Merge branch 'audit-ai'

README.md

@@ -141,11 +141,15 @@ _Libraries for building AI applications, LLM integrations, and autonomous agents
 - Orchestration
   - [langchain](https://github.com/langchain-ai/langchain) - A framework for building agents and LLM-powered applications.
   - [langgraph](https://github.com/langchain-ai/langgraph) - Low-level orchestration framework for building stateful, long-running LLM agents.
-  - [crewai](https://github.com/crewAIInc/crewAI) - A framework for orchestrating role-playing autonomous AI agents for collaborative task solving.
   - [pydantic-ai](https://github.com/pydantic/pydantic-ai) - A Python agent framework for building generative AI applications with structured schemas.
+  - [crewai](https://github.com/crewAIInc/crewAI) - A framework for orchestrating role-playing autonomous AI agents for collaborative task solving.
 - Vendor Agent SDKs
-  - [openai-agents](https://github.com/openai/openai-agents-python) - OpenAI's framework for building and managing AI agents.
   - [claude-agent-sdk](https://github.com/anthropics/claude-agent-sdk-python) - Anthropic's Python SDK for building AI agents on Claude Code's harness — custom tools, in-process MCP servers, hooks.
+  - [openai-agents](https://github.com/openai/openai-agents-python) - OpenAI's framework for building and managing AI agents.
+  - [google-adk](https://github.com/google/adk-python) - Google's code-first toolkit for building, evaluating, and deploying AI agents.
+- Model Context Protocol
+  - [mcp](https://github.com/modelcontextprotocol/python-sdk) - The official Python SDK for building Model Context Protocol servers and clients.
+  - [fastmcp](https://github.com/PrefectHQ/fastmcp) - A high-level, Pythonic framework for building MCP servers and clients.
 - Personal Assistants
   - [hermes-agent](https://github.com/nousresearch/hermes-agent) - An adaptive personal AI assistant that grows with you.
   - [AstrBot](https://github.com/AstrBotDevs/AstrBot) - A multi-platform AI assistant that connects LLMs to chat apps like Telegram, Slack, and QQ, extensible with Python plugins.
@@ -164,19 +168,19 @@ _Libraries for building AI applications, LLM integrations, and autonomous agents
   - [vllm](https://github.com/vllm-project/vllm) - A high-throughput and memory-efficient inference and serving engine for LLMs.
   - [mlx-lm](https://github.com/ml-explore/mlx-lm) - Run and fine-tune large language models on Apple Silicon with MLX.
 - LLM Gateways
-  - [LiteLLM](https://github.com/BerriAI/litellm) - Call 100+ LLMs using OpenAI format.
+  - [litellm](https://github.com/BerriAI/litellm) - Call 100+ LLMs using OpenAI format.
 - Image and Video Generation
   - [diffusers](https://github.com/huggingface/diffusers) - A library that provides pre-trained diffusion models for generating and editing images, audio, and video.
 - Fine-tuning
   - [peft](https://github.com/huggingface/peft) - A library for parameter-efficient fine-tuning of large pretrained models.
-  - [unsloth](https://github.com/unslothai/unsloth) - A library for faster LLM fine-tuning and training with reduced memory usage.
+  - [trl](https://github.com/huggingface/trl) - A library for post-training transformer language models with SFT, DPO, GRPO, and other trainers.
+  - [unsloth](https://github.com/unslothai/unsloth) - Faster, lower-memory LLM fine-tuning, as a Python library or a desktop app.
   - [axolotl](https://github.com/axolotl-ai-cloud/axolotl) - A framework for fine-tuning and post-training large language models.
 - Speech
+  - [faster-whisper](https://github.com/SYSTRAN/faster-whisper) - A Whisper reimplementation on CTranslate2, up to 4 times faster than openai-whisper with less memory.
   - [openai-whisper](https://github.com/openai/whisper) - A general-purpose automatic speech recognition model trained on 680k hours of multilingual and multitask supervised data.
-  - [funasr](https://github.com/modelscope/FunASR) - Industrial-grade speech recognition toolkit with speaker diarization and emotion detection.
-  - [vibevoice](https://github.com/microsoft/VibeVoice) - A family of open-source voice AI models from Microsoft for text-to-speech and long-form speech recognition.
   - [gTTS](https://github.com/pndurette/gTTS) - Python library and CLI tool for converting text to speech using Google Translate TTS.
-  - [kittentts](https://github.com/KittenML/KittenTTS) - Lightweight ONNX text-to-speech library with small CPU-friendly models.
+  - [funasr](https://github.com/modelscope/FunASR) - Industrial-grade speech recognition toolkit with speaker diarization and emotion detection.
 
 ### Deep Learning
 

docs/audit-logs.md

@@ -28,12 +28,14 @@ Entries admitted despite the Stable quality requirement, by maintainer decision:
 
 - zensical -- admitted 2026-08-23 at version 0.0.57, PyPI classifier `Development Status :: 3 - Alpha`, with no 1.0 target announced. Admitted as a challenger displacing mkdocs because the Documentation section was carrying a dying upstream: the Material for MkDocs team's announcement of 2025-11-05 put Material into maintenance mode for twelve months and called MkDocs itself unmaintained since 2024-08 and a supply chain risk. Re-check when 1.0 lands.
 - openviking -- admitted 2026-08-25 (PR #3302) at version 0.4.16, PyPI classifier `Development Status :: 3 - Alpha`, with a README that said "OpenViking is still in its early stages". Admitted as a Data Layer challenger because the adoption is real: 459,446 downloads/month on pypistats (mirror-excluded) on 2026-08-24, and releases every few days. Re-check when 1.0 lands.
+- claude-agent-sdk -- kept 2026-10-01 in the AI and Agents audit at version 0.2.163, PyPI classifier `Development Status :: 3 - Alpha`. Kept as a Vendor Agent SDKs obvious choice because the adoption is real: 30,033,689 downloads/month on ClickPy on 2026-10-01, up from 1.2M in March 2026 on pypistats. Re-check when 1.0 lands.
 
 ### Challenger Exceptions
 
 Entries admitted as challengers without adoption-trajectory evidence, by maintainer decision:
 
 - seleniumbase -- admitted 2026-08-16 (PR #3284) as a Browser Automation challenger. It wraps selenium, the successor trajectory in the use case belongs to playwright, and its distinct value is its stealth/UC mode. Admitted because that stealth value counts toward Browser Automation.
+- semantica -- kept 2026-10-01 in the AI and Agents audit as a Data Layer challenger at 16,582 downloads/month on ClickPy, with no adoption-trajectory evidence found. Kept by maintainer decision.
 
 ### Mature-stable Keeps