🚀 Feature: Install less packages

[InAnYan]

Which component is this feature for?

Traceloop SDK

🔖 Feature description

I've installed traceloop-sdk with uv, and it installed a lot of packages:

 + opentelemetry-instrumentation-alephalpha==0.39.0
 + opentelemetry-instrumentation-anthropic==0.39.0
 + opentelemetry-instrumentation-bedrock==0.39.0
 + opentelemetry-instrumentation-chromadb==0.39.0
 + opentelemetry-instrumentation-cohere==0.39.0
 + opentelemetry-instrumentation-crewai==0.39.0
 + opentelemetry-instrumentation-google-generativeai==0.39.0
 + opentelemetry-instrumentation-groq==0.39.0
 + opentelemetry-instrumentation-haystack==0.39.0
 + opentelemetry-instrumentation-lancedb==0.39.0
 + opentelemetry-instrumentation-langchain==0.39.0
 + opentelemetry-instrumentation-llamaindex==0.39.0
 + opentelemetry-instrumentation-logging==0.52b1
 + opentelemetry-instrumentation-marqo==0.39.0
 + opentelemetry-instrumentation-milvus==0.39.0
 + opentelemetry-instrumentation-mistralai==0.39.0
 + opentelemetry-instrumentation-ollama==0.39.0
 + opentelemetry-instrumentation-openai==0.39.0
 + opentelemetry-instrumentation-pinecone==0.39.0
 + opentelemetry-instrumentation-qdrant==0.39.0
 + opentelemetry-instrumentation-replicate==0.39.0
 + opentelemetry-instrumentation-requests==0.52b1
 + opentelemetry-instrumentation-sagemaker==0.39.0
 + opentelemetry-instrumentation-sqlalchemy==0.52b1
 + opentelemetry-instrumentation-threading==0.52b1
 + opentelemetry-instrumentation-together==0.39.0
 + opentelemetry-instrumentation-transformers==0.39.0
 + opentelemetry-instrumentation-urllib3==0.52b1
 + opentelemetry-instrumentation-vertexai==0.39.0
 + opentelemetry-instrumentation-watsonx==0.39.0
 + opentelemetry-instrumentation-weaviate==0.39.0
 + opentelemetry-semantic-conventions-ai==0.4.3
 + opentelemetry-util-http==0.52b1

Is there a way to limit number of packages and install only necessary?

🎤 Why is this feature needed ?

I don't really use all of these technologies in my app. This is too much.

Everywhere where openllmetry is mentioned (even your docs), they tell to use this command:

pip install traceloop-sdk

Many projects use this feature: https://stackoverflow.com/a/46775606/10037342, where extensions are listed in square brackets.

✌️ How do you aim to achieve this?

Use optional dependencies, so that only needed packages are installed.

And replace original command with:

pip install traceloop-sdk[full]

Something like that. Free to discuss, free to close 😄

🔄️ Additional Information

No response

👀 Have you spent some time to check if this feature request has been raised before?

• I checked and didn't find similar issue

Are you willing to submit PR?

None

[InAnYan]

Side note: I see that you can actually install separate packages: https://qdrant.tech/documentation/observability/openllmetry/

[nirga]

Hey @InAnYan, you're right. When we began this OSS we just had 5 packages and it has overgrown a bit. We'll prioritize this but feel free to take a stab at it if you have the capacity - we'd love the contribution :)

[brambhattabhishek]

/assign

[Rohan5kumar]

You're absolutely right — installing traceloop-sdk currently pulls in all optional OpenTelemetry instrumentation packages, even when users only need a subset. This can result in unnecessary bloat, especially for minimal or targeted setups.

It would definitely be more efficient to split these dependencies using extras_require in setup.py or pyproject.toml. For example:

pip install traceloop-sdk[openai]
pip install traceloop-sdk[langchain,llamaindex]
pip install traceloop-sdk[full] # for everything

This way, users can install only the integrations they need.

If the maintainers are open to this, I’d be happy to help contribute a PR to implement optional dependencies and update the docs accordingly.

[gauravbh024]

you can do it manually or you can use some technics for that and sprate them

[amitalokbera]

picking this up

[amitalokbera]

@nirga

I'm planning to modularize the dependencies like this, let me know if this looks good to you

• llm: openai, anthropic, cohere, groq, etc.
• frameworks: langchain, llamaindex, crewai, haystack, transformers
• vectorstores: pinecone, qdrant, weaviate, chromadb, lancedb, milvus
• cloud: bedrock, vertexai, sagemaker, watsonx, google-generativeai
• minimal: just the base SDK (opentelemetry + pydantic, logging, etc.)
• all: includes everything above

To handle cases where a user installs the minimal version but tries to use an integration from a missing group, I'm also thinking of adding a small utility like:

from .utils import require_dependency

# Ensure the 'openai' package is installed
require_dependency("openai", "llm")

This would raise a clear error like:

Missing optional dependency 'openai'. Please install with 'traceloop-sdk[llm]' to use this feature.

One drawback is that this check would need to be added in multiple places across the codebase, which isn’t ideal. Open to better suggestions or ideas for a cleaner solution before I move ahead with the PR.

[nirga]

Sounds good @amitalokbera! I might even take it one step further (not sure how easy / hard it will be) - have a module per LLM. So you can even just install traceloop-sdk[openai] - which is a common scenario imo. Same for frameworks - people are more likely to use just one framework, and not all of them.