COGENTS: Towards COGnitive aGENTic System
A comprehensive, modular ecosystem for building intelligent, cognitive agent systems. COGENTS provides foundational abstractions, specialized toolkits, intelligent agents, and powerful search & automation capabilities—all designed to work together seamlessly.
COGENTS is an initiative to develop a cognitive, computation-driven agentic system that combines theoretical foundations with practical implementations. The project emphasizes:
- Modular Architecture: Build systems with composable, reusable components
- Cognitive Design: Agent systems grounded in multi-agent system (MAS) philosophy
- Production-Ready: Battle-tested tools and agents for real-world applications
- Extensible Framework: Easy integration and customization for diverse use cases
COGENTS is organized into five interconnected subprojects, each addressing a specific aspect of cognitive agent development:
COGENTS
├── cogents-core # Foundational abstractions & core modules (BASE)
├── cogents-smith # Toolkits ecosystem & production agents
├── cogents-browser-use # AI-powered browser automation
├── wizsearch # Multi-engine search orchestration
└── wizagent # Web automation & research framework
1. cogents-core 🧠 (BASE)
The foundational layer providing core abstractions and essential modules for all COGENTS projects.
- Agent: Base classes for building custom agents (BaseAgent, BaseGraphicAgent, BaseConversationAgent, BaseResearcher)
- Goal Management: Goal decomposition, conflict detection, and dynamic replanning (Goalith)
- Tool Management: Centralized tool registry, execution engine, and repository system (Toolify)
- Memory Management: Persistent memory capabilities (under development)
- Orchestration: Global orchestration system (under development)
- Multi-Model LLM Support: OpenAI, Google GenAI, Ollama, LlamaCPP with dynamic routing
- Advanced Routing: Complexity-based and self-assessment routing strategies
- Observability: Built-in token tracking and Opik tracing integration
- Message Bus: Inter-agent communication system
- Extensible Design: Easy to add new providers and capabilities
pip install -U cogents-coreUse Cases: Building custom agents, goal-oriented planning, tool integration, LLM management
2. cogents-smith 🔨
Extensive toolkit ecosystem and production-ready agents built on cogents-core.
- Academic Research: arXiv integration for paper discovery and analysis
- Development Tools: Bash, file editing, GitHub, Python execution
- Media Processing: Image analysis, video processing, audio transcription
- Information Retrieval: Wikipedia, web search, knowledge extraction
- Data Management: Tabular data, memory systems, document handling
- Communication: Gmail integration
- Human Interaction: User communication and feedback collection
- Askura Agent: Dynamic conversational agent for structured information collection
- Seekra Agent: Deep research agent for comprehensive topic investigation and report generation
- Semantic Organization: Intuitive grouping for easy discovery
- Lazy Loading: Load only what you need
- Async-First Design: High-performance concurrent operations
- Error Resilience: Graceful handling of missing dependencies
pip install -U cogents-smithUse Cases: Rapid agent development, conversational data collection, deep research, multi-modal processing
3. cogents-browser-use 🌐
AI-powered browser automation adapted from browser-use on the COGENTS stack.
- Natural language browser control
- Intelligent web navigation
- Automated interaction with web elements
- Built on COGENTS core abstractions
- Support for headless and headed modes
pip install -U cogents-browser-useUse Cases: Web scraping, automated testing, data extraction, web interaction automation
4. wizsearch 🔍
Unified Python library for searching across multiple search engines with a consistent interface.
- Multiple Search Engines: Baidu, Bing, Brave, DuckDuckGo, Google, Google AI, SearxNG, Tavily, WeChat
- Unified Interface: Single API with consistent
SearchResultformat - Multi-Engine Aggregation: Concurrent searches with round-robin result merging
- Page Crawling: Built-in web page content extraction using Crawl4AI
- Semantic Search: Optional vector-based semantic search with local storage
- Full Async/Await Support: High-performance asynchronous operations
pip install -U wizsearchUse Cases: Multi-source search, web content aggregation, semantic search, research automation
5. wizagent 🧙
Powerful web automation and research framework combining multi-engine search, deep research, browser automation, and structured data extraction.
- Multi-Engine Web Search: Simultaneous search across multiple engines (Tavily, DuckDuckGo, Google AI, SearXNG, Brave, Baidu, WeChat)
- Deep Research Agent: Autonomous research with iterative query refinement and source aggregation
- Browser Automation: Intelligent browser control with natural language instructions (via cogents-browser-use)
- Structured Data Extraction: Extract data from websites using Pydantic models
- YAML-Based Schema Parser: Define data models declaratively with the Gem Parser
- LangGraph Workflows: Advanced agent orchestration with state management
Built on cogents-core, cogents-browser-use, wizsearch, LangGraph, and Pydantic for a complete web intelligence solution.
pip install -U wizagentUse Cases: Market research, competitive analysis, web data extraction, automated web intelligence gathering
- DeepWiki Documentation:
- Philosophy: Multi-Agent Systems Talk
We welcome contributions to any of the COGENTS subprojects! Each subproject follows its own contribution guidelines:
- Fork the specific subproject repository
- Create your feature branch
- Follow the project's code style and testing requirements
- Submit a pull request
For major changes, please open an issue first to discuss what you would like to change.
All COGENTS subprojects are released under the MIT License. See the LICENSE file in each subproject repository for details.
COGENTS builds upon and integrates with several excellent projects:
- Tencent Youtu-agent for toolkit integration
- browser-use for browser automation foundations
- LangGraph for workflow orchestration
- The open-source community for continuous inspiration and support
Built with ❤️ by the COGENTS Community