Subnet 20: BitAgent

Welcome to Subnet 20 (BitAgent), the leading decentralized AI subnet on Bittensor, pioneering the future of function-calling and agentic automation.

SN20 is the first and only Bittensor subnet focused on tool-calling AI models, enabling seamless execution of complex tasks across applications. Our models are designed for real-world AI agents, making them practical for businesses, developers, and researchers. Competing on the Berkeley Function Calling Leaderboard, we’re proving that open-source, Apache 2-licensed AI models (≤8B parameters) can outperform industry giants in tool execution.

Our cutting-edge products, GoGoAgent and MSPTech.ai, showcase the power of AI-driven automation, revolutionizing IT support, AI agent workflows, and enterprise productivity. Built for scalability, SN20’s models run on consumer-grade GPUs, making advanced AI accessible to miners, businesses, and developers alike.

Join the AI revolution. Stake with SN20 and be part of the future of decentralized AI-powered automation.

Evolution of AI Agency from Traditional Bots

SN 20’s AI agents redefine intelligence with adaptive precision and autonomous evolution, leaving traditional bots stagnant in a world that demands progress.

Subnet 20 AI Agents:

Adaptive learning & fine-tuning

Handles complex decision trees

Multi-tool orchestration

Semi-autonomous evolution & self-improvement

Optimized for consumer GPUs (≤8B parameters)

Integrates with multiple Bittensor subnets

Competes on Berkeley’s Function Calling Leaderboard

Traditional Bots:

Fixed response patterns

Limited decision-making

Single-purpose tasks

Requires constant updates

Requires proprietary infrastructure

Operates in isolation

No competitive benchmarking

Core Technology Stack

Subnet 20’s core technology stack powers AI-driven tool function calling, agent orchestration, and decentralized execution—enabling scalable, autonomous AI agents that seamlessly integrate, evaluate, and improve in real-world applications.

Tool Calling

Tool Function Calling
  • Tool function calling for automation
  • API integration framework for connectivity
  • User prompt processing for AI input
  • Tool descriptions for function execution

Decentralized / Distributed

Agent Orchestration & Performance

  • Agent orchestration for multi-tool execution
  • Multi-agent coordination for workflows
  • Offline evaluation (80%) via submissions
  • Online evaluation (20%) via live queries
  • Performance monitoring via leaderboard & validation
  • Datasets: BitAgent, Glaive, Berkeley

Decentralized Infrastructure

Agent Orchestration & Performance

  • Distributed computing via validators
  • Blockchain integration with Bittensor
  • Secure validation protocols for accuracy

Evolution & Growth

From pioneering AI-driven tool/function calling to shaping the future of decentralized intelligence, our journey reflects relentless innovation, technical excellence, and a commitment to advancing the Bittensor ecosystem.

Current Achievements

  • Ranked on Berkeley Function Calling Leaderboard
  • Building MSPTech.ai, an AI-driven IT automation platform
  • Launched GoGoAgent, an AI agent service for tool function calling

Future Development:

  • Expanded platform integration for broader adoption
  • Enhanced AI capabilities for tool execution
  • Performance optimizations for speed and scalability

Key Questions

Answers to the most common questions about Subnet 20, from its core technology to participation, performance, and real-world applications.
Subnet 20 is a specialized Bittensor subnet focused on AI-driven tool function calling, enabling AI agents to autonomously execute tasks by interacting with various tools and APIs.
Unlike general AI subnets, Subnet 20 is dedicated to function-calling models, allowing AI agents to perform specific actions by calling external functions and tools.
Subnet 20 powers platforms like MSPTech.ai for AI-driven IT automation and GoGoAgent for AI tool function calling, improving efficiency in real-world applications.
Participation is open to miners developing and hosting function-calling models and validators evaluating miner outputs. Models must be open-source (Apache 2.0) and ≤8B parameters.
Subnet 20’s models are optimized for consumer-grade GPUs, making mining accessible without the need for high-end infrastructure.
Evaluation is based on offline testing with datasets like BitAgent’s and Berkeley’s, as well as real-time online assessments to measure tool-calling accuracy.
Participants earn TAO tokens based on performance, with miners rewarded for high-quality models and validators for accurate assessments.
Subnet 20 interacts with other subnets, such as integrating SN45 (Gen42) for AI-driven code execution, strengthening the overall ecosystem.
Technical documentation, including guidelines for participation, can be found in the BitAgent Subnet repository on GitHub.

Follow official channels, such as Twitter, for updates on development, performance, and new integrations.

Subnet 20 prioritizes highly efficient function-calling models that are both lightweight (≤8B parameters) and competitive on Berkeley’s Function Calling Leaderboard.
Validation is distributed across a network of Bittensor validators, ensuring unbiased assessment and continuous model improvement.
By developing open-source, high-performance function-calling models, Subnet 20 enables AI automation without reliance on centralized providers.
Industries such as IT support, automation, and AI-driven enterprise solutions can integrate Subnet 20’s models to enhance efficiency and reduce manual workload.

Continuous competition on the Berkeley leaderboard, real-world testing through platforms like MSPTech.ai, and direct feedback loops drive ongoing improvements in performance and scalability.

Ready to Transform Your AI Infrastructure?

In our weight class of <=8B models, we are in the top! We are looking to fight several weight classes higher

Start Building
Stake Now
X-twitter Youtube Github