Python autoinstrumentation library for the Anthropic package
This package implements the following Anthropic clients:
Messages (create, stream, parse, count_tokens)AsyncMessagesMessages and AsyncMessages, including those of the Bedrock and Vertex AI clientsllm.provider is the platform hosting the model: aws for Bedrock, google for Vertex AI,
azure for Microsoft Foundry, and anthropic otherwise.
These traces are fully OpenTelemetry compatible and can be sent to an OpenTelemetry collector for viewing, such as Arize Phoenix or Arize AX.
openinference-instrumentation-anthropic |
anthropic |
Python |
|---|---|---|
>=3.2.0 |
>=1.8.0 |
>=3.10, <3.16 |
>=3.0.0, <3.2.0 |
>=1.8.0 |
>=3.10, <3.15 |
>=2.1.6, <3.0.0 |
>=1.0.0 |
>=3.10, <3.15 |
2.1.5 |
>=1.0.0, <1.8 |
>=3.10, <3.15 |
>=2.0, <2.1.5 |
>=1.0.0, <1.5 |
>=3.10, <3.15 |
>=1.0.1, <2.0 |
>=0.84.0, <1.0 |
>=3.10, <3.15 |
1.0.0 |
>=0.84.0, <1.0 |
>=3.9, <3.15 |
Anthropic 1.0 removed the legacy Text Completions API. Instrumentor 2.0 therefore
requires anthropic>=1.0.0 and instruments the Messages and Beta Messages APIs.
See Anthropic’s v1 migration guide
for the upstream breaking changes.
anthropic>=1.14.0 creates an OpenTelemetry span of its own for each API call
(anthropic.messages.create and so on), unless it is turned off with
open_telemetry=False or ANTHROPIC_OPEN_TELEMETRY=false. For a call that this
instrumentor traces, the SDK’s span is replaced by the OpenInference LLM span: the SDK
records no span, its request carries the LLM span’s trace context, and HTTP spans are
children of the LLM span. Calls that this instrumentor does not trace, such as
models.list(), keep the SDK’s span.
pip install openinference-instrumentation-anthropic
Through your terminal, install required packages.
pip install openinference-instrumentation-anthropic anthropic arize-phoenix opentelemetry-sdk opentelemetry-exporter-otlp
You can start Phoenix with the following terminal command:
python -m phoenix.server.main serve
By default, Phoenix listens on http://localhost:6006. You can visit the app via a browser at the same address. (Phoenix does not send data over the internet. It only operates locally on your machine.)
Try the following code in a Python file.
AnthropicInstrumentor to trace your application and sends the traces to Phoenix.http://localhost:6006!import os
from anthropic import Anthropic
from openinference.instrumentation.anthropic import AnthropicInstrumentor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk import trace as trace_sdk
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
# Configure AnthropicInstrumentor with Phoenix endpoint
endpoint = "http://127.0.0.1:6006/v1/traces"
tracer_provider = trace_sdk.TracerProvider()
tracer_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
AnthropicInstrumentor().instrument(tracer_provider=tracer_provider)
os.environ["ANTHROPIC_API_KEY"] = "YOUR_KEY_HERE"
client = Anthropic()
response = client.messages.create(
max_tokens=1024,
messages=[
{
"role": "user",
"content": "Tell me about the history of Iceland!",
}
],
model="claude-3-opus-20240229",
)
print(response)
Now, on the Phoenix UI on your browser, you should see the traces from your Anthropic application. Click on a trace, then the “Attributes” tab will provide you with in-depth information regarding execution!