Whitepaper of Supernet
Supernet is designed to break down AI barriers across economies, regions, and vendors through decentralization, creating a truly borderless AI ecosystem.
Last updated
Supernet is designed to break down AI barriers across economies, regions, and vendors through decentralization, creating a truly borderless AI ecosystem.
Last updated
Supernet is designed to break down AI barriers across economies, regions, and vendors through decentralization, creating a truly borderless AI ecosystem. By establishing a unified AI marketplace, we empower both developers and users to benefit from the convenience of an aggregated agent network.
With Agent Store, Supernet sets the standard for AI agent evaluation and ranking—becoming the "Michelin Guide" of the agent ecosystem. It serves as the gateway to the AI era, providing a trusted platform where users can seamlessly discover and interact with the most effective agents.
At the core of this vision is SuperAgent—an AI that utilizes AI, enabling users to solve problems through a single interface. Just as search engines revolutionized the internet era, SuperAgent will be the intelligence hub of the AI age, paving the way for AGI adoption.
By integrating advanced recommendation algorithms, transparent evaluation systems, and decentralized computational services, Supernet redefines how AI agents are aggregated, evaluated, and deployed. Through blockchain and machine learning, we are shaping a future where AI services are not only powerful and efficient but also transparent and community-driven—setting new benchmarks for AI performance and accessibility.
The current market is saturated with a plethora of agents, yet a critical challenge persists: identifying which agents are optimally suited to fulfill user-specific tasks. To address this essential, Supernet has undertaken the development of a user rating-based agent aggregator.
This innovative solution is designed to deliver superior performance by curating the highest-quality agents, while simultaneously incorporating real-time user feedback to iteratively fine-tune agent capabilities. Through this dual mechanism of rigorous evaluation and adaptive optimization, the platform ensures alignment with evolving user requirements and maximizes task execution efficacy.
The current model marketplaces primarily focus on offering as many models and datasets as possible, but they lack robust evaluation systems for the actual content produced. For example, MyShell's model library has amassed a vast collection of Agents, yet users are left wondering: Which one is truly optimal for image generation? Which one excels in video generation?
Therefore, in the following sections, we will demonstrate how Supernet achieves agent aggregation and optimization through three core innovations:
Multi-Path Recommendation Algorithm
Decentralized Scoring Mechanism
Light-Node Agent Services
The architecture integrates the three aforementioned core modules, designed to enable path validation and optimization through an intentionally minimalist approach. Each component operates with maximum decentralization to ensure scalable and resilient agent coordination.
In the rapidly expanding landscape of AI agents, the challenge isn't the availability of models but identifying the most effective one for a specific task. Supernet addresses this challenge with its Multi-Path Recommendation Algorithm, a cornerstone of its user rating-based agent aggregator.
Task-Specific Pathways: The algorithm maps user tasks to multiple potential agent pathways. Each pathway represents a different combination of agents and models that can fulfill the task, considering factors like task complexity, required output quality, and execution speed.
Dynamic User Feedback Integration: Real-time user ratings and feedback are continuously integrated into the algorithm. This dynamic feedback loop ensures that the recommendation engine remains responsive to changing user preferences and agent performance metrics.
Performance Metrics and Historical Data: The algorithm leverages historical performance data of agents across various tasks. Metrics such as success rates, execution times, and user satisfaction scores are analyzed to inform future recommendations.
Adaptive Optimization: The system employs machine learning techniques to adaptively refine its recommendation pathways. As more data is collected, the algorithm becomes increasingly adept at predicting the most effective agent combinations for diverse tasks.
Input Phase: Users submit their tasks, specifying parameters like desired output type (e.g., image, video), quality, and time constraints.
Pathway Generation: The algorithm generates multiple pathways, each representing a unique combination of agents tailored to the task.
Evaluation and Scoring: Each pathway is evaluated based on historical data and user feedback. The highest-scoring pathways are prioritized.
Recommendation Delivery: Users receive a ranked list of agent combinations, with detailed insights into why each recommendation is optimal.
Continuous Learning: Unlike static agent recommendations, Supernet’s Optimal Path Agents continuously evolve through iterative self-improvement. Post-task user feedback is seamlessly integrated into the system, enabling agents to refine their decision-making and enhance future task execution.
