Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
25 commits
Select commit Hold shift + click to select a range
e98650a
Initial plan
Copilot Feb 19, 2026
3476998
Add Cognee dependency and create cognee_memory.py, worldview_builder.py
Copilot Feb 19, 2026
b40b405
Integrate Cognee memory system and worldview builder with GUI
Copilot Feb 19, 2026
3b5e916
Add Cognee integration tests and fix database test configuration
Copilot Feb 19, 2026
9b7fc46
Update README English section with Cognee and Worldview features
Copilot Feb 19, 2026
030af1a
Fix code review issues: API key priority, event loop leaks, unused im…
Copilot Feb 20, 2026
661b9e8
Improve JSON parsing robustness and add raw output logging for debugging
Copilot Feb 20, 2026
f07e34f
Configure Cognee to use SiliconFlow as custom LLM provider instead of…
Copilot Feb 20, 2026
ca16b48
Improve error logging in CogneeMemoryManager to show detailed excepti…
Copilot Feb 20, 2026
ca7f2e0
Improve Cognee configuration: add timeout, better logging, API key va…
Copilot Feb 20, 2026
37425f5
Fix Cognee LLM configuration: use set_llm_config for complete setup, …
Copilot Feb 20, 2026
a659a60
Fix Cognee config: use LLMConfig and EmbeddingConfig classes with Sil…
Copilot Feb 20, 2026
d4463c9
Fix Cognee model names: use SiliconFlow format directly without opena…
Copilot Feb 20, 2026
6494942
Fix Cognee config: use environment variables only (compatible with Co…
Copilot Feb 20, 2026
104428a
Fix LiteLLM provider prefix: add openai/ prefix for SiliconFlow models
Copilot Feb 21, 2026
d1b8afd
Make LiteLLM model prefix configurable via COGNEE_MODEL_PREFIX env var
Copilot Feb 21, 2026
ab2db3a
Remove misleading openai/ prefix, use model names directly for custom…
Copilot Feb 21, 2026
3d3c8b3
Add LLM_INSTRUCTOR_MODE config for Cognee structured output control
Copilot Feb 21, 2026
65d3aa5
Fix Cognee LLM config: use openai provider with OPENAI_API_BASE for S…
Copilot Feb 21, 2026
f851640
Fix embedding tokenizer error: add HUGGINGFACE_TOKENIZER config for S…
Copilot Feb 21, 2026
e33e97e
Fix embedding tokenizer: register model variants in litellm.embedding…
Copilot Feb 21, 2026
c40c52d
Add log_warning method to DebugLogger class
Copilot Feb 21, 2026
ab1ce79
Fix tiktoken tokenizer error: register model variants in tiktoken.mod…
Copilot Feb 21, 2026
8ee9ed7
Add HuggingFace tokenizer with SSL fix and mirror support for China
Copilot Feb 21, 2026
184a165
Add standalone Cognee manager app with log viewer, remove tiktoken
Copilot Feb 21, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
53 changes: 51 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@ Neo Agent 是一个基于 LangChain + LangGraph 的智能对话代理系统,
#### 🏗️ 复合框架架构
- **LangChain**: 核心框架,提供LLM抽象和链式调用
- **LangGraph**: 状态图管理,实现复杂对话流程编排和动态多智能体协作
- **Cognee**: 智能持久化动态AI记忆引擎([查看详情](docs/COGNEE_INTEGRATION.md))
- **DeepAgents**: 子智能体生成、长期记忆和文件系统([查看详情](docs/DEEPAGENTS_INTEGRATION.md))
- **多层模型架构**: 根据任务类型智能选择模型
- 主模型 (DeepSeek-V3.2): 处理主要对话、复杂推理和任务编排
Expand All @@ -23,7 +24,8 @@ Neo Agent 是一个基于 LangChain + LangGraph 的智能对话代理系统,

### 主要特性

- 🧠 **分层记忆系统**: 短期记忆、长期记忆、知识库、基础知识
- 🧠 **Cognee智能记忆**: 基于向量搜索和图数据库的持久动态AI记忆
- 🌍 **世界观构建**: 基于Markdown的可编辑世界观系统,支持AI辅助生成
- 💭 **智能对话**: 角色扮演、连续对话、记忆检索、情感理解
- 📊 **情感分析**: 印象评估、累计评分、关系可视化
- 🖥️ **现代化GUI**: 基于Tkinter的友好界面
Expand Down Expand Up @@ -136,6 +138,36 @@ Neo Agent采用模块化的提示词工程系统,参考了SillyTavern的设计

详见 `prompts/README.md` 获取完整说明。

### Cognee 智能记忆系统

Neo Agent 集成了 [Cognee](https://docs.cognee.ai/) 作为智能记忆引擎:

- 🧠 **持久动态记忆**: 将原始数据转化为智能体的持久动态AI记忆
- 🔗 **知识图谱**: 自动构建实体关系图,使知识相互关联
- 🔍 **语义搜索**: 基于向量的语义检索,按含义搜索记忆
- 📊 **模块化知识**: 支持自定义知识块和记忆类型

### 世界观构建系统

基于Markdown的智能体世界观构建系统:

- 📝 **直接编辑**: 直接编辑Markdown源文件管理世界观
- 🤖 **AI生成**: 使用自然语言描述,AI辅助生成详细世界观
- 🔄 **知识同步**: 自动同步到Cognee记忆和知识库
- 📁 **模块化**: 将世界观解析为模块化知识块

世界观文件位于 `prompts/worldview/` 目录。

### 鸣谢

感谢以下开源项目对 Neo Agent 的支持:

- **[Cognee](https://github.com/topoteretes/cognee)** - 开源知识引擎,将原始数据转化为持久动态AI记忆
- **[LangChain](https://github.com/langchain-ai/langchain)** - 构建LLM应用的强大框架
- **[LangGraph](https://github.com/langchain-ai/langgraph)** - 状态图管理和多智能体协作
- **[DeepAgents](https://github.com/deepagents/deepagents)** - 子智能体生成和持久化状态管理
- **[SillyTavern](https://github.com/SillyTavern/SillyTavern)** - 提示词工程系统的设计参考

### 许可证

本项目采用 MIT 许可证。详见 [LICENSE](LICENSE) 文件。
Expand All @@ -153,14 +185,17 @@ Neo Agent is a LangChain + LangGraph-based intelligent conversation agent system
#### 🏗️ Composite Framework Architecture
- **LangChain**: Core framework providing LLM abstraction and chain invocation
- **LangGraph**: State graph management for complex conversation flow orchestration
- **Cognee**: Intelligent persistent and dynamic AI memory engine ([documentation](docs/COGNEE_INTEGRATION.md))
- **DeepAgents**: Sub-agent generation, long-term memory, and file system ([documentation](docs/DEEPAGENTS_INTEGRATION.md))
- **Multi-tier Model Architecture**: Intelligent model selection based on task type
- Main Model (DeepSeek-V3.2): Handles primary conversations and complex reasoning
- Tool Model (GLM-4.6V): Handles tool invocations, intent recognition, and lightweight tasks
- Multimodal Model (Qwen3-VL-32B): Handles multimodal recognition and reasoning

### Key Features

- 🧠 **Hierarchical Memory System**: Short-term memory, long-term memory, knowledge base, base knowledge
- 🧠 **Cognee Smart Memory**: Persistent and dynamic AI memory based on vector search and graph databases
- 🌍 **Worldview Builder**: Editable Markdown-based worldview system with AI-assisted generation
- 💭 **Intelligent Conversation**: Role-playing, continuous dialogue, memory retrieval, emotional understanding
- 📊 **Emotion Analysis**: Impression assessment, cumulative scoring, relationship visualization
- 🖥️ **Modern GUI**: User-friendly Tkinter-based interface
Expand Down Expand Up @@ -209,6 +244,8 @@ Neo_Agent/
│ ├── gui/ # GUI modules
│ ├── tools/ # Utility modules
│ └── nps/ # NPS tool system
├── prompts/ # Prompt templates
│ ├── worldview/ # Worldview settings
├── tests/ # Test files
├── examples/ # Example code
├── main.py # Main entry point (recommended)
Expand All @@ -220,6 +257,8 @@ Neo_Agent/

### Core Modules

- **cognee_memory**: Cognee smart memory management
- **worldview_builder**: Markdown worldview builder
- **chat_agent**: Conversation agent core
- **database_manager**: Unified database management
- **emotion_analyzer**: Emotional relationship analysis
Expand All @@ -228,6 +267,16 @@ Neo_Agent/
- **long_term_memory**: Long-term memory system
- **schedule_manager**: Schedule management

### Acknowledgments

Thanks to the following open-source projects for supporting Neo Agent:

- **[Cognee](https://github.com/topoteretes/cognee)** - Open-source knowledge engine for persistent dynamic AI memory
- **[LangChain](https://github.com/langchain-ai/langchain)** - Powerful framework for building LLM applications
- **[LangGraph](https://github.com/langchain-ai/langgraph)** - State graph management and multi-agent collaboration
- **[DeepAgents](https://github.com/deepagents/deepagents)** - Sub-agent generation and persistent state management
- **[SillyTavern](https://github.com/SillyTavern/SillyTavern)** - Design reference for prompt engineering system

### License

This project is licensed under the MIT License. See [LICENSE](LICENSE) file for details.
Loading
Loading