Big news for Python devs in VS Code: Pylance now lets you pick your type checker. π
Meet the Type Server Protocol (TSP), built with Meta so Pyrefly plugs straight into Pylance. Keep Copilot + IntelliSense, swap the engine.
Check it out πβ¨https://t.co/2X9Cymjptp
Pyrefly support for tensor shape checking is here! Using simple type annotations, Pyrefly can catch tricky bugs in your models. Validated on dozens of real-world models including nanoGPT. Works with PyTorch, JAX, NumPy and more!
https://t.co/35KTXxLcYP
If you've ever struggled with shape bugs in PyTorch, NumPy or JAX you can use Pyrefly shape checking to catch mismatches statically, before you ever run your code:
https://t.co/BY46qafjQS
#pyrefly#python#pytorch#numpy
The Pyrefly team wants to hear from you!
If you'd like to help make Pyrefly better and have a spare 10 mins to fill out this survey we would greatly appreciate it - whether youβre just getting started trying out Pyrefly or youβve been here since the beginning:
π https://t.co/MCFfP1Z3yk
π Python 3.15 is here! One feature worth trying is explicit lazy imports.
They let you defer loading a module until your code actually uses it. For applications with costly imports, that can mean a much faster startup thanks to the use of built-in Python syntax.
Lazy imports arrived in PyCharm 2026.2 and will be enhanced further in 2026.3. Try them in the 2026.3 EAP now.
Curious how they work? Cheuk walks you through the feature and benchmarks it in our blog post: https://t.co/EqnowFtbxN
@OneJumpInc@Meta Nice work! Thanks for putting Pyrefly through its paces and sharing this demo. We're eager to see if this catches more errors over the next weeks and months. Keep us posted! @OneJumpInc
@MetaOpenSource Read how our Python team transitioned from a homegrown checker to collaborating with the Pyrefly community on shared open-source infrastructure: https://t.co/HNldeSdPUY
The best open source tool for the job doesn't have to be built in-house. When evaluating Python type checkers, Google chose @MetaOpenSource's Pyrefly and started contributing upstream. π Link in first reply
Tried @pyrefly_dev from @Meta on a 135k-line scientific Python codebase: 637,003 lines checked in 0.50s, vs 22.3s cold before. All 149 findings were accurate.
Adopting it in CI. Thanks to the Meta team we met at @py_bay: Kyle Into, Steven Troxler, @__avik and Martin DeMello.
@TechWithTimm has a new video out featuring Pyrefly π₯
It covers the basics of getting started with type checking, including getting your agent to use Pyrefly
https://t.co/KNhKeykjVS
@TechWithTimm Really enjoyed this, @TechWithTimm. Especially appreciate that it starts with the bugs Python happily accepts, then shows a practical path to catching them earlier. Great walkthrough!
Python 3.15 is scheduled for tomorrow. Highlights include lazy imports, frozendict, sentinel, and new profiling tools.
Which feature are you trying first?
https://t.co/YytAZC9xaY
@PythonHub Very cool project! Noticed Python type hints, though I didnβt spot a checker in CI. Curious how folks are using hints these days: editor tooling, documentation, guiding agents, or running a checker locally? Are we seeing more annotations without CI checks?
@pycoders Appreciate the spotlight, @pycoders. Pyrefly is for anyone who wants to catch a bug before it becomes a debugging session- whether theyβve been type checking Python for years or never. Start small and keep the Python flowing π.
JetBrains Airβs thesis: as agents generate more code, the bottleneck moves to verification and ownership.
"The agent said it was done" is not a quality gate. Put deterministic checks in the loop.
@pycharm congrats on the launch!
https://t.co/3zEMG394gl
Just published my first blog! π₯³
I wrote about my open-source journey and how we migrated from `mypy` to `pyrefly` in @napari_imaging
https://t.co/plZDG9jqce
Hope it helps if you're planning a similar migration!
#OpenSource#Python#Typing#Pyrefly#Mypy
Congrats on the #Jev launch, @typesafeai π
We dug into the #Python adapter and were delighted to find Pyrefly running in strict mode across the source and tests.
Typed decisions π€ strictly typed client code. Typers unite.
https://t.co/msQAjMdbN0