langwatch 0.4.0__tar.gz → 0.5.0__tar.gz

{langwatch-0.4.0 → langwatch-0.5.0}/CHANGELOG.md

@@ -1,5 +1,32 @@
 # Changelog
 
+## [0.5.0](https://github.com/langwatch/langwatch/compare/python-sdk@v0.4.2...python-sdk@v0.5.0) (2025-10-13)
+
+
+### Features
+
+* add support for otel /metrics endpoint for genai metrics ([#680](https://github.com/langwatch/langwatch/issues/680)) ([14bec0d](https://github.com/langwatch/langwatch/commit/14bec0d70d4c645d409b2b18a8f6219515563aed))
+
+
+### Miscellaneous
+
+* update to allow first_token_ms ([#702](https://github.com/langwatch/langwatch/issues/702)) ([38ffd34](https://github.com/langwatch/langwatch/commit/38ffd34f90e49c866690747c0e36b43a867cc698))
+* update version 0.5 ([#704](https://github.com/langwatch/langwatch/issues/704)) ([615510f](https://github.com/langwatch/langwatch/commit/615510fa04ebe3c33635d0edcd3e8f7faeaabb52))
+
+## [0.4.2](https://github.com/langwatch/langwatch/compare/python-sdk@v0.4.1...python-sdk@v0.4.2) (2025-09-22)
+
+
+### Bug Fixes
+
+* downgrade litellm dependency due to https://github.com/BerriAI/litellm/issues/14145 preventing to build it on lambda ([066d97c](https://github.com/langwatch/langwatch/commit/066d97c26252c82f9143e36427782c7af19912a2))
+
+## [0.4.1](https://github.com/langwatch/langwatch/compare/python-sdk@v0.4.0...python-sdk@v0.4.1) (2025-09-19)
+
+
+### Bug Fixes
+
+* prompt tracing ([#653](https://github.com/langwatch/langwatch/issues/653)) ([9d39650](https://github.com/langwatch/langwatch/commit/9d39650435d6f32f040838935a89c037e47124f6))
+
 ## [0.4.0](https://github.com/langwatch/langwatch/compare/python-sdk@v0.3.2...python-sdk@v0.4.0) (2025-09-19)
 
 

{langwatch-0.4.0 → langwatch-0.5.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langwatch
-Version: 0.4.0
+Version: 0.5.0
 Summary: LangWatch Python SDK, for monitoring your LLMs
 Author-email: Langwatch Engineers <engineering@langwatch.ai>
 License: MIT

langwatch-0.5.0/examples/openai_bot_otel_metrics.py

@@ -0,0 +1,89 @@
+from dotenv import load_dotenv
+import os
+
+load_dotenv()
+
+import chainlit as cl
+from openai import OpenAI
+import langwatch
+
+from opentelemetry import metrics
+from opentelemetry.sdk.metrics import MeterProvider
+from opentelemetry.sdk.resources import Resource
+from opentelemetry.sdk.metrics.export import PeriodicExportingMetricReader
+from opentelemetry.exporter.otlp.proto.http.metric_exporter import OTLPMetricExporter
+
+import time
+
+client = OpenAI()
+
+exporter = OTLPMetricExporter(
+    endpoint=f"{os.environ.get('LANGWATCH_ENDPOINT', 'https://app.langwatch.ai')}/api/otel/v1/metrics",
+    headers={"Authorization": "Bearer " + os.environ["LANGWATCH_API_KEY"]},
+)
+
+
+reader = PeriodicExportingMetricReader(exporter)
+provider = MeterProvider(
+    resource=Resource.create({"service.name": "my-agent"}), metric_readers=[reader]
+)
+metrics.set_meter_provider(provider)
+meter = metrics.get_meter("gen_ai.server")
+
+# Create the histogram
+time_to_first_token_hist = meter.create_histogram(
+    name="gen_ai.server.time_to_first_token",
+    unit="s",
+    description="Time to generate first token for successful responses",
+)
+
+
+@langwatch.trace(type="llm")
+async def do_call_llm(content: str, msg):
+    start_time = time.time()
+    completion = client.chat.completions.create(
+        model="gpt-5",
+        messages=[
+            {
+                "role": "system",
+                "content": "You are a helpful assistant that only reply in short tweet-like responses, using lots of emojis.",
+            },
+            {"role": "user", "content": content},
+        ],
+        stream=True,
+        stream_options={"include_usage": True},
+    )
+
+    first_token_time = None
+    full_response = ""
+    for part in completion:
+        if len(part.choices) == 0:
+            continue
+        if token := part.choices[0].delta.content or "":
+            if first_token_time is None:
+                first_token_time = time.time()
+                record_first_token_latency((time.time() - start_time) * 1000)
+            await msg.stream_token(token)
+            full_response += token
+
+    langwatch.get_current_trace().update(output=full_response)
+
+
+@cl.on_message
+@langwatch.trace()
+async def main(message: cl.Message):
+    msg = cl.Message(
+        content="",
+    )
+
+    await do_call_llm(message.content, msg)
+
+    await msg.update()
+
+
+def record_first_token_latency(milliseconds: float):
+    seconds = milliseconds / 1000.0
+    print(f"Record first token latency: {seconds} seconds")
+    time_to_first_token_hist.record(
+        seconds, attributes={"model": "gpt-5", "request.status": "success"}
+    )

