Skip to main content

​
What are Instrumentations?

Glass is built on OpenTelemetry, the industry-standard for observability. OpenTelemetry uses instrumentations to automatically capture traces from libraries and frameworks — without you changing any code. When you call init(), Glass automatically instruments your AI providers. Every API call is traced with request/response data, token usage, latency, and errors.

​
Default Instrumentations

Glass automatically instruments these AI providers out of the box:
ProviderPackageWhat’s Traced
OpenAIopenaiChat completions, embeddings, images, audio
AnthropicanthropicMessages, completions
Google Generative AIgoogle-generativeaiGemini generate, chat, embeddings
Default instrumentations are enabled automatically. No configuration needed.
from glass import init
from openai import OpenAI

init(api_key="your-api-key")

# This call is automatically traced
client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)
Each traced call captures:
  • Request and response payloads
  • Token usage (input/output/total)
  • Model name and parameters
  • Latency
  • Errors and exceptions

​
Adding Custom Instrumentations

Need to trace other libraries? You can add any OpenTelemetry-compatible instrumentation.

​
Example: Adding HTTP Request Tracing

from glass import init
from opentelemetry.instrumentation.requests import RequestsInstrumentor

init(
    api_key="your-api-key",
    instrumentations=[RequestsInstrumentor()]
)
Now all requests library calls are traced alongside your AI calls.

​
Example: Multiple Custom Instrumentations

from glass import init
from opentelemetry.instrumentation.requests import RequestsInstrumentor
from opentelemetry.instrumentation.httpx import HTTPXClientInstrumentor
from opentelemetry.instrumentation.sqlalchemy import SQLAlchemyInstrumentor

init(
    api_key="your-api-key",
    instrumentations=[
        RequestsInstrumentor(),
        HTTPXClientInstrumentor(),
        SQLAlchemyInstrumentor(),
    ]
)
Custom instrumentations are added on top of the defaults. Your AI providers are still traced automatically.

​
Disabling Default Instrumentations

If you want full control over what gets instrumented, disable the defaults: 
from glass import init
from opentelemetry.instrumentation.openai import OpenAIInstrumentor

init(
    api_key="your-api-key",
    skip_default_instrumentations=True,
    instrumentations=[
        OpenAIInstrumentor(),  # Only instrument OpenAI
    ]
)
Here are commonly used instrumentations for AI/ML workflows:
LibraryInstrumentation PackageInstall
LangChainopentelemetry-instrumentation-langchainpip install opentelemetry-instrumentation-langchain
LlamaIndexopentelemetry-instrumentation-llamaindexpip install opentelemetry-instrumentation-llamaindex
Cohereopentelemetry-instrumentation-coherepip install opentelemetry-instrumentation-cohere
Bedrockopentelemetry-instrumentation-bedrockpip install opentelemetry-instrumentation-bedrock
Replicateopentelemetry-instrumentation-replicatepip install opentelemetry-instrumentation-replicate
Pineconeopentelemetry-instrumentation-pineconepip install opentelemetry-instrumentation-pinecone
Chromaopentelemetry-instrumentation-chromadbpip install opentelemetry-instrumentation-chromadb
Weaviateopentelemetry-instrumentation-weaviatepip install opentelemetry-instrumentation-weaviate
Qdrantopentelemetry-instrumentation-qdrantpip install opentelemetry-instrumentation-qdrant

OpenTelemetry Registry

Browse all available Python instrumentations in the OpenTelemetry Registry.