DOCSEARCHbyAlgolia
Submit Docs

Hello-Agents Tutorial

repository·main·Indexed 13 days ago

https://github.com/datawhalechina/hello-agents

A comprehensive, hands-on tutorial for building AI-native agent architectures. The project includes various co-creation examples such as DataAnalysisAgent for tabular data analysis, a Database Agent Assistant for natural language to Oracle SQL conversion, YingQian movie recommender, and the GuessWhoAmI interactive game.

Tokens
479.8K
Snippets
1.1K
Records
1.8K
Agent score
96%

What's inside Hello-Agents

  1. Overview of FastGPT features and modules

    main

    FastGPT is an open-source, LLM-based knowledge base Q&A platform and Agent building tool designed as an "enterprise-grade AI productivity engine." It focuses on RAG (Retrieval-Augmented Generation) optimization through a complete pipeline: Data Import → Intelligent Chunking → Vector Retrieval → Dialogue Generation.

    Core Modules

    • Dialogue Portal: Interface for interacting with Agents.
    • Workbench: Central area for managing and creating Agents.
    • Knowledge Base: Management of uploaded documents and vector indices.
    • Account: User settings and subscription management.

    Agent Types

    In the Agent module, you can choose between:

    • Workflow: For complex, multi-step logic.
    • Dialogue Agent
    • Dialogue Agent V2 (Beta)

  2. Overview of Memory and RAG in HelloAgents

    main
    Chapter 8 introduces the expansion of the HelloAgents framework to include two critical capabilities: a Memory System and Retrieval-Augmented Generation (RAG). These systems allow agents to retain information from previous interactions and retrieve external knowledge, overcoming the limitations of stateless agents in continuous dialogues or complex tasks.

  3. Overview of Intelligent Stock Analysis Assistant

    main

    The Intelligent Stock Analysis Assistant is an A-share investment analysis tool built on the HelloAgents multi-agent collaboration framework. It integrates market data, financial analysis, news sentiment, intelligent stock screening, and simulated trading to provide data-driven investment decision support.

    Core Capabilities

    • Market & Financial Data: Real-time quotes, financial indicators, and company profiles.
    • AI-Driven Analysis: Sentiment analysis, data analysis reports, and a 'Buffett Evaluation' (value investing framework) using ReflectionAgents.
    • Agentic Features: An AI Chat Assistant that uses a Coordinator Agent to route user intent to specialized sub-agents (data, sentiment, or advisor).
    • Trading & Management: Simulated trading (buy/sell/cancel), watchlist management, and personalized investment preference settings.
    • Data Persistence: Local file caching for stock data and a memory system for daily snapshots.

    ⚠️ Disclaimer: All analysis results are for reference only and do not constitute investment advice.

  4. Overview of FastGPT features and workspace

    main

    FastGPT is an open-source, LLM-based knowledge base Q&A platform and Agent building tool designed as an "Enterprise-grade AI Productivity Engine." It specializes in RAG (Retrieval-Augmented Generation) solutions and visual workflow orchestration.

    Core Modules:

    • Dialog Portal: Interface for interacting with agents.
    • Workspace: Central area for managing agents and workflows.
    • Knowledge Base: Management of uploaded documents and vector indexes.
    • Account: User and plan management.

    Agent Types:

    • Workflow: For complex, logic-driven processes.
    • Dialog Agent: Standard conversational agents.
    • Dialog Agent V2 (Beta): Next-generation conversational agents.

    Free Plan Limits:

    • 100 credits
    • 600 knowledge base indexes
    • 1 team member
    • 10 Agents
    • 3 knowledge bases
    • 30-day conversation record retention
    • 30 QPM (Queries Per Minute) call rate
    • File upload: up to 5 files of 50MB each per batch.

  5. Overview of the Hello-Agents project

    main

    Hello-Agents is a systematic learning tutorial designed to teach developers how to build AI-native agent systems from scratch. Unlike software-engineering-type agents (e.g., Dify, Coze) that focus on process-driven workflows, this project focuses on the core principles, architectures, and paradigms of truly AI-driven agents.

