braintrust 0.5.0__tar.gz → 0.5.3__tar.gz

{braintrust-0.5.0 → braintrust-0.5.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: braintrust
-Version: 0.5.0
+Version: 0.5.3
 Summary: SDK for integrating Braintrust
 Home-page: https://www.braintrust.dev
 Author: Braintrust

{braintrust-0.5.0 → braintrust-0.5.3}/src/braintrust/__init__.py

@@ -49,7 +49,21 @@ BRAINTRUST_API_KEY=<YOUR_BRAINTRUST_API_KEY> braintrust eval eval_hello.py
 ### API Reference
 """
 
+# Check env var at import time for auto-instrumentation
+import os
+
+if os.getenv("BRAINTRUST_INSTRUMENT_THREADS", "").lower() in ("true", "1", "yes"):
+    try:
+        from .wrappers.threads import setup_threads
+
+        setup_threads()
+    except Exception:
+        pass  # Never break on import
+
 from .audit import *
+from .auto import (
+    auto_instrument,  # noqa: F401 # type: ignore[reportUnusedImport]
+)
 from .framework import *
 from .framework2 import *
 from .functions.invoke import *

{braintrust-0.5.0 → braintrust-0.5.3}/src/braintrust/_generated_types.py

@@ -167,6 +167,10 @@ AsyncScoringState: TypeAlias = AsyncScoringStateAsyncScoringState | AsyncScoring
 class PreprocessorPreprocessor(TypedDict):
     type: Literal['function']
     id: str
+    version: NotRequired[str | None]
+    """
+    The version of the function
+    """
 
 
 class PreprocessorPreprocessor2(TypedDict):
@@ -638,6 +642,10 @@ class ExperimentEventContext(TypedDict):
 class ExtendedSavedFunctionIdExtendedSavedFunctionId(TypedDict):
     type: Literal['function']
     id: str
+    version: NotRequired[str | None]
+    """
+    The version of the function
+    """
 
 
 class ExtendedSavedFunctionIdExtendedSavedFunctionId2(TypedDict):
@@ -668,6 +676,10 @@ class ExternalAttachmentReference(TypedDict):
 class Preprocessor1Preprocessor1(TypedDict):
     type: Literal['function']
     id: str
+    version: NotRequired[str | None]
+    """
+    The version of the function
+    """
 
 
 class Preprocessor1Preprocessor12(TypedDict):
@@ -803,7 +815,7 @@ FunctionOutputType: TypeAlias = Literal['completion', 'score', 'facet', 'classif
 
 
 FunctionTypeEnum: TypeAlias = Literal[
-    'llm', 'scorer', 'task', 'tool', 'custom_view', 'preprocessor', 'facet', 'classifier'
+    'llm', 'scorer', 'task', 'tool', 'custom_view', 'preprocessor', 'facet', 'classifier', 'tag'
 ]
 """
 The type of global function. Defaults to 'scorer'.
@@ -811,7 +823,7 @@ The type of global function. Defaults to 'scorer'.
 
 
 FunctionTypeEnumNullish: TypeAlias = Literal[
-    'llm', 'scorer', 'task', 'tool', 'custom_view', 'preprocessor', 'facet', 'classifier'
+    'llm', 'scorer', 'task', 'tool', 'custom_view', 'preprocessor', 'facet', 'classifier', 'tag'
 ]
 
 
@@ -1108,6 +1120,12 @@ class GroupScope(TypedDict):
 IfExists: TypeAlias = Literal['error', 'ignore', 'replace']
 
 
+ImageRenderingMode: TypeAlias = Literal['auto', 'click_to_load', 'blocked']
+"""
+Controls how images are rendered in the UI: 'auto' loads images automatically, 'click_to_load' shows a placeholder until clicked, 'blocked' prevents image loading entirely
+"""
+
+
 class InvokeFunctionInvokeFunction(TypedDict):
     function_id: str
     """
@@ -1311,6 +1329,10 @@ class ModelParamsModelParams4(TypedDict):
 class NullableSavedFunctionIdNullableSavedFunctionId(TypedDict):
     type: Literal['function']
     id: str
+    version: NotRequired[str | None]
+    """
+    The version of the function
+    """
 
 
 class NullableSavedFunctionIdNullableSavedFunctionId1(TypedDict):
@@ -1390,6 +1412,7 @@ class Organization(TypedDict):
     """
     Date of organization creation
     """
+    image_rendering_mode: NotRequired[ImageRenderingMode | None]
 