{langwatch-0.4.0 → langwatch-0.5.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "langwatch"
-version = "0.4.0" # remember to also update it in src/langwatch/__version__.py
+version = "0.5.0" # remember to also update it in src/langwatch/__version__.py
 description = "LangWatch Python SDK, for monitoring your LLMs"
 authors = [{ name = "Langwatch Engineers", email = "engineering@langwatch.ai" }]
 requires-python = ">=3.10,<3.14"
@@ -95,8 +95,7 @@ examples = [
     "opentelemetry-instrumentation-langchain>=0.40.4",
     "openinference-instrumentation-litellm>=0.1.19",
     "pandas>=2.2.3",
-    "strands-agents[otel]>=1.0.0,<2.0.0",
+    "strands-agents[otel]>=1.9.0,<2.0.0",
     "unstructured[pdf]>=0.18.3",
-    "strands-agents>=1.6.0",
 ]
 tests = ["pytest-asyncio>=0.21.2"]

{langwatch-0.4.0 → langwatch-0.5.0}/src/langwatch/__version__.py

@@ -1,3 +1,3 @@
 """Version information for LangWatch."""
 
-__version__ = "0.4.0"
+__version__ = "0.5.0"

{langwatch-0.4.0 → langwatch-0.5.0}/src/langwatch/domain/__init__.py

@@ -157,6 +157,7 @@ class SpanMetrics(TypedDict, total=False):
     prompt_tokens: Optional[int]
     completion_tokens: Optional[int]
     cost: Optional[float]
+    first_token_ms: Optional[int]
 
 
 class SpanParams(TypedDict, total=False):

{langwatch-0.4.0 → langwatch-0.5.0}/src/langwatch/prompts/decorators/prompt_tracing.py

@@ -1,6 +1,7 @@
 from functools import wraps
 from opentelemetry import trace
-from typing import TYPE_CHECKING, TypeVar, Callable, Any
+from opentelemetry.trace import Span
+from typing import TYPE_CHECKING, TypeVar, Callable, Any, Optional, Union, Dict
 import json
 
 from langwatch.attributes import AttributeKey
@@ -11,9 +12,37 @@ if TYPE_CHECKING:
 T = TypeVar("T", bound="Prompt")
 
 
+def _set_attribute_if_not_none(
+    span: Span, key: str, value: Optional[Union[str, int, float, bool]]
+) -> None:
+    """Set span attribute only if value is not None."""
+    if value is not None:
+        span.set_attribute(key, value)
+
+
 class PromptTracing:
     """Namespace for Prompt method tracing decorators"""
 
+    @staticmethod
+    def _set_prompt_attributes(span: Span, prompt: "Prompt") -> None:
+        """Set prompt-related attributes on the span."""
+        _set_attribute_if_not_none(
+            span, AttributeKey.LangWatchPromptId, getattr(prompt, "id", None)
+        )
+        _set_attribute_if_not_none(
+            span, AttributeKey.LangWatchPromptHandle, getattr(prompt, "handle", None)
+        )
+        _set_attribute_if_not_none(
+            span,
+            AttributeKey.LangWatchPromptVersionId,
+            getattr(prompt, "version_id", None),
+        )
+        _set_attribute_if_not_none(
+            span,
+            AttributeKey.LangWatchPromptVersionNumber,
+            getattr(prompt, "version", None),
+        )
+
     @staticmethod
     def _create_compile_decorator(
         span_name: str,
@@ -28,26 +57,15 @@ class PromptTracing:
                 with trace.get_tracer(__name__).start_as_current_span(
                     PromptTracing._create_span_name(span_name)
                 ) as span:
-                    # Set base prompt
-                    span.set_attributes(
-                        {
-                            AttributeKey.LangWatchPromptId: getattr(self, "id", None),
-                            AttributeKey.LangWatchPromptHandle: getattr(
-                                self, "handle", None
-                            ),
-                            AttributeKey.LangWatchPromptVersionId: getattr(
-                                self, "version_id", None
-                            ),
-                            AttributeKey.LangWatchPromptVersionNumber: getattr(
-                                self, "version", None
-                            ),
-                        }
-                    )
+                    # Set prompt attributes
+                    PromptTracing._set_prompt_attributes(span, self)
 
                     # Create variables dict from args and kwargs
-                    variables_dict: dict[str, Any] = {}
-                    if args and args[0] is not None:
-                        variables_dict.update(args[0])
+                    variables_dict: Dict[str, Any] = {}
+                    if args:
+                        first_arg = args[0]
+                        if first_arg is not None and hasattr(first_arg, "update"):
+                            variables_dict.update(first_arg)
                     variables_dict.update(kwargs)
 
                     span.set_attribute(
@@ -88,7 +106,7 @@ class PromptTracing:
     @staticmethod
     def _create_span_name(span_name: str) -> str:
         """Create a span name for the prompt"""
-        return "Prompt" + "." + span_name
+        return f"Prompt.{span_name}"
 
 
 prompt_tracing = PromptTracing()