This adaptive learning mechanism ensures that agents not only provide optimal recommendations at any given moment but also grow over time, consistently improving their performance to align with evolving user needs.
Precision: Tailors agent recommendations to specific user needs, ensuring higher task success rates.
Adaptability: Continuously evolves with user feedback and performance data, maintaining alignment with user expectations.
Transparency: Provides clear reasoning behind each recommendation, fostering user trust.
To further enhance transparency and trust in agent performance evaluations, Supernet integrates a Decentralized Scoring Mechanism powered by blockchain technology. This approach ensures that all user feedback and agent performance data are securely recorded, immutable, and publicly verifiable.
Blockchain-Based Feedback: Every piece of user feedback and agent performance metric is recorded on a block by node cluster. This ensures that the data is tamper-proof and accessible to all stakeholders, fostering trust in the scoring process.
Smart Contract-Driven Evaluation: Smart contracts automate the scoring process by aggregating feedback and performance data. These contracts execute predefined rules to calculate scores, ensuring consistency and eliminating human bias.
Token Incentives for Feedback: Users are incentivized to provide honest and constructive feedback through a token-based reward system. Tokens are distributed via smart contracts to users whose feedback aligns with broader consensus, promoting integrity in the scoring system.
Decentralized Governance: The scoring mechanism is governed by a decentralized community of stakeholders. Voting rights are distributed based on token holdings, allowing the community to propose and vote on changes to the scoring algorithms and rules.
Feedback Submission: After completing a task, users submit feedback on agent performance. This feedback is cryptographically signed and recorded on the blockchain.
Automated Scoring: Smart contracts aggregate feedback and performance data, calculate scores, and update the blockchain ledger in real-time.
Transparency and Verification: All scoring data is publicly accessible on the blockchain, allowing users to verify the authenticity and accuracy of agent ratings.
Community Governance: Stakeholders participate in governance decisions, ensuring the scoring mechanism evolves in line with community values and technological advancements.
Security: Cryptographic signatures and decentralized storage protect data from tampering and unauthorized access.
Fairness: Smart contracts eliminate human bias, ensuring consistent and objective evaluations.
Community-Driven: Decentralized governance empowers the community to shape the evolution of the scoring system.
To ensure efficient and scalable task execution, Supernet introduces Light-Node Agent Services. These lightweight nodes are designed to connect seamlessly with various inference environments and large language models, providing decentralized and reliable computational power for executing user tasks.
Decentralized Node Network: Light nodes are distributed across a global network, ensuring redundancy and high availability. Each node is capable of executing specific tasks, reducing dependency on centralized servers.
Inference Environment Integration: Light nodes are equipped to interface with diverse inference environments, enabling them to handle tasks that require complex computations or model-specific operations.
Direct Model Connectivity: Nodes are connected to large language models (LLMs) and other AI models, allowing for real-time processing and execution of tasks such as natural language generation, image recognition, and data analysis.
Resource-Efficient Design: Light nodes are optimized for low resource consumption, making them suitable for deployment on a variety of devices, from personal computers to edge devices, without compromising performance.
Task Allocation: When a user submits a task, the system identifies the most suitable light node based on factors like proximity, available resources, and required inference environment.
Node Execution: The selected node connects to the relevant inference environment and AI models to execute the task efficiently.
Result Delivery: Upon task completion, results are securely transmitted back to the user, with all interactions logged on the blockchain for transparency.
Scalability and Redundancy: The decentralized nature of the node network ensures tasks are distributed effectively, preventing bottlenecks and enhancing system resilience.
In the rapidly evolving landscape of AI technology, Supernet envisions a future where users can effortlessly identify, evaluate, and leverage the best AI agents tailored to their specific needs. Our mission is to address the critical challenge of selecting optimal agents from a saturated market, ensuring that users not only find the right tools but also experience superior performance and trust in their AI solutions.
At the heart of Supernet’s approach is the Multi-Path Recommendation Algorithm. This innovative system maps user tasks to multiple potential agent pathways, considering task complexity, output quality, and execution speed. By integrating real-time user feedback and historical performance data, the algorithm continuously refines its recommendations. Machine learning techniques further enhance this process, ensuring that users receive precise, adaptable, and transparent agent suggestions tailored to their unique requirements.
