deepeval 3.7.4__tar.gz → 3.7.5__tar.gz

{deepeval-3.7.4 → deepeval-3.7.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: deepeval
-Version: 3.7.4
+Version: 3.7.5
 Summary: The LLM Evaluation Framework
 Home-page: https://github.com/confident-ai/deepeval
 License: Apache-2.0
@@ -13,13 +13,10 @@ Classifier: Programming Language :: Python :: 3.9
 Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Requires-Dist: aiohttp
-Requires-Dist: anthropic
 Requires-Dist: click (>=8.0.0,<8.3.0)
-Requires-Dist: google-genai (>=1.9.0,<2.0.0)
 Requires-Dist: grpcio (>=1.67.1,<2.0.0)
 Requires-Dist: jinja2
 Requires-Dist: nest_asyncio
-Requires-Dist: ollama
 Requires-Dist: openai
 Requires-Dist: opentelemetry-api (>=1.24.0,<2.0.0)
 Requires-Dist: opentelemetry-exporter-otlp-proto-grpc (>=1.24.0,<2.0.0)

deepeval-3.7.5/deepeval/_version.py

@@ -0,0 +1 @@
+__version__: str = "3.7.5"

{deepeval-3.7.4 → deepeval-3.7.5}/deepeval/dataset/golden.py

@@ -1,6 +1,6 @@
-from pydantic import BaseModel, Field, PrivateAttr
+from pydantic import BaseModel, Field, PrivateAttr, model_validator
 from typing import Optional, Dict, List
-from deepeval.test_case import ToolCall, Turn
+from deepeval.test_case import ToolCall, Turn, MLLMImage
 
 
 class Golden(BaseModel):
@@ -32,10 +32,40 @@ class Golden(BaseModel):
     custom_column_key_values: Optional[Dict[str, str]] = Field(
         default=None, serialization_alias="customColumnKeyValues"
     )
+    multimodal: bool = Field(False, exclude=True)
     _dataset_rank: Optional[int] = PrivateAttr(default=None)
     _dataset_alias: Optional[str] = PrivateAttr(default=None)
     _dataset_id: Optional[str] = PrivateAttr(default=None)
 
+    @model_validator(mode="after")
+    def set_is_multimodal(self):
+        import re
+
+        if self.multimodal is True:
+            return self
+
+        pattern = r"\[DEEPEVAL:IMAGE:(.*?)\]"
+        self.multimodal = (
+            any(
+                [
+                    (
+                        re.search(pattern, self.input) is not None
+                        if self.input
+                        else False
+                    ),
+                    (
+                        re.search(pattern, self.actual_output) is not None
+                        if self.actual_output
+                        else False
+                    ),
+                ]
+            )
+            if isinstance(self.input, str)
+            else self.multimodal
+        )
+
+        return self
+
 
 class ConversationalGolden(BaseModel):
     scenario: str
@@ -55,6 +85,28 @@ class ConversationalGolden(BaseModel):
         default=None, serialization_alias="customColumnKeyValues"
     )
     turns: Optional[List[Turn]] = Field(default=None)
+    multimodal: bool = Field(False, exclude=True)
     _dataset_rank: Optional[int] = PrivateAttr(default=None)
     _dataset_alias: Optional[str] = PrivateAttr(default=None)
     _dataset_id: Optional[str] = PrivateAttr(default=None)
+
+    @model_validator(mode="after")
+    def set_is_multimodal(self):
+        import re
+
+        if self.multimodal is True:
+            return self
+
+        pattern = r"\[DEEPEVAL:IMAGE:(.*?)\]"
+        self.multimodal = (
+            any(
+                [
+                    re.search(pattern, turn.content) is not None
+                    for turn in self.turns
+                ]
+            )
+            if self.turns
+            else self.multimodal
+        )
+
+        return self