{langwatch-0.4.0 → langwatch-0.5.0}/src/langwatch/telemetry/__tests__/test_tracing.py

@@ -46,3 +46,45 @@ def test_metadata_not_lost_on_multiple_updates():
     # None update should not clear
     trace.update(metadata=None)
     assert trace.metadata == {"x": 1, "y": 99, "z": 3}
+
+
+def test_metrics_update():
+    # Test updating metrics including first_token_ms
+    trace = LangWatchTrace()
+
+    # Update with first_token_ms
+    trace.update(metrics={"first_token_ms": 150})
+
+    # Update with additional metrics
+    trace.update(metrics={"prompt_tokens": 100, "completion_tokens": 50})
+
+    # Update first_token_ms again
+    trace.update(metrics={"first_token_ms": 200})
+
+    # Verify the metrics are properly handled (they should be passed to root_span.update)
+    # Since we can't easily test the internal span state without mocking,
+    # we'll test that the update method doesn't raise any errors
+    assert True  # If we get here, the updates succeeded
+
+
+def test_metrics_update_with_root_span():
+    # Test metrics update when there's an active root span
+    trace = LangWatchTrace()
+
+    with trace:
+        # Update metrics while span is active
+        trace.update(metrics={"first_token_ms": 150, "prompt_tokens": 100})
+
+        # Verify metrics were set on the root span
+        assert trace.root_span is not None
+        assert trace.root_span.metrics is not None
+        assert trace.root_span.metrics["first_token_ms"] == 150
+        assert trace.root_span.metrics["prompt_tokens"] == 100
+
+        # Update again with different values
+        trace.update(metrics={"first_token_ms": 200, "completion_tokens": 50})
+
+        # Verify the metrics were updated (merged, not replaced)
+        assert trace.root_span.metrics["first_token_ms"] == 200  # Updated
+        assert trace.root_span.metrics["prompt_tokens"] == 100  # Preserved
+        assert trace.root_span.metrics["completion_tokens"] == 50  # Added

{langwatch-0.4.0 → langwatch-0.5.0}/uv.lock

@@ -2862,7 +2862,7 @@ wheels = [
 
 [[package]]
 name = "langwatch"
-version = "0.4.0"
+version = "0.5.0"
 source = { editable = "." }
 dependencies = [
     { name = "attrs" },
@@ -3002,8 +3002,7 @@ examples = [
     { name = "pydantic", specifier = ">=1,<2.10.2" },
     { name = "pytest", specifier = ">=7.4.2,<8.0.0" },
     { name = "python-dotenv", specifier = "==1.0.1" },
-    { name = "strands-agents", specifier = ">=1.6.0" },
-    { name = "strands-agents", extras = ["otel"], specifier = ">=1.0.0,<2.0.0" },
+    { name = "strands-agents", extras = ["otel"], specifier = ">=1.9.0,<2.0.0" },
     { name = "streamlit", specifier = ">=1.37.1,<2.0.0" },
     { name = "unstructured", extras = ["pdf"], specifier = ">=0.18.3" },
     { name = "uvicorn", specifier = ">=0.25.0,<0.26.0" },
@@ -6274,7 +6273,7 @@ wheels = [
 
 [[package]]
 name = "strands-agents"
-version = "1.6.0"
+version = "1.9.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "boto3" },
@@ -6288,9 +6287,9 @@ dependencies = [
     { name = "typing-extensions" },
     { name = "watchdog" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/12/cc/e4c0324895c15fd1f0507616d7346697e0810ac431e588860a16d6916cf4/strands_agents-1.6.0.tar.gz", hash = "sha256:bb3ce55e29630210ba19ac13b5b6aa7c92d1e403a80bbf0738e58d092dba8036", size = 268023, upload-time = "2025-08-26T17:28:54.701Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/40/e0/65952b5c8767b62e9e5b4da1df8203bbaaa1e0d8753b5bc0a35674e0d607/strands_agents-1.9.0.tar.gz", hash = "sha256:84610b855bc79236ce822626b852c6f5a3c1a1de63592445a704a4f8a7249695", size = 394868, upload-time = "2025-09-17T19:13:01.751Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/27/c3/a8beae822a7960eaa16bbc41d3854f5a0bf9753336e6a659bf78c96ec6ea/strands_agents-1.6.0-py3-none-any.whl", hash = "sha256:c7ce5f061df94d7081d1368aa4037c99cf2239f94a8151eee48fac57bf7b0707", size = 186311, upload-time = "2025-08-26T17:28:52.338Z" },
+    { url = "https://files.pythonhosted.org/packages/64/af/dec0c4c71eeafb07fdbf24f74f2d0e7a6796931a3a192e19d0ed0e9bb106/strands_agents-1.9.0-py3-none-any.whl", hash = "sha256:a7a7c51d71278ee6daf7d47089456abc6de3a62e7cd433c08031a97c6b2e59f3", size = 205328, upload-time = "2025-09-17T19:13:00.22Z" },
 ]
 
 [package.optional-dependencies]