Python Weekly (Issue 712 September 4 2025)

Welcome to issue 712 of Python Weekly. Let's get straight to the links this week.

Articles, Tutorials and Talks

Building your own CLI Coding Agent with Pydantic-AI

CLI coding agents are a fundamentally different tool to chatbots or autocomplete tools - they're agents that can read code, run tests, and update a codebase. While commercial tools are impressive, they don't understand the particular context of our environment and the eccentricities of our specific project. Instead we can build our own coding agent by assembling open source tools, using our specific development standards for: testing, documentation production, code reasoning, and file system operations.

Polars GPU Execution. (70% speed up)

Polars' new GPU engine, powered by NVIDIA RAPIDS cuDF, accelerates data processing up to 70% compared to CPU-based execution, enabling faster handling of large datasets. The beta release supports common operations, leveraging GPU parallel processing for significant performance gains in data analytics workflows.

TIL: Using SQLModel Asynchronously with FastAPI (and Air) with PostgreSQL

This post explains how to leverage SQLModel with FastAPI and PostgreSQL to enable fully asynchronous database operations, improving scalability and efficiency for concurrent web applications. Key steps include setting up async database engines and sessions, using dependency injection in FastAPI, and aligning everything with non-blocking patterns.

Speeding up PyTorch inference by 87% on Apple devices with AI-generated Metal kernels

The post describes how AI models can automatically generate optimized Metal GPU kernels that speed up PyTorch inference on Apple devices by an average of 87% across 215 modules, with some kernels running hundreds of times faster than baseline. Using an agentic swarm approach and adding context like CUDA references and profiling data, the system outperforms standalone models, making kernel optimization more accessible and efficient without requiring kernel engineering expertise.

How to Send SMS Messages With Python, Flask and Vonage

The tutorial explains how to send SMS messages using Python and Flask by integrating the Vonage API, covering both SMS sending and basic web app setup. It provides a step-by-step guide including environment setup, API key management, Flask view creation, HTML form design, and message delivery verification. SPONSOR

Sharing a mutable reference with Python

The post explains how to share mutable references between Rust and Python using PyO3, addressing lifetime and ownership challenges by leveraging Rust’s std::mem::take, std::mem::replace, and smart pointers like Arc and Mutex. This approach enables safe mutable access to Rust data from Python, preserving correctness and concurrency while overcoming PyO3’s limitations with Rust lifetimes.

PydanticAI: the AI Agent Framework Winner

The video showcases how to use Pydantic AI to build Python applications with AI-powered agents that provide validated, structured outputs by integrating large language models like GPT-5. It demonstrates a healthcare triage assistant that personalizes responses using domain data, dependencies, and customizable prompts, enabling robust, real-world AI integration beyond simple chatbots.

How I write Django views

The author advocates using Django's base View class over generic class-based or function-based views for simplicity and flexibility in handling HTTP requests. By avoiding complex mixins and leveraging straightforward helper methods, developers can write clearer, more maintainable view code with minimal cognitive overhead.

Python: capture stdout and stderr in unittest

The article explains how to capture stdout and stderr during Python unittest runs using contextlib.redirect_stdout and redirect_stderr, enabling tests to programmatically access console output. It also provides examples and custom context managers to simplify capturing both streams simultaneously, improving test logging and debugging capabilities.

Python has had async for 10 years -- why isn't it more popular?

Python's async/await has been available for 10 years but remains less popular due to its complexity, limited use cases mostly for I/O-bound tasks, and challenges like the Global Interpreter Lock (GIL) which restricts true parallelism. Additionally, maintaining both sync and async codebases, limited async support in key libraries, and confusing APIs hinder widespread adoption despite success in async-native frameworks like FastAPI.

Scheduling Background Tasks in Python with Celery and RabbitMQ

We'll build background tasks using Celery and RabbitMQ to create a weather notification service.

vLLM with torch.compile: Efficient LLM inference on PyTorch

Learn how to optimize PyTorch code with minimal effort using torch.compile, a just-in-time compiler that generates optimized kernels automatically.

