The UFO project is organized into a well-defined directory structure to facilitate development, deployment, and documentation. Below is an overview of each directory and file, along with their purpose:

📦project
 ┣ 📂documents               # Folder to store project documentation
 ┣ 📂learner                 # Folder to build the vector database for help documents
 ┣ 📂model_worker            # Folder to store tools for deploying your own model
 ┣ 📂record_processor        # Folder to parse human demonstrations from Windows Step Recorder and build the vector database
 ┣ 📂dataflow                # Folder for the code of data collection pipeline for Large Action Model (LAM)
 ┣ 📂vetordb                 # Folder to store all data in the vector database for RAG (Retrieval-Augmented Generation)
 ┣ 📂logs                    # Folder to store logs, generated after the program starts
 ┗ 📂ufo                     # Directory containing main project code
    ┣ 📂module               # Directory for the basic module of UFO, e.g., session and round
    ┣ 📂agents               # Code implementation of agents in UFO
    ┣ 📂automator            # Implementation of the skill set of agents to automate applications
    ┣ 📂experience           # Parse and save the agent's self-experience
    ┣ 📂llm                  # Folder to store the LLM (Large Language Model) implementation
    ┣ 📂prompter             # Prompt constructor for the agent
    ┣ 📂prompts              # Prompt templates and files to construct the full prompt
    ┣ 📂rag                  # Implementation of RAG from different sources to enhance agents' abilities
    ┣ 📂trajectory           # Implementation of loading and parsing trajectories of task completion
    ┣ 📂utils                # Utility functions
    ┣ 📂config               # Configuration files
        ┣ 📜config.yaml      # User configuration file for LLM and other settings
        ┣ 📜config_dev.yaml  # Configuration file for developers
        â”— ...
    ┗ 📄ufo.py               # Main entry point for the UFO client

Directory and File Descriptions

documents

  • Purpose: Stores all the project documentation.
  • Details: This may include design documents, user manuals, API documentation, and any other relevant project documentation.

learner

  • Purpose: Used to build the vector database for help documents.
  • Details: This directory contains scripts and tools to process help documents and create a searchable vector database, enhancing the agents' ability for task completion.

model_worker

  • Purpose: Contains tools and scripts necessary for deploying custom models.
  • Details: This includes model deployment configurations, and management tools for integrating custom models into the project.

dataflow

  • Purpose: Contains the code for the data collection pipeline for the Large Action Model (LAM).
  • Details: This directory includes scripts and tools for collecting and processing data to train the Large Action Model, improving the agents' performance and capabilities.

record_processor

  • Purpose: Parses human demonstrations recorded using the Windows Step Recorder and builds the vector database.
  • Details: This directory includes parsers, data processing scripts, and tools to convert human demonstrations into a format suitable for agent's retrieval.

vetordb

  • Purpose: Stores all data within the vector database for Retrieval-Augmented Generation (RAG).
  • Details: This directory is essential for maintaining the data that enhances the agents' ability to retrieve relevant information and generate more accurate responses.

logs

  • Purpose: Stores log files generated by the application.
  • Details: This directory helps in monitoring, debugging, and analyzing the application's performance and behavior. Logs are generated dynamically as the application runs.

ufo

  • Purpose: The core directory containing the main project code.

  • Details: This directory is further subdivided into multiple subdirectories, each serving a specific purpose within the project.

    module

    • Purpose: Contains the basic modules of the UFO project, such as session management and rounds.
    • Details: This includes foundational classes and functions that are used throughout the project.

    agents

    • Purpose: Houses the code implementations of various agents in the UFO project.
    • Details: Agents are components that perform specific tasks within the system, and this directory contains their logic, components, and behavior.

    automator

    • Purpose: Implements the skill set of agents to automate applications.
    • Details: This includes scripts and tools that enable agents to interact with and automate tasks in various applications, such as mouse and keyboard actions and API calls.

    experience

    • Purpose: Parses and saves the agent's self-experience.
    • Details: This directory contains mechanisms for agents to learn from their actions and outcomes, improving their performance over time.

    llm

    • Purpose: Stores the implementation of the Large Language Model (LLM).
    • Details: This includes the implementation of APIs for different language models, such as GPT, Genimi, QWEN, etc., that are used by the agents.

    prompter

    • Purpose: Constructs prompts for the agents.
    • Details: This directory includes prompt construction logic and tools that help agents generate meaningful prompts for user interactions.

    prompts

    • Purpose: Contains prompt templates and files used to construct the full prompt.
    • Details: This includes predefined prompt structures and content that are used to create meaningful interactions with the agents.

    rag

    • Purpose: Implements Retrieval-Augmented Generation (RAG) from different sources to enhance the agents' abilities.
    • etails: This directory includes scripts and tools for integrating various data sources into the RAG framework, improving the accuracy and relevance of the agents' outputs.

    trajectory

    • Purpose: Implements loading and parsing of task completion trajectories.
    • Details: This directory includes tools and scripts to load and parse task completion trajectories, enabling agents to learn from past experiences or for evaluation purposes.

    utils

    • Purpose: Contains utility functions.
    • Details: This directory includes helper functions, common utilities, and other reusable code snippets that support the project's operations.

    config

    • Purpose: Stores configuration files.
    • Details: This directory includes different configuration files for various environments and purposes.
    • config.yaml: User configuration file for LLM and other settings. You need to rename config.yaml.template to config.yaml and edit the configuration settings as needed.
    • config_dev.yaml: Developer-specific configuration file with settings tailored for development purposes.

    ufo.py

    • Purpose: Main entry point for the UFO client.
    • Details: This script initializes and starts the UFO application.