 
 Permission: TypeAlias = Literal[
@@ -2078,6 +2101,10 @@ class RunEvalMcpAuth(TypedDict):
 class SavedFunctionIdSavedFunctionId(TypedDict):
     type: Literal['function']
     id: str
+    version: NotRequired[str | None]
+    """
+    The version of the function
+    """
 
 
 class SavedFunctionIdSavedFunctionId1(TypedDict):
@@ -2590,6 +2617,14 @@ class DatasetEvent(TypedDict):
     """
     Optional list of audit entries attached to this event
     """
+    facets: NotRequired[Mapping[str, Any] | None]
+    """
+    Facets for categorization (dictionary from facet id to value)
+    """
+    classifications: NotRequired[Mapping[str, Any] | None]
+    """
+    Classifications for this event (dictionary from classification name to items)
+    """
 
 
 class EvalStatusPage(TypedDict):
@@ -3101,6 +3136,14 @@ class ExperimentEvent(TypedDict):
     """
     Optional list of audit entries attached to this event
     """
+    facets: NotRequired[Mapping[str, Any] | None]
+    """
+    Facets for categorization (dictionary from facet id to value)
+    """
+    classifications: NotRequired[Mapping[str, Any] | None]
+    """
+    Classifications for this event (dictionary from classification name to items)
+    """
 
 
 class GraphNodeGraphNode7(TypedDict):
@@ -3223,6 +3266,14 @@ class ProjectLogsEvent(TypedDict):
     """
     The async scoring state for this event
     """
+    facets: NotRequired[Mapping[str, Any] | None]
+    """
+    Facets for categorization (dictionary from facet id to value)
+    """
+    classifications: NotRequired[Mapping[str, Any] | None]
+    """
+    Classifications for this event (dictionary from classification name to items)
+    """
 
 
 class ProjectScore(TypedDict):
@@ -3330,7 +3381,9 @@ class View(TypedDict):
         'classifiers',
         'logs',
         'monitor',
-        'for_review',
+        'for_review_project_log',
+        'for_review_experiments',
+        'for_review_datasets',
     ]
     """
     Type of object that the view corresponds to.

