fix: Duplicate LLM calls

[dot-agi]

📥 Pull Request

📘 Description
Fixes duplicate LLM calls in CrewAI due to instrumentation of the LLM class causing another span to be produced
Closes #959

🧪 Testing
The OG job posting example

📋 Future work

• Link with the current span which is being called from the SDK
• Add the custom attributes in the current span
• Integrate LiteLLM and remove both above

[Copilot]

Pull Request Overview

This PR addresses duplicate LLM call issues by enhancing the instrumentation logic for CrewAI by adding provider detection logic.

• Introduces a mapping from LLM providers to specific instrumentors to selectively skip instrumentation.
• Wraps provider detection logic in a try/except block to handle potential errors during instrumentation setup.

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 48 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
agentops/instrumentation/crewai/instrumentation.py	0.00%	48 Missing ⚠️

📢 Thoughts on this report? Let us know!

[Copilot]

Pull Request Overview

This PR fixes duplicate LLM calls by detecting when a provider’s own instrumentation is in use and skipping the extra span, refactors token attribute handling, and adds error handling around provider detection.

• Added mapping from model identifiers to known instrumentors and early-return if matched
• Reformatted callback token attribute setting for readability
• Wrapped provider-detection logic in a try/except to log and avoid crashing

Comments suppressed due to low confidence (2)

agentops/instrumentation/crewai/instrumentation.py:439

• The with block appears unindented relative to the preceding else: and may execute even when an instrumentor is found, breaking the intended skip logic. Please correct the indentation so that the span is only started in the else branch.

with tracer.start_as_current_span(f"{llm}.llm", kind=SpanKind.CLIENT, attributes={}) as span:

agentops/instrumentation/crewai/instrumentation.py:431

• [nitpick] New behavior for both matching and non-matching providers should be covered by unit tests to ensure correct skipping or instrumentation. Please add tests for cases where the provider is in the map and where it is not.

instrumentor = provider_instrumentor.get(provider.lower())

[Dwij1704]

Crewai instrumentation is so much better now, Thanks @the-praxs !!