With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
Bittensor Subnet 36, known as Web Agents, is a groundbreaking initiative within the Bittensor ecosystem designed to revolutionize web-based automation by creating a network of autonomous web agents capable of interacting with any website. The subnet’s primary function is to incentivize the development and deployment of AI-driven agents that can perform tasks on the web—such as navigating websites, filling out forms, extracting data, or completing transactions—without human intervention. These agents are built to adapt to the dynamic nature of the internet, handling changes in website layouts or content with ease, which is a significant improvement over traditional automation tools that often break when websites update.
Web Agents leverages Bittensor’s decentralized network to distribute the workload across miners, who develop and run these agents, while validators assess their performance on real-world web tasks. The subnet aims to transform how industries leverage web-based software by automating low-value, routine tasks that often slow down operations, as highlighted on autoppia.com. For example, a business could use a Web Agent to automatically scrape pricing data from e-commerce sites or manage customer support inquiries on a web platform. By rewarding miners with TAO for creating effective agents, Web Agents fosters a competitive environment where the best-performing solutions rise to the top, ultimately benefiting users with reliable, adaptable automation tools that work across the web’s vast and ever-changing landscape.
Bittensor Subnet 36, known as Web Agents, is a groundbreaking initiative within the Bittensor ecosystem designed to revolutionize web-based automation by creating a network of autonomous web agents capable of interacting with any website. The subnet’s primary function is to incentivize the development and deployment of AI-driven agents that can perform tasks on the web—such as navigating websites, filling out forms, extracting data, or completing transactions—without human intervention. These agents are built to adapt to the dynamic nature of the internet, handling changes in website layouts or content with ease, which is a significant improvement over traditional automation tools that often break when websites update.
Web Agents leverages Bittensor’s decentralized network to distribute the workload across miners, who develop and run these agents, while validators assess their performance on real-world web tasks. The subnet aims to transform how industries leverage web-based software by automating low-value, routine tasks that often slow down operations, as highlighted on autoppia.com. For example, a business could use a Web Agent to automatically scrape pricing data from e-commerce sites or manage customer support inquiries on a web platform. By rewarding miners with TAO for creating effective agents, Web Agents fosters a competitive environment where the best-performing solutions rise to the top, ultimately benefiting users with reliable, adaptable automation tools that work across the web’s vast and ever-changing landscape.
The product of Subnet 36, Web Agents, is a decentralized network of autonomous AI agents that can interact with any website, offering a fully permissionless web operator system powered by Bittensor’s infrastructure. The “build” is a protocol where miners develop and deploy web agents—AI models trained to perform specific web tasks—and validators evaluate their performance using a benchmark called the Infinite Web Arena (IWA). The IWA, as described on autoppia.com, is a standardized set of web tasks that tests agents on their ability to navigate, interact, and complete objectives on real websites, such as booking a flight or extracting data from a blog. Miners in Web Agents create these agents using machine learning techniques, often leveraging large language models (LLMs) or multimodal AI to understand and interact with web content, including text, images, and HTML structures. Validators then challenge these agents with tasks from the IWA, scoring them based on accuracy, efficiency, and adaptability. Key features of the Web Agents build include:
The autoppia.com website emphasizes that Web Agents “keeps getting better over time” through this competitive process, as miners refine their agents to achieve higher scores. The result is a swarm of autonomous agents that can automate web operations for businesses, researchers, or individuals, making processes more efficient in a world where the web is central to nearly all functions, as noted in the subnet’s documentation.
Technical Architecture
The technical architecture of Web Agents (Subnet 36) is a sophisticated system that combines AI, web interaction, and Bittensor’s decentralized framework to create a scalable and efficient network of autonomous agents. The subnet operates with up to 256 nodes (neurons), with 64 slots reserved for validators and the rest for miners, following Bittensor’s standard structure. Miners develop web agents using machine learning models, often LLMs or multimodal AI, capable of understanding web content (e.g., HTML, text, images) and performing tasks like clicking buttons, filling forms, or extracting data. These agents are deployed on miners’ systems, where they interact with real websites through a browser interface, such as a headless browser (e.g., Selenium or Playwright), as inferred from the GitHub repository’s focus on web automation. Validators evaluate the agents using the Infinite Web Arena (IWA) benchmark, which provides a diverse set of web tasks to test the agents’ capabilities. The evaluation process involves:
The scoring results are aggregated using Bittensor’s Yuma Consensus mechanism, which determines the TAO rewards for each miner. The subnet connects to the Subtensor blockchain via the Bittensor SDK, enabling miners and validators to register, stake TAO, and log their activities. Web interactions occur off-chain for speed, but the incentive mechanism is recorded on-chain, ensuring transparency. The GitHub repository (autoppia/autoppia_web_agents_subnet) indicates that the system is designed to be permissionless, meaning anyone can join as a miner or validator, and it leverages Bittensor’s parallel processing capabilities to scale across a global network of participants, as noted in the autoppia.com papers.
