[BUG] crewai 1.12.2 uninstallable on Intel Macs: mandatory lancedb>=0.29.2 dependency has no macOS x86_64 wheels

[ecoop]

Description

crewai 1.12.2 declares lancedb>=0.29.2 as a mandatory core dependency
(lib/crewai/pyproject.toml, line 44). lancedb dropped macOS x86_64
(Intel) wheel builds at 0.29.2, so crewai is uninstallable on all
Intel Macs — even for projects that never use memory features.

Steps to Reproduce

On an Intel Mac (macOS x86_64):

1. pip install crewai==1.12.2
   → ResolutionImpossible error

2. For a clearer diagnosis, use uv:
   uv pip install "lancedb>=0.29.2"

   Output confirms:
   "lancedb>=0.29.2 has no wheels with a matching platform tag
   (e.g., macosx_15_0_x86_64)"

   Available platforms for lancedb 0.29.2–0.30.2:
   manylinux_2_17_aarch64, manylinux_2_17_x86_64,
   manylinux_2_28_aarch64, manylinux_2_28_x86_64,
   manylinux2014_aarch64, manylinux2014_x86_64,
   macosx_11_0_arm64, win_amd64

   macOS x86_64 is absent.

Expected behavior

crewai should be installable on Intel Macs. lancedb is only required
for memory features, which many projects do not use. It should be an
optional dependency, e.g.:

pip install crewai # no lancedb required
pip install crewai[memory] # pulls in lancedb for memory features

Screenshots/Code snippets

% uv pip install "lancedb>=0.29.2"
Using Python 3.13.1 environment at: venv-crewai
× No solution found when resolving dependencies:
╰─▶ Because only the following versions of lancedb are available:
lancedb<=0.29.2
lancedb==0.30.0
lancedb==0.30.1
lancedb==0.30.2
and lancedb>=0.29.2 has no wheels with a matching platform tag
(e.g., macosx_15_0_x86_64), we can conclude that
lancedb>=0.29.2 cannot be used.
And because you require lancedb>=0.29.2, we can conclude that
your requirements are unsatisfiable.

  hint: Wheels are available for `lancedb` (v0.30.2) on the
  following platforms: `manylinux_2_17_aarch64`,
  `manylinux_2_17_x86_64`, `manylinux_2_28_aarch64`,
  `manylinux_2_28_x86_64`, `manylinux2014_aarch64`,
  `manylinux2014_x86_64`, `macosx_11_0_arm64`, `win_amd64`

Operating System

macOS Sonoma

Python Version

3.10

crewAI Version

1.12.2

crewAI Tools Version

1.9.3

Virtual Environment

Venv

Evidence

Confirmed via uv pip install "lancedb>=0.29.2" — error message
explicitly lists supported platforms, macOS x86_64 is absent.

Smoking gun: lib/crewai/pyproject.toml line 44:
"lancedb>=0.29.2"
Listed as a core dependency with no extras qualifier.

Source:
https://raw.githubusercontent.com/crewAIInc/crewAI/1.12.2/lib/crewai/pyproject.toml

Possible Solution

Make lancedb an optional dependency under a [memory] extras group:

[project.optional-dependencies]
memory = ["lancedb>=0.29.2"]

This is a one-line change in pyproject.toml and unblocks all Intel
Mac users without affecting Apple Silicon or Linux users at all.

Additional context

Workaround for projects not using memory features:
pip install crewai==1.12.2 --no-deps

This is not a viable general solution as it bypasses all dependency
checks. It works only if you manually manage all other dependencies.

Note: lancedb 0.29.2 does exist on PyPI (tagged python-v0.29.2 on
GitHub) but simply has no Intel Mac wheels. This has been reported
separately to the lancedb project to confirm whether the platform
drop was intentional.

Note: this bug template does not include macOS Sequoia (15.x) or
Python 3.13 — both current releases. Selected Sonoma / 3.12 as
the closest available options. Actual environment: macOS Sequoia
15.x, Python 3.13.1.

[ecoop]

Tested crewai 1.14.0 — still affected. The constraint has changed to lancedb>=0.29.2,<0.30.1 but the platform problem remains. Looking forward to the crewai[memory] fix in #5328.

[rehan243]

I've encountered similar issues with platform-specific wheels in the past, particularly when working with voice AI systems that needed to support a wide range of hardware. Here are a few suggestions:

• Check for alternative builds: Sometimes, the maintainers provide alternative builds or source installations. For lancedb, you might want to check if there's a source distribution that can be built locally on Intel Macs.
• Conditional dependencies: Modify the pyproject.toml to make lancedb an optional dependency. This way, users who don't need the memory features won't be affected. You can use the extras feature in pyproject.toml to achieve this.
  • Example:

    [project.optional-dependencies]
    memory = ["lancedb>=0.29.2"]

• Local wheel building: As a temporary workaround, users can build the wheel locally using tools like cibuildwheel or delocate. This approach worked well for us when we needed to support older hardware for our voice AI systems.
• Reach out to maintainers: Engage with the lancedb maintainers to understand why they dropped support for macOS x86_64. They might have insights or plan to reinstate support.