The Tryolabs 2025 Python library review, their 11th annual list, highlights the significant influence of Large Language Models (LLMs) and agent frameworks on the Python ecosystem while maintaining a balanced focus across general development and AI/ML/Data domains. Key trends observed include enhanced performance through Rust integration, advanced LLM integration, agent automation, and improved security and maintainability.

The report categorizes notable libraries into "General Development Top 10," "AI/ML/Data Top 10," "Runners-up," and a "Long Tail" of niche tools.

General Development Top 10 Libraries:

1. ty: A Rust-written, ultra-fast Python type checker. It leverages Salsa for function-level incremental analysis, which significantly enhances IDE responsiveness. It automatically recognizes project structures, detects .venv environments, and supports pyproject.toml. This tool, from the Astral team, continues the trend of modernizing Python tooling alongside Ruff and uv.
2. complexipy: A Rust-implemented tool designed to measure the cognitive complexity of code. It quantifies code structures that are difficult for humans to comprehend, based on SonarSource research, enabling fast analysis of large codebases via CLI, Python API, VS Code extensions, and CI/CD integration.
3. Kreuzberg: A multilingual document intelligence framework capable of processing over 50 file formats including PDF, Office documents, images, and HTML. It provides language bindings for Python, TypeScript, and Go, with deployment options including CLI, REST API, Docker, and MCP servers.
4. throttled-py: A request rate limiting library offering 2.5 to 4.5 times faster performance. It supports five algorithms: Fixed Window, Sliding Window, Token Bucket, Leaky Bucket, and Generic Cell Rate Algorithm (GCRA), with support for in-memory or Redis storage for both synchronous and asynchronous contexts.
5. httptap: A tool for detailed timing analysis and visualization of HTTP requests. It measures distinct phases such as DNS resolution, TCP connection, TLS handshake, server wait time, and response transfer, presenting data in a terminal waterfall view, JSON, or metrics output.
6. fastapi-guard: A security middleware solution for FastAPI, providing functionalities like IP whitelisting/blacklisting, rate limiting, XSS/SQL injection detection, and geographical filtering. It integrates with Redis for distributed environments and automatically configures OWASP security headers.
7. modshim: Enables extension of existing libraries through module overlay, a robust alternative to monkey-patching. It hooks into the Python import system to create virtual merged modules, allowing feature additions without modifying source code.
8. Spec Kit: GitHub's spec-driven development tool, translating specifications into executable blueprints that AI agents can use for implementation. It is compatible with AI tools like Copilot and Claude Code.
9. skylos: A dead code detection and security vulnerability analysis tool. It identifies unused functions, classes, and imports, and scans for dangerous patterns like SQL injection, providing a confidence score (0-100) and integrating with VS Code and CI/CD pipelines.
10. FastOpenAPI: Automates OpenAPI document generation across various web frameworks (e.g., Flask, Django, Tornado). It leverages FastAPI-style decorator routing and Pydantic v2 validation for schema generation.

AI/ML/Data Top 10 Libraries:

11. MCP Python SDK & FastMCP: Implementations of the Model Context Protocol (MCP) for connecting LLMs with external data. The Anthropic official SDK and Prefect's FastMCP 2.0 offer complementary capabilities, supporting OAuth 2.1, enterprise authentication, and OpenAPI/FastAPI integration.
12. TOON (Token-Oriented Object Notation): A compact JSON alternative for LLMs, achieving 40-60% token reduction through YAML-like indentation and CSV-like array structures. It maintains full JSON compatibility and is undergoing multi-language implementation.
13. Deep Agents: A LangChain-based framework for long-running LLM agents. It integrates LangGraph for streaming and persistent memory, featuring internal capabilities for planning, filesystem access, and sub-agent delegation.
14. smolagents: Hugging Face's lightweight, code-executing agent framework, designed with approximately 1,000 lines of Python code. It executes agent behaviors as Python code within secure sandboxed environments like E2B, Docker, and WebAssembly.
15. LlamaIndex Workflows: An event-driven AI workflow framework. It constructs asynchronous workflows using @step decorators and Event objects, supporting parallel execution, context object-based state management, and checkpoint restoration.
16. Batchata: A unified batch processing API for OpenAI, Anthropic, and Gemini models. It manages cost limits, automatic retries, interruption recovery, and structured output generation using Pydantic models.
17. MarkItDown: Microsoft's document-to-Markdown converter, supporting a wide range of formats including PDF, Word, PPT, Excel, images, and audio. It maintains an LLM-friendly structure and integrates with Azure Document Intelligence.
18. Data Formulator: Microsoft Research's AI-powered data visualization tool. It combines a visual interface with natural language input, automatically generating data transformation code. Visualizations are Vega-Lite based, with transparent exposure of underlying pandas or SQL code.
19. LangExtract: Google's precise text structure extraction library. It maps extracted text back to original character positions, providing visual evidence for extraction. It supports multiple LLM models (Gemini, OpenAI, Ollama) and is optimized for parallel processing.
20. GeoAI: OpenGeos's framework for integrating AI with geospatial information. It combines PyTorch, Transformers, and Leafmap for satellite image training and visualization, simplifying common geospatial analysis tasks such as land cover classification and change detection.

Runners-up and Long Tail:

Notable runners-up include AuthTuna for async Python authentication/authorization, FastRTC for real-time audio/video streaming from Python functions, hexora for static malware pattern detection, opentemplate for project templates, and Pyrefly (Meta's Rust-based type checker, similar to ty). The "Long Tail" comprises hundreds of niche libraries across various domains like AI agents, async processing, data pipelines, web development, and testing, indicating widespread experimentation and generational shifts within the Python ecosystem.

In conclusion, the 2025 Python ecosystem demonstrates a strong trend towards performance optimization via Rust, deep integration with LLMs and agent automation, and enhanced security and maintainability. The listed libraries collectively affirm Python's central role at the intersection of AI innovation and general software development.