To build trust and transparency in agent evaluations, Supernet employs a Decentralized Scoring Mechanism powered by blockchain technology. Every user feedback and performance metric is securely recorded on an immutable blockchain ledger, ensuring data integrity and public verifiability. Smart contracts automate the evaluation process, eliminating human bias and maintaining consistent scoring standards. Users are incentivized to provide honest feedback through token rewards, while decentralized governance allows the community to shape and evolve the scoring system. This mechanism not only enhances fairness and security but also aligns with the principles of community-driven innovation.
Supernet’s vision extends to the actual execution of tasks through our Light-Node Agent Services. These decentralized, resource-efficient nodes are globally distributed, providing scalable and reliable computational power. Light nodes connect seamlessly with diverse inference environments and large language models, ensuring efficient task processing. Their optimized design allows deployment across various devices, from personal computers to edge devices, without compromising performance. By distributing tasks effectively and maintaining secure, blockchain-logged interactions, Supernet guarantees high availability, fault tolerance, and data integrity.
Supernet offers a highly flexible customization mechanism, enabling users to select the most suitable AI agents based on their specific needs. Users can define task requirements, including computational cost, model efficiency, and task complexity, allowing the system to generate optimal agent combinations accordingly. Additionally, Supernet supports pre-trained agents, which have been optimized for specific tasks or domains, reducing adaptation costs and enhancing execution efficiency. For example, in text processing tasks, users can choose an agent trained on news articles for automatic summarization or an agent specialized in legal documents for contract analysis. This approach not only provides standardized recommendation solutions but also allows for deep customization, ensuring that AI agents meet diverse application scenarios effectively.
Supernet employs a dual-layer scoring system consisting of agent scoring and pathway scoring to ensure optimal agent selection and execution efficiency.
Agent scoring evaluates the performance of individual AI agents based on multiple factors, such as task success rate, user feedback, computational cost, and execution stability. For instance, Llama, GPT-4, and Claude may excel in text-based tasks but might not perform as effectively in specialized domains such as code generation or medical diagnosis. Therefore, Supernet allows users to assess and select the most suitable AI agents based on specific task requirements.
Pathway scoring assesses the overall effectiveness of multiple agents working together in a recommendation pathway. This ensures that not only the best individual agents are selected, but also that the execution pipeline is optimized for efficiency. For example, in AI-generated content (AIGC), a pathway may include Llama for text generation, Stable Diffusion for image synthesis, and RunwayML for video processing. Supernet dynamically analyzes historical performance data of these agent combinations and adjusts the pathway scores accordingly, ensuring users receive the most effective task execution solutions.
Supernet integrates professional benchmark datasets with user ratings to create a robust and credible AI agent evaluation framework. Professional datasets provide standardized benchmarking for AI models, ensuring scientific validation of their capabilities. For instance, in computer vision, datasets such as ImageNet, COCO, and OpenImages are used for model evaluation; in natural language processing, GLUE, SuperGLUE, and MT-Bench serve as industry benchmarks; in code generation, HumanEval and APPS are utilized to assess programming ability.
User ratings, on the other hand, reflect real-world performance, capturing aspects that benchmark datasets may overlook, such as whether an AI agent retains key information in long-text summarization or whether its data analysis results are prone to errors. By combining expert-curated datasets with dynamic user feedback, Supernet ensures that agent recommendations are based on both industry-standard evaluations and real-world effectiveness, thereby optimizing the reliability of AI agents across various applications.
Supernet implements a structured domain categorization system, similar to how Google Play or the App Store categorizes applications and ranks them within different domains. This approach ensures that AI agents are effectively classified and recommended based on their suitability for specific tasks. Agents on the Supernet platform are categorized into domains such as natural language processing (NLP), computer vision (CV), financial analysis, code generation, medical diagnosis, legal compliance, and game development. Within each category, AI agents are ranked based on their performance, user ratings, and historical data, providing users with a transparent and reliable selection process.
Furthermore, Supernet supports cross-domain AI agent collaboration, allowing users to combine agents from multiple fields to create comprehensive workflows. For example, in automated content creation, users can select Llama for article writing, Stable Diffusion for image generation, and Whisper for speech synthesis, forming an integrated AI-powered content pipeline. By offering both domain-specific optimization and multi-domain interoperability, Supernet enhances the usability and efficiency of AI agents for a wide range of applications.