{deepeval-3.7.4 → deepeval-3.7.5}/deepeval/evaluate/evaluate.py

@@ -54,7 +54,6 @@ from deepeval.metrics.indicator import (
 from deepeval.test_case import (
     LLMTestCase,
     ConversationalTestCase,
-    MLLMTestCase,
 )
 from deepeval.test_run import (
     global_test_run_manager,
@@ -71,9 +70,7 @@ from deepeval.evaluate.execute import (
 
 
 def assert_test(
-    test_case: Optional[
-        Union[LLMTestCase, ConversationalTestCase, MLLMTestCase]
-    ] = None,
+    test_case: Optional[Union[LLMTestCase, ConversationalTestCase]] = None,
     metrics: Optional[
         Union[
             List[BaseMetric],
@@ -175,7 +172,7 @@ def assert_test(
                 try:
                     if not metric_data.success:
                         failed_metrics_data.append(metric_data)
-                except:
+                except Exception:
                     failed_metrics_data.append(metric_data)
 
         failed_metrics_str = ", ".join(
@@ -188,9 +185,7 @@ def assert_test(
 
 
 def evaluate(
-    test_cases: Union[
-        List[LLMTestCase], List[ConversationalTestCase], List[MLLMTestCase]
-    ],
+    test_cases: Union[List[LLMTestCase], List[ConversationalTestCase]],
     metrics: Optional[
         Union[
             List[BaseMetric],
@@ -272,6 +267,19 @@ def evaluate(
         test_run.hyperparameters = process_hyperparameters(hyperparameters)
         test_run.prompts = process_prompts(hyperparameters)
         global_test_run_manager.save_test_run(TEMP_FILE_PATH)
+
+        # In CLI mode (`deepeval test run`), the CLI owns finalization and will
+        # call `wrap_up_test_run()` once after pytest finishes. Finalizing here
+        # as well would double finalize the run and consequently result in
+        # duplicate uploads / local saves and temp file races, so only
+        # do it when we're NOT in CLI mode.
+        if get_is_running_deepeval():
+            return EvaluationResult(
+                test_results=test_results,
+                confident_link=None,
+                test_run_id=None,
+            )
+
         res = global_test_run_manager.wrap_up_test_run(
             run_duration, display_table=False
         )

{deepeval-3.7.4 → deepeval-3.7.5}/deepeval/evaluate/execute.py

@@ -58,6 +58,13 @@ from deepeval.metrics import (
     BaseConversationalMetric,
     BaseMultimodalMetric,
     TaskCompletionMetric,
+    # RAG metrics that support both single-turn and multimodal
+    ContextualPrecisionMetric,
+    ContextualRecallMetric,
+    ContextualRelevancyMetric,
+    AnswerRelevancyMetric,
+    FaithfulnessMetric,
+    ToolCorrectnessMetric,
 )
 from deepeval.metrics.indicator import (
     measure_metrics_with_indicator,
@@ -70,7 +77,6 @@ from deepeval.models.retry_policy import (
 from deepeval.test_case import (
     LLMTestCase,
     ConversationalTestCase,
-    MLLMTestCase,
 )
 from deepeval.test_case.api import create_api_test_case
 from deepeval.test_run import (
@@ -110,6 +116,15 @@ from deepeval.test_run.hyperparameters import (
 
 logger = logging.getLogger(__name__)
 
+MLLM_SUPPORTED_METRICS = [
+    ContextualPrecisionMetric,
+    ContextualRecallMetric,
+    ContextualRelevancyMetric,
+    AnswerRelevancyMetric,
+    FaithfulnessMetric,
+    ToolCorrectnessMetric,
+]
+
 
 def _skip_metrics_for_error(
     span: Optional[BaseSpan] = None,
@@ -263,9 +278,7 @@ async def _await_with_outer_deadline(obj, *args, timeout: float, **kwargs):
 