braintrust-0.5.3/src/braintrust/auto.py

@@ -0,0 +1,179 @@
+"""
+Auto-instrumentation for AI/ML libraries.
+
+Provides one-line instrumentation for supported libraries.
+"""
+
+from __future__ import annotations
+
+import logging
+from contextlib import contextmanager
+
+__all__ = ["auto_instrument"]
+
+logger = logging.getLogger(__name__)
+
+
+@contextmanager
+def _try_patch():
+    """Context manager that suppresses ImportError and logs other exceptions."""
+    try:
+        yield
+    except ImportError:
+        pass
+    except Exception:
+        logger.exception("Failed to instrument")
+
+
+def auto_instrument(
+    *,
+    openai: bool = True,
+    anthropic: bool = True,
+    litellm: bool = True,
+    pydantic_ai: bool = True,
+    google_genai: bool = True,
+    agno: bool = True,
+    claude_agent_sdk: bool = True,
+    dspy: bool = True,
+) -> dict[str, bool]:
+    """
+    Auto-instrument supported AI/ML libraries for Braintrust tracing.
+
+    Safe to call multiple times - already instrumented libraries are skipped.
+
+    Note on import order: If you use `from openai import OpenAI` style imports,
+    call auto_instrument() first. If you use `import openai` style imports,
+    order doesn't matter since attribute lookup happens dynamically.
+
+    Args:
+        openai: Enable OpenAI instrumentation (default: True)
+        anthropic: Enable Anthropic instrumentation (default: True)
+        litellm: Enable LiteLLM instrumentation (default: True)
+        pydantic_ai: Enable Pydantic AI instrumentation (default: True)
+        google_genai: Enable Google GenAI instrumentation (default: True)
+        agno: Enable Agno instrumentation (default: True)
+        claude_agent_sdk: Enable Claude Agent SDK instrumentation (default: True)
+        dspy: Enable DSPy instrumentation (default: True)
+
+    Returns:
+        Dict mapping integration name to whether it was successfully instrumented.
+
+    Example:
+        ```python
+        import braintrust
+        braintrust.auto_instrument()
+
+        # OpenAI
+        import openai
+        client = openai.OpenAI()
+        client.chat.completions.create(model="gpt-4o-mini", messages=[...])
+
+        # Anthropic
+        import anthropic
+        client = anthropic.Anthropic()
+        client.messages.create(model="claude-sonnet-4-20250514", messages=[...])
+
+        # LiteLLM
+        import litellm
+        litellm.completion(model="gpt-4o-mini", messages=[...])
+
+        # DSPy
+        import dspy
+        lm = dspy.LM("openai/gpt-4o-mini")
+        dspy.configure(lm=lm)
+
+        # Pydantic AI
+        from pydantic_ai import Agent
+        agent = Agent("openai:gpt-4o-mini")
+        result = agent.run_sync("Hello!")
+
+        # Google GenAI
+        from google.genai import Client
+        client = Client()
+        client.models.generate_content(model="gemini-2.0-flash", contents="Hello!")
+        ```
+    """
+    results = {}
+
+    if openai:
+        results["openai"] = _instrument_openai()
+    if anthropic:
+        results["anthropic"] = _instrument_anthropic()
+    if litellm:
+        results["litellm"] = _instrument_litellm()
+    if pydantic_ai:
+        results["pydantic_ai"] = _instrument_pydantic_ai()
+    if google_genai:
+        results["google_genai"] = _instrument_google_genai()
+    if agno:
+        results["agno"] = _instrument_agno()
+    if claude_agent_sdk:
+        results["claude_agent_sdk"] = _instrument_claude_agent_sdk()
+    if dspy:
+        results["dspy"] = _instrument_dspy()
+
+    return results
+
+
+def _instrument_openai() -> bool:
+    with _try_patch():
+        from braintrust.oai import patch_openai
+
+        return patch_openai()
+    return False
+
+
+def _instrument_anthropic() -> bool:
+    with _try_patch():
+        from braintrust.wrappers.anthropic import patch_anthropic
+
+        return patch_anthropic()
+    return False
+
+
+def _instrument_litellm() -> bool:
+    with _try_patch():
+        from braintrust.wrappers.litellm import patch_litellm
+
+        return patch_litellm()
+    return False
+
+
+def _instrument_pydantic_ai() -> bool:
+    with _try_patch():
+        from braintrust.wrappers.pydantic_ai import setup_pydantic_ai
+
+        return setup_pydantic_ai()
+    return False
+
+
+def _instrument_google_genai() -> bool:
+    with _try_patch():
+        from braintrust.wrappers.google_genai import setup_genai
+
+        return setup_genai()
+    return False
+
+
+def _instrument_agno() -> bool:
+    with _try_patch():
+        from braintrust.wrappers.agno import setup_agno
+
+        return setup_agno()
+    return False
+
+
+def _instrument_claude_agent_sdk() -> bool:
+    with _try_patch():
+        from braintrust.wrappers.claude_agent_sdk import setup_claude_agent_sdk
+
+        return setup_claude_agent_sdk()
+    return False
+
+
+def _instrument_dspy() -> bool:
+    with _try_patch():
+        from braintrust.wrappers.dspy import patch_dspy
+
+        return patch_dspy()
+    return False

{braintrust-0.5.0 → braintrust-0.5.3}/src/braintrust/conftest.py

@@ -48,16 +48,29 @@ def reset_braintrust_state():
     logger._state = logger.BraintrustState()
 
