Description

I’m hitting repeated dependency-resolution failures because several constraints in pyproject.toml are much stricter than the broader Python AI ecosystem now expects.

Concrete examples:

• openai is pinned to a very narrow window (openai>=1.83.0,<1.84.dev0). This makes CrewAI incompatible with packages that require newer OpenAI SDKs (e.g. langchain-openai>=1.0.3 requires openai>=1.109.1,<3.0.0, and tools/observability integrations like openlit require openai>=1.92.0).

Steps to Reproduce

uv add "crewai[tools]" "langchain-openai>=1.0.3"

Resolver fails because CrewAI pins OpenAI SDK to <1.84.dev0 while langchain-openai needs >=1.109.1.

Expected behavior

CrewAI should remain installable alongside common ecosystem packages (LangChain integrations, LiteLLM proxy stacks, observability tools, transformers/sentence-transformers), unless CrewAI truly requires a breaking API change.

Screenshots/Code snippets

• Package manager: uv

Operating System

macOS Sonoma

Python Version

3.12

crewAI Version

1.9.2

crewAI Tools Version

1.9.2

Virtual Environment

Poetry

Evidence

Error: Exit code 1
× No solution found when resolving dependencies for split (markers:
│ python_full_version >= '3.14'):
╰─▶ Because crewai==1.9.2 depends on openai>=1.83.0,<1.84.dev0 and your
project depends on crewai[litellm]==1.9.2, we can conclude that your
project depends on openai>=1.83.0,<1.84.dev0.
And because your project depends on openai>=1.90.0, we can conclude that
your project's requirements are unsatisfiable.

Possible Solution

Relax openai constraint

• If CrewAI is compatible with newer OpenAI SDKs: widen to something like:
  • openai>=1.83.0,<3.0.0 (or <2.0.0 if 2.x is truly incompatible)

Additional context

Strict pins force users to fork or vendor CrewAI just to integrate with standard tooling. Relaxing constraints (plus a small compatibility test suite) would massively improve install success rates without sacrificing stability.