What is Co-Sight
Co-Sight is ZTE's open source super agent project, serving as a collaborative visual analysis platform and intelligent automation foundation. It adopts a multi-agent architecture to build a "digital team" collaboration system, driven by a DAG task engine to achieve efficient task scheduling and execution. Co-Sight possesses self-evolution capabilities, automatically generating intelligent summary reports through execution records and model reasoning, forming a continuous improvement loop. It emphasizes security and reliability, with all operations running in a sandbox environment, supporting log tracing, permission control, and compliance auditing.
Main Features of Co-Sight
- Intelligent Summarization and Reflection: Based on execution process records and model reasoning, it can automatically conduct retrospectives and generate intelligent summary reports, including reviews of key success points, identification of failure causes, and feedback on optimization suggestions, forming a continuous self-evolution loop.
- Multiple Security Protections: All operations are conducted in a controlled sandbox environment to prevent data breaches or leaks; system logs are recorded in real time, with the entire operation path traceable; it supports permission control and compliance auditing, ensuring worry-free enterprise operations.
- Real-time Monitoring and Intelligent Recognition: It can analyze surveillance video in real time and quickly identify abnormal behavior, accurately recognizing multiple behaviors and objects through deep learning technology.
- Flexible Deployment: It supports multiple operating systems and hardware platforms, and is easy to deploy and scale.
Technical Principles of Co-Sight
- Multi-Agent Architecture: Co-Sight adopts a Multi-Agent architecture and builds a "digital team" collaboration system. The supervisor agent coordinates and dispatches multiple execution agents, each performing its own role, to collaboratively complete complex tasks. This gives the system excellent task decomposition and process collaboration capabilities, enabling dynamic replanning and adaptive fault tolerance, automatically responding to emergencies during task execution, flexibly adjusting strategies, and ensuring continuous and stable processes.
- DAG Task Engine: Co-Sight introduces a DAG (Directed Acyclic Graph) task scheduling mechanism, intelligently identifying concurrent nodes in tasks and combining them with the concurrent execution capabilities of multiple agents to significantly shorten the overall process cycle. The rich tool components provided cover multiple search engines, voice and image recognition tools, document processing tools, etc., enabling one-stop completion of various tasks from planning to execution.
- Deep Learning and Natural Language Processing: Co-Sight integrates deep learning models and Natural Language Processing (NLP) technology, performing reasoning and decision-making based on large-scale neural networks. The system can process massive image data, textual information, and diverse sensor data
Co-Sight Project Address
- Github Repository：https://github.com/ZTE-AICloud/Co-Sight
Application Scenarios of Co-Sight
- Industry Research: Automatically generate visual reports, such as SWOT analysis.
- News Analysis: Integrate multi-source information for public opinion insights.
- Travel Planning: Integrate multi-platform APIs to optimize itineraries.
- Academic Research: Automatic generation of literature retrieval and reviews.