 
 def execute_test_cases(
-    test_cases: Union[
-        List[LLMTestCase], List[ConversationalTestCase], List[MLLMTestCase]
-    ],
+    test_cases: Union[List[LLMTestCase], List[ConversationalTestCase]],
     metrics: Union[
         List[BaseMetric],
         List[BaseConversationalMetric],
@@ -307,6 +320,8 @@ def execute_test_cases(
         metric.async_mode = False
         if isinstance(metric, BaseMetric):
             llm_metrics.append(metric)
+            if type(metric) in MLLM_SUPPORTED_METRICS:
+                mllm_metrics.append(metric)
         elif isinstance(metric, BaseConversationalMetric):
             conversational_metrics.append(metric)
         elif isinstance(metric, BaseMultimodalMetric):
@@ -325,12 +340,12 @@ def execute_test_cases(
         )
         for i, test_case in enumerate(test_cases):
             # skip what we know we won't run
-            if isinstance(test_case, LLMTestCase):
+            if isinstance(test_case, LLMTestCase) and not test_case.multimodal:
                 if not llm_metrics:
                     update_pbar(progress, pbar_id)
                     continue
                 per_case_total = len(llm_metrics)
-            elif isinstance(test_case, MLLMTestCase):
+            elif isinstance(test_case, LLMTestCase) and test_case.multimodal:
                 if not mllm_metrics:
                     update_pbar(progress, pbar_id)
                     continue
@@ -349,10 +364,16 @@ def execute_test_cases(
 
             metrics_for_case = (
                 llm_metrics
-                if isinstance(test_case, LLMTestCase)
+                if (
+                    isinstance(test_case, LLMTestCase)
+                    and not test_case.multimodal
+                )
                 else (
                     mllm_metrics
-                    if isinstance(test_case, MLLMTestCase)
+                    if (
+                        isinstance(test_case, LLMTestCase)
+                        and test_case.multimodal
+                    )
                     else conversational_metrics
                 )
             )
@@ -360,10 +381,16 @@ def execute_test_cases(
                 test_case=test_case,
                 index=(
                     llm_test_case_count + 1
-                    if isinstance(test_case, LLMTestCase)
+                    if (
+                        isinstance(test_case, LLMTestCase)
+                        and not test_case.multimodal
+                    )
                     else (
                         mllm_test_case_count + 1
-                        if isinstance(test_case, MLLMTestCase)
+                        if (
+                            isinstance(test_case, LLMTestCase)
+                            and test_case.multimodal
+                        )
                         else conversational_test_case_count + 1
                     )
                 ),
@@ -383,7 +410,10 @@ def execute_test_cases(
                         for metric in metrics:
                             metric.error = None  # Reset metric error
 
-                        if isinstance(test_case, LLMTestCase):
+                        if (
+                            isinstance(test_case, LLMTestCase)
+                            and not test_case.multimodal
+                        ):
                             llm_test_case_count += 1
                             cached_test_case = None
                             if cache_config.use_cache:
@@ -436,7 +466,10 @@ def execute_test_cases(
                                 update_pbar(progress, pbar_test_case_id)
 
                         # No caching and not sending test cases to Confident AI for multimodal metrics yet
-                        elif isinstance(test_case, MLLMTestCase):
+                        elif (
+                            isinstance(test_case, LLMTestCase)
+                            and test_case.multimodal
+                        ):
                             mllm_test_case_count += 1
                             for metric in mllm_metrics:
                                 current_index = index_of[id(metric)]
@@ -560,9 +593,7 @@ def execute_test_cases(
 