 
-@pytest.fixture(scope="session")
-def vcr_config():
+@pytest.fixture(autouse=True)
+def skip_vcr_tests_in_wheel_mode(request):
+    """Skip VCR tests when running from an installed wheel.
+
+    Wheel mode (BRAINTRUST_TESTING_WHEEL=1) is a pre-release sanity check
+    that verifies the built package installs and runs correctly. It's not
+    intended to be a full test suite - VCR cassettes are not included in
+    the wheel, so we skip those tests here. The full test suite with VCR
+    tests runs against source code during normal CI.
+    """
+    if os.environ.get("BRAINTRUST_TESTING_WHEEL") == "1":
+        if request.node.get_closest_marker("vcr"):
+            pytest.skip("VCR tests skipped in wheel mode (pre-release sanity check only)")
+
+
+def get_vcr_config():
     """
-    VCR configuration for recording/playing back HTTP interactions.
+    Get VCR configuration for recording/playing back HTTP interactions.
 
     In CI, use "none" to fail if cassette is missing.
     Locally, use "once" to record new cassettes if they don't exist.
     """
     record_mode = "none" if (os.environ.get("CI") or os.environ.get("GITHUB_ACTIONS")) else "once"
-
     return {
         "record_mode": record_mode,
         "filter_headers": [
@@ -70,3 +83,9 @@ def vcr_config():
             "x-bt-auth-token",
         ],
     }
+
+
+@pytest.fixture(scope="session")
+def vcr_config():
+    """Pytest fixture wrapper for get_vcr_config()."""
+    return get_vcr_config()

{braintrust-0.5.0 → braintrust-0.5.3}/src/braintrust/db_fields.py

@@ -15,3 +15,13 @@ PARENT_ID_FIELD = "_parent_id"
 
 ASYNC_SCORING_CONTROL_FIELD = "_async_scoring_control"
 SKIP_ASYNC_SCORING_FIELD = "_skip_async_scoring"
+
+# Keys that identify which object (experiment, dataset, project logs, etc.) a row belongs to.
+OBJECT_ID_KEYS = (
+    "experiment_id",
+    "dataset_id",
+    "prompt_session_id",
+    "project_id",
+    "log_id",
+    "function_data",
+)

{braintrust-0.5.0 → braintrust-0.5.3}/src/braintrust/framework.py

@@ -673,6 +673,7 @@ def _EvalCommon(
     stream: Callable[[SSEProgressEvent], None] | None = None,
     parent: str | None = None,
     state: BraintrustState | None = None,
+    enable_cache: bool = True,
 ) -> Callable[[], Coroutine[Any, Any, EvalResultWithSummary[Input, Output]]]:
     """
     This helper is needed because in case of `_lazy_load`, we need to update
@@ -759,7 +760,7 @@ def _EvalCommon(
         async def run_to_completion():
             with parent_context(parent, state):
                 try:
-                    ret = await run_evaluator(experiment, evaluator, 0, [], stream, state)
+                    ret = await run_evaluator(experiment, evaluator, 0, [], stream, state, enable_cache)
                     reporter.report_eval(evaluator, ret, verbose=True, jsonl=False)
                     return ret
                 finally:
@@ -798,6 +799,7 @@ async def EvalAsync(
     stream: Callable[[SSEProgressEvent], None] | None = None,
     parent: str | None = None,
     state: BraintrustState | None = None,
+    enable_cache: bool = True,
 ) -> EvalResultWithSummary[Input, Output]:
     """
     A function you can use to define an evaluator. This is a convenience wrapper around the `Evaluator` class.
@@ -855,6 +857,8 @@ async def EvalAsync(
     :param parent: If specified, instead of creating a new experiment object, the Eval() will populate
     the object or span specified by this parent.
     :param state: Optional BraintrustState to use for the evaluation. If not specified, the global login state will be used.
+    :param enable_cache: Whether to enable the span cache for this evaluation. Defaults to True. The span cache stores
+    span data on disk to minimize memory usage and allow scorers to read spans without server round-trips.
     :return: An `EvalResultWithSummary` object, which contains all results and a summary.
     """
     f = _EvalCommon(
@@ -883,6 +887,7 @@ async def EvalAsync(
         stream=stream,
         parent=parent,
         state=state,
+        enable_cache=enable_cache,
     )
 
