace_id=ag.tracing.active_trace,
+                trace_id=tracing.active_trace,
             )
             try:
                 is_coroutine_function = inspect.iscoroutinefunction(func)
@@ -33,7 +35,7 @@ async def wrapper(*args, **kwargs):
             finally:
                 if not isinstance(result, dict):
                     result = {"message": result}
-                ag.tracing.end_span(output=result, span=span)
+                tracing.end_span(output=result, span=span)
             return result
 
         return wrapper
diff --git a/examples/app_with_observability/app_async.py b/examples/app_with_observability/app_async.py
index cf423750e4..6f0612c15d 100644
--- a/examples/app_with_observability/app_async.py
+++ b/examples/app_with_observability/app_async.py
@@ -8,6 +8,7 @@
 )
 
 ag.init()
+tracing = ag.llm_tracing()
 ag.config.default(
     temperature=ag.FloatParam(0.2), prompt_template=ag.TextParam(default_prompt)
 )
@@ -18,7 +19,7 @@ async def llm_call(prompt):
     chat_completion = await client.chat.completions.create(
         model="gpt-3.5-turbo", messages=[{"role": "user", "content": prompt}]
     )
-    ag.tracing.set_span_attribute(
+    tracing.set_span_attribute(
         "model_config", {"model": "gpt-3.5-turbo", "temperature": ag.config.temperature}
     )  # translate to {"model_config": {"model": "gpt-3.5-turbo", "temperature": 0.2}}
     tokens_usage = chat_completion.usage.dict()
diff --git a/examples/app_with_observability/app_nested_async.py b/examples/app_with_observability/app_nested_async.py
index 291796df10..fe94789e78 100644
--- a/examples/app_with_observability/app_nested_async.py
+++ b/examples/app_with_observability/app_nested_async.py
@@ -19,6 +19,7 @@
 ]
 
 ag.init()
+tracing = ag.llm_tracing()
 ag.config.default(
     temperature_1=ag.FloatParam(default=1, minval=0.0, maxval=2.0),
     model_1=ag.MultipleChoiceParam("gpt-3.5-turbo", CHAT_LLM_GPT),
@@ -56,7 +57,7 @@ async def llm_call(
         frequency_penalty=frequency_penalty,
         presence_penalty=presence_penalty,
     )
-    ag.tracing.set_span_attribute(
+    tracing.set_span_attribute(
         "model_config", {"model": model, "temperature": temperature}
     )
     tokens_usage = response.usage.dict()  # type: ignore