Example Use Case
Consider an e-commerce scenario where the goal is to purchase a product under certain constraints.
Phase 1: Data & Test Generation
LLMs generate tasks and tests without human input. For example:
Task: “Buy a red dress for less than $10.”
The tests might verify that a “Purchase()” event occurred with parameters (item: “red dress”, price: 10) to confirm successful completion.
Phase 2: Agent Execution & Action Tracing
The agent is run in a virtual browser environment. It navigates, searches, clicks, and adds items to the cart, all recorded for evaluation. This ensures no real-world side effects while preserving fidelity.
Phase 3: Test Execution & Evaluation
Once the agent finishes its attempts, the pre-generated tests run automatically. If the tests pass, the agent succeeds. If not, it fails. Subtasks and milestones can provide partial credit, helping guide iterative improvement.
Key Insight
By generating both tasks and validation criteria upfront, they eliminate the need for a human or omniscient AI evaluator at runtime. Objective tests enable IWA to scale indefinitely, continually challenging agents as website configurations and complexities evolve.
The product of Subnet 36, Web Agents, is a decentralized network of autonomous AI agents that can interact with any website, offering a fully permissionless web operator system powered by Bittensor’s infrastructure. The “build” is a protocol where miners develop and deploy web agents—AI models trained to perform specific web tasks—and validators evaluate their performance using a benchmark called the Infinite Web Arena (IWA). The IWA, as described on autoppia.com, is a standardized set of web tasks that tests agents on their ability to navigate, interact, and complete objectives on real websites, such as booking a flight or extracting data from a blog. Miners in Web Agents create these agents using machine learning techniques, often leveraging large language models (LLMs) or multimodal AI to understand and interact with web content, including text, images, and HTML structures. Validators then challenge these agents with tasks from the IWA, scoring them based on accuracy, efficiency, and adaptability. Key features of the Web Agents build include:
The autoppia.com website emphasizes that Web Agents “keeps getting better over time” through this competitive process, as miners refine their agents to achieve higher scores. The result is a swarm of autonomous agents that can automate web operations for businesses, researchers, or individuals, making processes more efficient in a world where the web is central to nearly all functions, as noted in the subnet’s documentation.
Technical Architecture
The technical architecture of Web Agents (Subnet 36) is a sophisticated system that combines AI, web interaction, and Bittensor’s decentralized framework to create a scalable and efficient network of autonomous agents. The subnet operates with up to 256 nodes (neurons), with 64 slots reserved for validators and the rest for miners, following Bittensor’s standard structure. Miners develop web agents using machine learning models, often LLMs or multimodal AI, capable of understanding web content (e.g., HTML, text, images) and performing tasks like clicking buttons, filling forms, or extracting data. These agents are deployed on miners’ systems, where they interact with real websites through a browser interface, such as a headless browser (e.g., Selenium or Playwright), as inferred from the GitHub repository’s focus on web automation. Validators evaluate the agents using the Infinite Web Arena (IWA) benchmark, which provides a diverse set of web tasks to test the agents’ capabilities. The evaluation process involves:
The scoring results are aggregated using Bittensor’s Yuma Consensus mechanism, which determines the TAO rewards for each miner. The subnet connects to the Subtensor blockchain via the Bittensor SDK, enabling miners and validators to register, stake TAO, and log their activities. Web interactions occur off-chain for speed, but the incentive mechanism is recorded on-chain, ensuring transparency. The GitHub repository (autoppia/autoppia_web_agents_subnet) indicates that the system is designed to be permissionless, meaning anyone can join as a miner or validator, and it leverages Bittensor’s parallel processing capabilities to scale across a global network of participants, as noted in the autoppia.com papers.
Example Use Case
Consider an e-commerce scenario where the goal is to purchase a product under certain constraints.
Phase 1: Data & Test Generation
LLMs generate tasks and tests without human input. For example:
Task: “Buy a red dress for less than $10.”
The tests might verify that a “Purchase()” event occurred with parameters (item: “red dress”, price: 10) to confirm successful completion.
Phase 2: Agent Execution & Action Tracing
The agent is run in a virtual browser environment. It navigates, searches, clicks, and adds items to the cart, all recorded for evaluation. This ensures no real-world side effects while preserving fidelity.
Phase 3: Test Execution & Evaluation
Once the agent finishes its attempts, the pre-generated tests run automatically. If the tests pass, the agent succeeds. If not, it fails. Subtasks and milestones can provide partial credit, helping guide iterative improvement.
Key Insight
By generating both tasks and validation criteria upfront, they eliminate the need for a human or omniscient AI evaluator at runtime. Objective tests enable IWA to scale indefinitely, continually challenging agents as website configurations and complexities evolve.