Build an AI Coding Agent in Python

This tutorial teaches how to build a functional agentic AI coding assistant in Python using the free Gemini Flash API, covering agentic loops, tool-calling, file manipulation, and autonomous debugging. By constructing an agent that can read, modify, and execute code, viewers gain practical skills and deep insight into how modern coding agents operate beneath the surface.

When You No Longer Need That Object • Dealing With Garbage in Python

Let's explore reference counting and cyclic garbage collection in Python.

Interesting Projects, Tools, and Libraries

Elysia

Elysia is an agentic platform designed to use tools in a decision tree. A decision agent decides which tools to use dynamically based on its environment and context.

Zuban

Zuban is a high-performance Python Language Server and type checker implemented in Rust, by the author of Jedi. Zuban is 20–200× faster than Mypy, while using roughly half the memory and CPU compared to Ty and Pyrefly. It offers both a PyRight-like mode and a Mypy-compatible mode, which behaves just like Mypy; supporting the same config files, command-line flags, and error messages.

toolfront

Simple data retrieval for AI with unmatched control, precision, and speed.

WhisperLiveKit

Real-time & local speech-to-text, translation, and speaker diarization. With server & web UI.

Youtu-agent

A simple yet powerful agent framework that delivers with open-source models.

LEANN

RAG on Everything with LEANN. Enjoy 97% storage savings while running a fast, accurate, and 100% private RAG application on your personal device.

GENIE

Experience near-instantaneous speech synthesis on your CPU.

oLLM

oLLM is a lightweight Python library for large-context LLM inference, built on top of Huggingface Transformers and PyTorch. It enables running models like Llama-3.1-8B-Instruct on 100k context using ~$200 consumer GPU with 8GB VRAM. Example performance: ~20 min for the first token, ~17s per subsequent token.

Kronos

A Foundation Model for the Language of Financial Markets.

sync-with-uv

The sync-with-uv package automates version synchronization between uv.lock and .pre-commit-config.yaml, ensuring consistent dependency management for tools like black, ruff, and mypy. It integrates as a pre-commit hook, streamlining workflows by aligning versions from a single source while leaving unspecified tools unchanged.

PageIndex

PageIndex is a reasoning-based RAG system that simulates how human experts navigate and extract knowledge from long documents through tree search, enabling LLMs to think and reason their way to the most relevant document sections.

yesglot

LLM-powered Django translations. Just call me "python manage.py translatemessages"

New Releases

Django security releases issued: 5.2.6, 5.1.12, and 4.2.24

The Django team released security updates for versions 5.2.6, 5.1.12, and 4.2.24 to fix a high-severity SQL injection vulnerability in FilteredRelation column aliases affecting multiple supported branches. Users are strongly encouraged to upgrade promptly to these patched versions to mitigate the risk.

Upcoming Events and Webinars

San Francisco Python Meetup September 2025

There will be following talks

• Powerful Doctesting

• See Python Logs, Traces, and Errors in One Place

PyData Lisbon Meetup September 2025

There will be following talks

• Software Design: why it matters and how should we approach it?

• Agentic AI: Hype, Reality, and What’s Next – Round Table

PyData Cornwall - Meetup #2

There will be following talks

• How to manage a Series A startup's cloud infrastructure using Python

• Would I want to talk to me right now? Own awkward and feel confident

• PAWS and start analysing Wikipedia's data in 5 minutes

PyData Amsterdam Meetup September 2025

There will be a talk, Building an agent network on your own device (like Jetson Nano).

MadPy Meetup September 2025

There will be A Crash Course in Causal Inference.

Programmer Weekly - A free weekly newsletter for programmers.

Founder Weekly - A free weekly newsletter for entrepreneurs featuring best curated content, must read articles, how to guides, tips and tricks, resources, events and more.