 
 async def a_execute_test_cases(
-    test_cases: Union[
-        List[LLMTestCase], List[ConversationalTestCase], List[MLLMTestCase]
-    ],
+    test_cases: Union[List[LLMTestCase], List[ConversationalTestCase]],
     metrics: Union[
         List[BaseMetric],
         List[BaseConversationalMetric],
@@ -605,6 +636,8 @@ async def a_execute_test_cases(
     for metric in metrics:
         if isinstance(metric, BaseMetric):
             llm_metrics.append(metric)
+            if type(metric) in MLLM_SUPPORTED_METRICS:
+                mllm_metrics.append(metric)
         elif isinstance(metric, BaseMultimodalMetric):
             mllm_metrics.append(metric)
         elif isinstance(metric, BaseConversationalMetric):
@@ -613,7 +646,7 @@ async def a_execute_test_cases(
     llm_test_case_counter = -1
     mllm_test_case_counter = -1
     conversational_test_case_counter = -1
-    test_results: List[Union[TestResult, MLLMTestCase]] = []
+    test_results: List[Union[TestResult, LLMTestCase]] = []
     tasks = []
 
     if display_config.show_indicator and _use_bar_indicator:
@@ -632,7 +665,10 @@ async def a_execute_test_cases(
         with progress:
             for test_case in test_cases:
                 with capture_evaluation_run("test case"):
-                    if isinstance(test_case, LLMTestCase):
+                    if (
+                        isinstance(test_case, LLMTestCase)
+                        and not test_case.multimodal
+                    ):
                         if len(llm_metrics) == 0:
                             update_pbar(progress, pbar_id)
                             continue
@@ -660,7 +696,10 @@ async def a_execute_test_cases(
                         )
                         tasks.append(asyncio.create_task(task))
 
-                    elif isinstance(test_case, MLLMTestCase):
+                    elif (
+                        isinstance(test_case, LLMTestCase)
+                        and test_case.multimodal
+                    ):
                         mllm_test_case_counter += 1
                         copied_multimodal_metrics: List[
                             BaseMultimodalMetric
@@ -724,7 +763,10 @@ async def a_execute_test_cases(
     else:
         for test_case in test_cases:
             with capture_evaluation_run("test case"):
-                if isinstance(test_case, LLMTestCase):
+                if (
+                    isinstance(test_case, LLMTestCase)
+                    and not test_case.multimodal
+                ):
                     if len(llm_metrics) == 0:
                         continue
                     llm_test_case_counter += 1
@@ -772,7 +814,9 @@ async def a_execute_test_cases(
                     )
                     tasks.append(asyncio.create_task((task)))
 
-                elif isinstance(test_case, MLLMTestCase):
+                elif (
+                    isinstance(test_case, LLMTestCase) and test_case.multimodal
+                ):
                     mllm_test_case_counter += 1
                     copied_multimodal_metrics: List[BaseMultimodalMetric] = (
                         copy_metrics(mllm_metrics)
@@ -815,7 +859,7 @@ async def _a_execute_llm_test_cases(
     metrics: List[BaseMetric],
     test_case: LLMTestCase,
     test_run_manager: TestRunManager,
-    test_results: List[Union[TestResult, MLLMTestCase]],
+    test_results: List[Union[TestResult, LLMTestCase]],
     count: int,
     test_run: TestRun,
     ignore_errors: bool,
@@ -934,9 +978,9 @@ async def _a_execute_llm_test_cases(
 
 async def _a_execute_mllm_test_cases(
     metrics: List[BaseMultimodalMetric],
-    test_case: MLLMTestCase,
+    test_case: LLMTestCase,
     test_run_manager: TestRunManager,
-    test_results: List[Union[TestResult, MLLMTestCase]],
+    test_results: List[Union[TestResult, LLMTestCase]],
     count: int,
     ignore_errors: bool,
     skip_on_missing_params: bool,
@@ -1013,7 +1057,7 @@ async def _a_execute_conversational_test_cases(
     ],
     test_case: ConversationalTestCase,
     test_run_manager: TestRunManager,
-    test_results: List[Union[TestResult, MLLMTestCase]],
+    test_results: List[Union[TestResult, LLMTestCase]],
     count: int,
     ignore_errors: bool,
     skip_on_missing_params: bool,
@@ -1776,7 +1820,7 @@ async def a_execute_agentic_test_cases(
 async def _a_execute_agentic_test_case(
     golden: Golden,
     test_run_manager: TestRunManager,
-    test_results: List[Union[TestResult, MLLMTestCase]],
+    test_results: List[Union[TestResult, LLMTestCase]],
     count: int,
     verbose_mode: Optional[bool],
     ignore_errors: bool,
@@ -3205,7 +3249,7 @@ async def _evaluate_test_case_pairs(
 
