Agentuniverse

Added
- New embedding components
- Azure-OpenAI embedding component
- Gemini embedding component
- Doubao embedding component
- New knowledge loading reader components
- Web page reader component based on BeautifulSoup (bs4)
- Image reader component based on OCR technology
- CSV format reader component
- New tool plugins
- Arxiv paper retrieval and query tool plugin
- Jina AI intelligent search tool plugin
- Added standard LLM output parser for reasoning-type models
- Added standard logger sink for agents, LLMs, and tools
- Added retry function annotation

Changed
- Improved project configuration experience, with the following key optimizations:
- Project configuration now supports global PACKAGE path replacement. For details, see the PACKAGE_PATH_INFO parameter in the sample project's config.toml file.
- Supports user-defined configuration extensions and YAML function extensions. For details, see the EXTENSION_MODULES parameter in the sample project's config.toml file.
- Updated custom_key.toml.sample with more detailed parameter content.
- Submit a set of commonly used model configuration YAML files. For details, see the llm directory in the sample project.
- YAML configuration now supports user-defined loading of attributes using environment variables and extension functions. For details, see the api_key comments in the llm sample of the sample project.
- Simplified LLM configuration. By default, it reads the global default_llm configuration. For details, see the default_llm.toml configuration in the llm sample of the sample project.
- Simplified component extension configuration. The module part in the component metadata can be left blank, and it will default to searching for a Python extension file with the same name in the same directory.
- Optimized project startup process, with the following key improvements:
- Improved LLM component startup registration. Only LLM instances that are dependent on by components like agents and tools are started.
- Enhanced error info during project startup.
- Other changes:
- Renamed LocalMemoryStorage to RamMemoryStorage for memory storage objects.

Note
- Added image_build image packaging files and tutorial documents for application projects.
- Added CONTRIBUTING.md documentation for the project.
- Standardized PR and ISSUE templates for the project.
- Added a sample practice for agents based on multimodal capabilities. For details, see the examples/multimodal_app directory in the project.
- Our project was included in the awesome-deepseek-integration list released by DeepSeek in February.

Added
- aU Memory Component Version Update
The new version provides comprehensive memory management and utilization capabilities for multi-agent applications through a global memory module. It can automatically collect, record, and process memory interactions among multiple agents, including conversations, model calls, tool calls, and knowledge retrieval. Additionally, it offers functionalities such as memory addition, memory retrieval, memory modification, memory compression, memory pruning, memory extraction, and multi-source persistence for both single-agent and multi-agent.

- Add AgentTemplate Component and WorkPattern Component
The AgentTemplate component helps users rapidly create agents according to pre-defined templates, while the WorkPattern component assists agent teams in selecting specific collaboration methods. In versions prior to V0.0.13, the Planner component offered both the customization logic of intelligent agents and their work patterns. The customization logic for agents was often modified due to the task, whereas work patterns were validated and stability. Consequently, users frequently needed to rewrite the Planner and focus on the code logic related to work patterns, which significantly increased their usage and development costs. With the new version, we have layered the customization logic and work patterns in the agent-building process. It will make the research on collaboration modes and the construction of intelligent agents more focused and efficient. The original Planner component will still be retained, but it will no longer be recommended.

- Knowledge Components Update
Add: Markdown format knowledge loading component, document chapter and paragraph segmentation component, knowledge graph storage component (beta version).

Changed
- Standard Project Scaffolding Update
The scaffolding Project Scaffolding has been standardized. You can read the [「Application Project Structure and Explanation」](./docs/guidebook/en/Get_Start/1.Application_Project_Structure_and_Explanation.md) for more details. For users of older versions, we have provided some guidelines and tools for this update, detailed information can be found in [「aU Old and New Project Migration Guide」](./docs/guidebook/en/Get_Start/6.aU_Old_and_New_Project_Migration_Guide.md).

Note
- Update docs for new memory version
- Optimization official docs and examples project
- Ask for high-priority features

Added
- RAG(Retrieval-Augmented Generation) Component Version Update
This version provides a standard operating procedure for knowledge base construction and the RAG retrieval recall stage. The component covers a series of RAG atomic capabilities, including data loading, data processing, index construction, knowledge storage, intent rewriting, and retrieval re-ranking, helping users to quickly build a general RAG intelligent agent solution in open-source scenarios.

- Intelligent Agent Product Platform Update
This version introduces new capabilities such as intelligent agent canvas orchestration, private knowledge base construction, and custom plugin support, enabling users to quickly build and orchestrate intelligent agents through a low-code, visual approach.

- Added GLM Default Model Component for Zhipu
- Added SQLiteStore Storage Component
- Added Flow Orchestration Execution Engine

Note
- Default path optimization for system_db_uri
The default path is already compatible with the Windows platform, for more details, please refer to [issue142](https://github.com/antgroup/agentUniverse/issues/142)
- Support for configurable chain stop words in ReactAgent
The ReactAgent YAML configuration now supports the stop_sequence keyword, allowing users to customize chain stop words. For more details, please refer to [issue127](https://github.com/antgroup/agentUniverse/issues/127)
- Added an introduction to RAG principles and a quick guide for building RAG, please pay attention to the corresponding parts in the README and user guide.
- Added advanced guidance documents for the intelligent agent productization platform, please pay attention to the corresponding parts in the README and user guide.
- Various code optimizations and documentation updates.

Added
- agentUniverse Product Version Offering
- The current version provides basic capabilities for agent construction, modification, and debugging, jointly launched by the difizen project. For more details, please refer to the documentation in the product platform section.
- Monitor Component: Added knowledge and tool instance collection, supporting full-link trace sequence concatenation and providing token consumption monitoring.
- New Web Session Module: Provides session and message persistence management capabilities.

Note
- Optimized Knowledge Component: Users can configure and specify any number of recall results (similarity_top_k).
- Fixed Chroma Component: Resolved issues where the embedding module was not specified.
- Various code optimizations and documentation updates.

Added
- DataAgent Autonomous Data Agent MVP Version Released
- Minimum Viable Product version, DataAgent aims to empower your agent with the capability of self-assessment and evolution through intelligent agent abilities. For detailed information, please refer to the user documentation.
- Added intermediate information streaming output capabilities in PEER and ReAct modes

Note
- Latest PEER research findings released
- This paper provides a detailed introduction to the mechanisms and principles of the PEER multi-agent framework. Experimental validation proves the advancement of the PEER model. For detailed information, please refer to the user documentation.
- Added use cases
- Andrew Ng's Reflexive Workflow Translation Agent Replication
- Some code optimizations and documentation updates.

Added
- Added standard integration for the DeepSeek model in the LLM module.
- Added a new OpenAI general protocol wrapper class, OpenAIStyleLLM.
- Models using the OpenAI protocol can be configured directly.
- Added a new LangChain tool wrapper class, LangChainTool, with several example tools for search and execution.
- LangChain tools can be configured directly.
- Added Agent information collection capability in the monitor module.

Note
- Added use cases.
- Supplemented documentation with a financial event analysis case study using PEER collaborative mode.
- Added several new documents for LLM components, tool components, and the Monitor module.
- Updated the new README.
- Some code optimizations and documentation updates.