Web Agents (Subnet 36) is developed by Autoppia, a company focused on autonomous web automation, as confirmed by the autoppia.com website and GitHub repository. While specific team members are not publicly named, Autoppia is the driving force behind the subnet, with a mission to revolutionize how industries use web-based software by automating routine tasks. The autoppia.com website highlights the team’s goal of creating a network of web agents that “adapts to changes and keeps getting better over time,” reflecting a deep understanding of both AI and web technologies.
The project’s GitHub repository is well-maintained, with detailed documentation and setup guides, suggesting a team with expertise in software engineering, machine learning, and blockchain integration. Community discussions on Bittensor’s Discord also indicate that Autoppia engages with users to provide support and updates, showing a commitment to community-driven development. The team’s work has been recognized within the ecosystem, with Web Agents being praised for its potential to transform web automation, as evidenced by its detailed documentation and active presence in Bittensor’s community channels.
Web Agents (Subnet 36) is developed by Autoppia, a company focused on autonomous web automation, as confirmed by the autoppia.com website and GitHub repository. While specific team members are not publicly named, Autoppia is the driving force behind the subnet, with a mission to revolutionize how industries use web-based software by automating routine tasks. The autoppia.com website highlights the team’s goal of creating a network of web agents that “adapts to changes and keeps getting better over time,” reflecting a deep understanding of both AI and web technologies.
The project’s GitHub repository is well-maintained, with detailed documentation and setup guides, suggesting a team with expertise in software engineering, machine learning, and blockchain integration. Community discussions on Bittensor’s Discord also indicate that Autoppia engages with users to provide support and updates, showing a commitment to community-driven development. The team’s work has been recognized within the ecosystem, with Web Agents being praised for its potential to transform web automation, as evidenced by its detailed documentation and active presence in Bittensor’s community channels.
Benchmark Pipeline:
The core benchmarking infrastructure is now operational and actively running across a limited number of manually deployed demo websites. This initial rollout validates the foundational capabilities of the subnet.
Automated Website Generation & Deployment:
They are working towards full meta-programming automation—enabling the system to autonomously generate and deploy websites. This unlocks an infinite testing space without manual setup, drastically increasing scale and variability.
Advanced Validation Frameworks:
Enhancements to the validation pipeline are underway. The goal is to enable more robust testing of agents across highly complex, dynamic, and unpredictable tasks, ensuring meaningful performance benchmarks.
Refinement of Synthetic Pipelines:
New filtering layers and constraint logic are being implemented to improve the quality of LLM-generated content—minimizing hallucinations and increasing alignment with realistic outputs.
Launch of Web Playground Application:
A public-facing interactive web interface is in development, giving users the ability to test and observe intelligent web agents in real time.
Agent Ranking Dashboard:
A transparent leaderboard system will soon be released, providing live scoring and ranking of agent performance. This is designed to foster competition and highlight top contributors.
Beyond the Browser:
Plans are underway to extend the IWA (Infinite Web Agent) framework beyond browser environments—targeting operating system-level interactions and other interfaces that support infinite, procedurally generated tasks.
Benchmark Pipeline:
The core benchmarking infrastructure is now operational and actively running across a limited number of manually deployed demo websites. This initial rollout validates the foundational capabilities of the subnet.
Automated Website Generation & Deployment:
They are working towards full meta-programming automation—enabling the system to autonomously generate and deploy websites. This unlocks an infinite testing space without manual setup, drastically increasing scale and variability.
Advanced Validation Frameworks:
Enhancements to the validation pipeline are underway. The goal is to enable more robust testing of agents across highly complex, dynamic, and unpredictable tasks, ensuring meaningful performance benchmarks.
Refinement of Synthetic Pipelines:
New filtering layers and constraint logic are being implemented to improve the quality of LLM-generated content—minimizing hallucinations and increasing alignment with realistic outputs.
Launch of Web Playground Application:
A public-facing interactive web interface is in development, giving users the ability to test and observe intelligent web agents in real time.
Agent Ranking Dashboard:
A transparent leaderboard system will soon be released, providing live scoring and ranking of agent performance. This is designed to foster competition and highlight top contributors.
Beyond the Browser:
Plans are underway to extend the IWA (Infinite Web Agent) framework beyond browser environments—targeting operating system-level interactions and other interfaces that support infinite, procedurally generated tasks.
Keep ahead of the Bittensor exponential development curve…
Subnet Alpha is an informational platform for Bittensor Subnets.
This site is not affiliated with the Opentensor Foundation or TaoStats.
The content provided on this website is for informational purposes only. We make no guarantees regarding the accuracy or currency of the information at any given time.
Subnet Alpha is created and maintained by The Realistic Trader. If you have any suggestions or encounter any issues, please contact us at [email protected].
Copyright 2024