 def _execute_metric(
     metric: BaseMetric,
-    test_case: Union[LLMTestCase, ConversationalTestCase, MLLMTestCase],
+    test_case: Union[LLMTestCase, ConversationalTestCase],
     show_metric_indicator: bool,
     in_component: bool,
     error_config: ErrorConfig,

{deepeval-3.7.4 → deepeval-3.7.5}/deepeval/evaluate/utils.py

@@ -16,7 +16,6 @@ from deepeval.metrics import (
 from deepeval.test_case import (
     LLMTestCase,
     ConversationalTestCase,
-    MLLMTestCase,
 )
 from deepeval.test_run import (
     LLMApiTestCase,
@@ -129,17 +128,14 @@ def create_test_result(
             turns=api_test_case.turns,
         )
     else:
-        multimodal = (
-            api_test_case.multimodal_input is not None
-            and api_test_case.multimodal_input_actual_output is not None
-        )
+        multimodal = api_test_case.images_mapping
         if multimodal:
             return TestResult(
                 name=name,
                 success=api_test_case.success,
                 metrics_data=api_test_case.metrics_data,
-                input=api_test_case.multimodal_input,
-                actual_output=api_test_case.multimodal_input_actual_output,
+                input=api_test_case.input,
+                actual_output=api_test_case.actual_output,
                 conversational=False,
                 multimodal=True,
                 additional_metadata=api_test_case.additional_metadata,
@@ -222,9 +218,9 @@ def validate_assert_test_inputs(
         )
 
     if test_case and metrics:
-        if isinstance(test_case, LLMTestCase) and not all(
-            isinstance(metric, BaseMetric) for metric in metrics
-        ):
+        if (
+            isinstance(test_case, LLMTestCase) and not test_case.multimodal
+        ) and not all(isinstance(metric, BaseMetric) for metric in metrics):
             raise ValueError(
                 "All 'metrics' for an 'LLMTestCase' must be instances of 'BaseMetric' only."
             )
@@ -234,11 +230,17 @@ def validate_assert_test_inputs(
             raise ValueError(
                 "All 'metrics' for an 'ConversationalTestCase' must be instances of 'BaseConversationalMetric' only."
             )
-        if isinstance(test_case, MLLMTestCase) and not all(
-            isinstance(metric, BaseMultimodalMetric) for metric in metrics
+        if (
+            isinstance(test_case, LLMTestCase) and test_case.multimodal
+        ) and not all(
+            (
+                isinstance(metric, BaseMultimodalMetric)
+                or isinstance(metric, BaseMetric)
+            )
+            for metric in metrics
         ):
             raise ValueError(
-                "All 'metrics' for an 'MLLMTestCase' must be instances of 'BaseMultimodalMetric' only."
+                "All 'metrics' for multi-modal LLMTestCase must be instances of 'BaseMultimodalMetric' only."
             )
 