    Key Learning Objectives:

    • Core Principles: Understand the history and classic paradigms of agents.
    • Hands-on Implementation: Master low-code platforms and agent code frameworks.
    • Custom Framework Development: Build a custom agent framework (HelloAgents) based on the OpenAI native API.
    • Advanced Agent Technologies: Implement context engineering, Memory, protocols, and evaluation systems.
    • Model Training: Learn Agentic RL, covering the full process from SFT to GRPO.
    • Real-world Projects: Develop practical applications like intelligent travel assistants and cyber towns.

  6. Overview of Geometry Problem Solving Agent (GPSAgent)

    main

    The Geometry Problem Solving Agent (GPSAgent) is a neuro-symbolic reasoning framework designed for high-precision geometry theorem proving and automatic problem solving. It integrates Large Language Models (LLMs) as 'planners' for high-level semantic understanding and reflection with a formal symbolic solver (FormalGeo) as the 'executor' for rigorous verification and theorem application.

    Key features include:

    • Neuro-symbolic Collaboration: Uses a 'Reasoning — Execution — Reflection — Memory' loop to eliminate LLM hallucinations by ensuring every derivation step is verifiable.
    • Unified Bidirectional Symbolic Reasoning Engine: Supports both forward reasoning (deriving conclusions from known conditions) and backward reasoning (decomposing goals into sub-goals), searching from both directions to find a meeting point in the state space.
    • Out-of-the-box Integration: Works on local machines without specific problem-annotated data and supports multiple LLM backends via tool-calling interfaces (theorem application, goal decomposition, fact querying, etc.).

  7. Overview of Memory and Retrieval in HelloAgents

    main
    Chapter 8 introduces two critical capabilities to the HelloAgents framework: a Memory System and Retrieval-Augmented Generation (RAG). These systems address the limitation of agents that cannot remember previous interactions or learn from historical experiences, which is essential for continuous conversations and complex task execution. The implementation follows a 'framework extension + knowledge popularization' approach, building upon the core architecture established in Chapter 7.

  8. Overview of Hello-Agents

    main
    Hello-Agents is a systematic learning tutorial from the Datawhale community designed to teach the design and implementation of AI-native Agent systems. Unlike workflow-driven software engineering approaches (like Dify or Coze), this project focuses on building agents driven by AI principles. It covers core principles, architectures, classic paradigms, and the hands-on construction of multi-agent applications.

  9. Overview of the Love Analyst Agent

    main

    The Love Analyst Agent is an LLM-based agent designed to help users navigate romantic confusion during ambiguous or established relationships. It analyzes uploaded chat logs (e.g., WeChat or QQ) to determine a partner's level of interest, interpret subtext, and provide emotional insights.

    Key Capabilities:

    • Chat Log Deep Analysis: Automatically cleans and extracts key emotional interaction nodes from text logs.
    • Heartbeat Index Assessment: Provides a quantitative score (0-100) and radar chart analysis based on dialogue frequency, intimacy of wording, and emotional value.
    • High-EQ Reply Generation: Suggests multiple response styles (e.g., humorous, deep, or 'push-pull') to break deadlocks or optimize daily conversation.

  10. Overview of Code Agent Prompts

    main

    The prompts/ directory centralizes the management of Code Agent prompts to facilitate independent iteration and comparison. The prompt architecture is divided into several specialized files:

    • system.md: Defines global behavior, security boundaries (e.g., confirmation for sensitive operations), and patch formats.
    • react.md: Defines the ReAct (Reasoning and Acting) loop format and tool input conventions.
    • plan.md: Dedicated prompt for the planning tool (plan[...]).
    • summarize_observation.md: Prompt used for summarizing tool outputs.

  12. Overview of Dify LLM Application Development Platform

    main

    Dify is an open-source LLM application development platform that combines Backend-as-a-Service (BaaS) and LLMOps. It provides a full lifecycle support from prototyping to production deployment using a layered modular architecture: Data Layer, Development Layer, Orchestration Layer, and Foundation Layer.

    Key features include:

    • Model Neutrality: Supports hundreds of LLMs (GPT, Deepseek, Llama, etc.) and any model compatible with the OpenAI API.
    • Deployment Flexibility: Supports local deployment via Docker Compose (for data privacy) or official SaaS cloud services.
    • Marketplace Ecosystem: Offers a one-stop plugin management system including Models, Tools, Agent Strategies, Extensions, and Bundles.
    • Developer Friendly: Supports remote debugging with popular IDEs by forwarding SaaS plugin operations to local environments.