AI框架与项目教程合集

This repository will have different projects using AutoGen and Tutorials

AI Projects & Learning Repository A collection of AI agent projects, framework tutorials, and integrations. This is the main hub for code from my YouTube channel. Looking for full courses? Check out ai-courses for structured learning paths (LangChain, LangGraph, Atomic Agents, etc.) Repository Structure ├── frameworks/ # AI framework projects (simple examples, not full courses) │ ├── agentops/ # Agent monitoring/observability │ ├── autogen/ # AutoGen projects (includes v0.7+ examples) │ ├── crewai/ # CrewAI projects (includes v1.15+ features) │ ├── openai/ # OpenAI SDK, Swarm, Agents, Image Gen │ ├── pydanticai/ # PydanticAI examples (includes v2.x features) │ └── smolagents/ # Smolagents examples (includes v1.26+ features) │ ├── mini_projects/ # Standalone tutorial projects │ ├── 00_getting_started │ ├── 01_workout │ ├── 02_lmstudio │ ├── 03_video_captions │ └── ...more │ ├── integrations/ # Third-party integrations │ ├── google_drive_monitor │ ├── mcp_crewai │ ├── mcp_supabase │ ├── n8n_crewai │ └── n8n_runner │ ├── tools/ # Utilities (not frameworks) │ ├── crawl4ai/ # Web scraping tool │ └── repo_images/ │ ├── youtube_shorts/ # Code for YouTube Shorts videos │ └── claude_skill_agent_builder/ # Claude skill for building AI agents Framework Versions (October 2026) Every project in the repo was updated for these versions and smoke-tested offline (LLM calls faked; external services stubbed). A few projects run on older pins because of upstream caps: CrewAI 1.15 requires openai<3 and mcp~=1.28, litellm requires openai<3, and the archived Swarm needs openai 3.3. Framework Package / version Highlights AutoGen autogen-agentchat 0.7.5 ⚠️ Breaking change from v0.2! New async API, event-driven CrewAI crewai[tools] 1.15 Conversational flows, LLM overlay, streaming PydanticAI pydantic-ai 2.58 Typed outputs (output_type), tools with context, streaming Smolagents smolagents 1.26 Code agents, managed agents, LiteLLM models OpenAI Agents SDK openai-agents 0.24 Agents, tools, handoffs OpenAI Agents API (hosted) openai 3.35 Hosted sandbox, sessions, subagents (openai_agents_058) AgentOps agentops 0.4 Agent monitoring and observability Default model in the examples is gpt-5.4-mini. Image examples use gpt-image-2.5-flare / -sunburst (gpt-image-1 shuts down 2026-10-23). Framework Quick Reference AutoGen v0.7+ (Breaking Changes!) The old import autogen API is deprecated. New pattern: from autogen_agentchat.agents import AssistantAgent from autogen_agentchat.ui import Console from autogen_ext.models.openai import OpenAIChatCompletionClient model_client = OpenAIChatCompletionClient( model="gpt-5.4-mini", # AutoGen 0.7.5 predates gpt-5.4-mini, so its capabilities must be given explicitly model_info={"vision": True, "function_calling": True, "json_output": True, "family": "gpt-5", "structured_output": True}, ) agent = AssistantAgent(name="my_agent", model_client=model_client) await Console(agent.run_stream(task="...")) # or: result = await agent.run(task="...") See frameworks/autogen/autogen_v07_quickstart/ for examples. All AutoGen examples in this repo use the v0.7 API. Many were first recorded on AutoGen 0.2, so the code may differ from what's on screen in older videos. CrewAI v1.15+ from crewai import Agent, Crew, Task, LLM from crewai.flow import Flow, listen, start See frameworks/crewai/crewai_v115_features/ for new features. PydanticAI v2.x from pydantic_ai import Agent agent = Agent("openai:gpt-5.4-mini", output_type=MyModel) result = await agent.run("...") print(result.output) See frameworks/pydanticai/pydanticai_v2_features/ for examples. Skills / Claude Code agent-builder - A Claude Code skill that scaffolds complete AI agent projects. Give it a task and framework, get a working project with proper structure, dependencies, and best practices. Getting Started Prerequisites Python 3.11–3.13 (3.12 recommended; CrewAI doesn't support 3.14 yet) Ollama (for local LLMs) LM Studio (alternative local LLM option) IDE Options VS Code with Python extension PyCharm Cursor (AI-powered IDE) Running Examples Most examples are standalone. Each project folder has its own README; check it first. There are two layouts. Plain script folders ship a requirements.txt: cd frameworks/crewai/crewai_tools_11 pip install -r requirements.txt python csv_file_rag_search.py CrewAI crew and flow projects ship a pyproject.toml (with [tool.crewai]) instead and run through the CrewAI CLI: cd frameworks/crewai/crewai_first_crew/first_crew crewai install crewai run Put your OPENAI_API_KEY (and any other keys the project README lists) in a .env file in the project folder. Related Repositories Repository Description ai-courses Full structured courses (LangChain, LangGraph, Atomic Agents) master-crewai-course Comprehensive CrewAI course YouTube • GitHub