     return await f()
@@ -918,6 +923,7 @@ def Eval(
     stream: Callable[[SSEProgressEvent], None] | None = None,
     parent: str | None = None,
     state: BraintrustState | None = None,
+    enable_cache: bool = True,
 ) -> EvalResultWithSummary[Input, Output]:
     """
     A function you can use to define an evaluator. This is a convenience wrapper around the `Evaluator` class.
@@ -975,6 +981,8 @@ def Eval(
     :param parent: If specified, instead of creating a new experiment object, the Eval() will populate
     the object or span specified by this parent.
     :param state: Optional BraintrustState to use for the evaluation. If not specified, the global login state will be used.
+    :param enable_cache: Whether to enable the span cache for this evaluation. Defaults to True. The span cache stores
+    span data on disk to minimize memory usage and allow scorers to read spans without server round-trips.
     :return: An `EvalResultWithSummary` object, which contains all results and a summary.
     """
 
@@ -1005,6 +1013,7 @@ def Eval(
         stream=stream,
         parent=parent,
         state=state,
+        enable_cache=enable_cache,
     )
 
     # https://stackoverflow.com/questions/55409641/asyncio-run-cannot-be-called-from-a-running-event-loop-when-using-jupyter-no
@@ -1249,10 +1258,11 @@ async def run_evaluator(
     filters: list[Filter],
     stream: Callable[[SSEProgressEvent], None] | None = None,
     state: BraintrustState | None = None,
+    enable_cache: bool = True,
 ) -> EvalResultWithSummary[Input, Output]:
     """Wrapper on _run_evaluator_internal that times out execution after evaluator.timeout."""
     results = await asyncio.wait_for(
-        _run_evaluator_internal(experiment, evaluator, position, filters, stream, state), evaluator.timeout
+        _run_evaluator_internal(experiment, evaluator, position, filters, stream, state, enable_cache), evaluator.timeout
     )
 
     if experiment:
@@ -1280,6 +1290,7 @@ async def _run_evaluator_internal(
     filters: list[Filter],
     stream: Callable[[SSEProgressEvent], None] | None = None,
     state: BraintrustState | None = None,
+    enable_cache: bool = True,
 ):
     # Start span cache for this eval (it's disabled by default to avoid temp files outside of evals)
     if state is None:
@@ -1287,13 +1298,15 @@ async def _run_evaluator_internal(
 
         state = _internal_get_global_state()
 
-    state.span_cache.start()
+    if enable_cache:
+        state.span_cache.start()
     try:
         return await _run_evaluator_internal_impl(experiment, evaluator, position, filters, stream, state)
     finally:
         # Clean up disk-based span cache after eval completes and stop caching
-        state.span_cache.dispose()
-        state.span_cache.stop()
+        if enable_cache:
+            state.span_cache.dispose()
+            state.span_cache.stop()
 
 
 async def _run_evaluator_internal_impl(

{braintrust-0.5.0 → braintrust-0.5.3}/src/braintrust/generated_types.py

@@ -1,4 +1,4 @@
-"""Auto-generated file (internal git SHA 21146f64bf5ad1eadd3a99d186274728e25e5399) -- do not modify"""
+"""Auto-generated file (internal git SHA 7178200dd3c7869f27b677260303cda0b798bf42) -- do not modify"""
 
 from ._generated_types import (
     Acl,
@@ -53,6 +53,7 @@ from ._generated_types import (
     Group,
     GroupScope,
     IfExists,
+    ImageRenderingMode,
     InvokeFunction,
     InvokeParent,
     MCPServer,
@@ -163,6 +164,7 @@ __all__ = [
     "Group",
     "GroupScope",
     "IfExists",
+    "ImageRenderingMode",
     "InvokeFunction",
     "InvokeParent",
     "MCPServer",