     if not ((golden and observed_callback) or (test_case and metrics)):
@@ -251,9 +253,7 @@ def validate_evaluate_inputs(
     goldens: Optional[List] = None,
     observed_callback: Optional[Callable] = None,
     test_cases: Optional[
-        Union[
-            List[LLMTestCase], List[ConversationalTestCase], List[MLLMTestCase]
-        ]
+        Union[List[LLMTestCase], List[ConversationalTestCase]]
     ] = None,
     metrics: Optional[
         Union[
@@ -292,9 +292,10 @@ def validate_evaluate_inputs(
     if test_cases and metrics:
         for test_case in test_cases:
             for metric in metrics:
-                if isinstance(test_case, LLMTestCase) and not isinstance(
-                    metric, BaseMetric
-                ):
+                if (
+                    isinstance(test_case, LLMTestCase)
+                    and not test_case.multimodal
+                ) and not isinstance(metric, BaseMetric):
                     raise ValueError(
                         f"Metric {metric.__name__} is not a valid metric for LLMTestCase."
                     )
@@ -305,11 +306,14 @@ def validate_evaluate_inputs(
                     raise ValueError(
                         f"Metric {metric.__name__} is not a valid metric for ConversationalTestCase."
                     )
-                if isinstance(test_case, MLLMTestCase) and not isinstance(
-                    metric, BaseMultimodalMetric
+                if (
+                    isinstance(test_case, LLMTestCase) and test_case.multimodal
+                ) and not (
+                    isinstance(metric, BaseMultimodalMetric)
+                    or isinstance(metric, BaseMetric)
                 ):
                     raise ValueError(
-                        f"Metric {metric.__name__} is not a valid metric for MLLMTestCase."
+                        f"Metric {metric.__name__} is not a valid metric for multi-modal LLMTestCase."
                     )
 
 

deepeval-3.7.5/deepeval/integrations/pydantic_ai/agent.py

@@ -0,0 +1,38 @@
+import warnings
+from typing import TYPE_CHECKING, Any
+
+try:
+    from pydantic_ai.agent import Agent as _BaseAgent
+
+    is_pydantic_ai_installed = True
+except ImportError:
+    is_pydantic_ai_installed = False
+
+    class _BaseAgent:
+        """Dummy fallback so imports don't crash when pydantic-ai is missing."""
+
+        def __init__(self, *args: Any, **kwargs: Any) -> None:
+            # No-op: for compatibility
+            pass
+
+
+if TYPE_CHECKING:
+    # For type checkers: use the real Agent if available.
+    from pydantic_ai.agent import Agent  # type: ignore[unused-ignore]
+else:
+    # At runtime we always have some base: real Agent or our dummy.
+    # This is just to avoid blow-ups.
+    Agent = _BaseAgent
+
+
+class DeepEvalPydanticAIAgent(Agent):
+
+    def __init__(self, *args, **kwargs):
+        warnings.warn(
+            "instrument_pydantic_ai is deprecated and will be removed in a future version. "
+            "Please use the new ConfidentInstrumentationSettings instead. Docs: https://www.confident-ai.com/docs/integrations/third-party/pydantic-ai",
+            DeprecationWarning,
+            stacklevel=2,
+        )
+
+        super().__init__(*args, **kwargs)

{deepeval-3.7.4 → deepeval-3.7.5}/deepeval/integrations/pydantic_ai/instrumentator.py

@@ -1,40 +1,58 @@
+from __future__ import annotations
+
 import json
 import logging
 import os
 from time import perf_counter
-from typing import Literal, Optional, List
+from typing import Any, List, Optional, TYPE_CHECKING
 
 from deepeval.config.settings import get_settings
 from deepeval.confident.api import get_confident_api_key
 from deepeval.metrics.base_metric import BaseMetric
 from deepeval.prompt import Prompt
 from deepeval.tracing.context import current_trace_context
-from deepeval.tracing.types import Trace
-from deepeval.tracing.otel.utils import to_hex_string
-from deepeval.tracing.tracing import trace_manager
-from deepeval.tracing.otel.utils import normalize_pydantic_ai_messages
 from deepeval.tracing.otel.exporter import ConfidentSpanExporter
-
+from deepeval.tracing.otel.test_exporter import test_exporter
+from deepeval.tracing.otel.utils import (
+    normalize_pydantic_ai_messages,
+    to_hex_string,
+)
+from deepeval.tracing.perf_epoch_bridge import init_clock_bridge
+from deepeval.tracing.tracing import trace_manager
+from deepeval.tracing.types import (
+    AgentSpan,
+    Trace,
+    TraceSpanStatus,
+    ToolCall,
+)
 
 logger = logging.getLogger(__name__)
 
-
 try:
-    from pydantic_ai.models.instrumented import InstrumentationSettings
-    from opentelemetry.sdk.trace import SpanProcessor, TracerProvider
+    # Optional dependencies
+    from opentelemetry.sdk.trace import (
+        ReadableSpan as _ReadableSpan,
+        SpanProcessor as _SpanProcessor,
+        TracerProvider,
+    )
     from opentelemetry.sdk.trace.export import BatchSpanProcessor
     from opentelemetry.exporter.otlp.proto.http.trace_exporter import (
         OTLPSpanExporter,
     )
-    from opentelemetry.sdk.trace import ReadableSpan
+    from pydantic_ai.models.instrumented import (
+        InstrumentationSettings as _BaseInstrumentationSettings,
+    )
 
     dependency_installed = True
 except ImportError as e:
+    dependency_installed = False
+
+    # Preserve previous behavior: only log when verbose mode is enabled.
     if get_settings().DEEPEVAL_VERBOSE_MODE:
         if isinstance(e, ModuleNotFoundError):
             logger.warning(
                 "Optional tracing dependency not installed: %s",
-                e.name,
+                getattr(e, "name", repr(e)),
                 stacklevel=2,
             )
         else:
@@ -43,26 +61,47 @@ except ImportError as e:
                 e,
                 stacklevel=2,
             )
-    dependency_installed = False
+
+    # Dummy fallbacks so imports and class definitions don't crash when
+    # optional deps are missing. Actual use is still guarded by
+    # is_dependency_installed().
+    class _BaseInstrumentationSettings:
+        def __init__(self, *args: Any, **kwargs: Any) -> None:
+            pass
+
+    class _SpanProcessor:
+        def __init__(self, *args: Any, **kwargs: Any) -> None:
+            pass
+
+        def on_start(self, span: Any, parent_context: Any) -> None:
+            pass
+
+        def on_end(self, span: Any) -> None:
+            pass
+
+    class _ReadableSpan:
+        pass
 
 
-def is_dependency_installed():
+def is_dependency_installed() -> bool:
     if not dependency_installed:
         raise ImportError(
-            "Dependencies are not installed. Please install it with `pip install pydantic-ai opentelemetry-sdk opentelemetry-exporter-otlp-proto-http`."
+            "Dependencies are not installed. Please install it with "
+            "`pip install pydantic-ai opentelemetry-sdk "
+            "opentelemetry-exporter-otlp-proto-http`."
         )
     return True
 
 
-from deepeval.tracing.types import AgentSpan
-from deepeval.confident.api import get_confident_api_key
-from deepeval.prompt import Prompt
-from deepeval.tracing.otel.test_exporter import test_exporter
-from deepeval.tracing.context import current_trace_context
-from deepeval.tracing.types import Trace
-from deepeval.tracing.otel.utils import to_hex_string
-from deepeval.tracing.types import TraceSpanStatus, ToolCall
-from deepeval.tracing.perf_epoch_bridge import init_clock_bridge
+if TYPE_CHECKING:
+    # For type checkers, use real types
+    from opentelemetry.sdk.trace import ReadableSpan, SpanProcessor
+    from pydantic_ai.models.instrumented import InstrumentationSettings
+else:
+    # At runtime we always have something to subclass / annotate with
+    InstrumentationSettings = _BaseInstrumentationSettings
+    SpanProcessor = _SpanProcessor
+    ReadableSpan = _ReadableSpan
 
 # OTLP_ENDPOINT = "http://127.0.0.1:4318/v1/traces"
 OTLP_ENDPOINT = "https://otel.confident-ai.com/v1/traces"