Agent Builder SDK

The Agent Builder SDK is the foundational layer of Hybrid’s development ecosystem, enabling developers to build AI agents with access to essential tools and resources. It simplifies the setup process by providing streamlined access to large language models (LLMs), compute resources, and the ability to train agents with integrated datasets and APIs.

The SDK focuses on covering the core needs for AI agent development while allowing developers to customize and expand functionality.

Core Components

1. Access to Leading LLMs

   • The SDK provides seamless integration with the most popular LLMs, enabling developers to select the best model for their specific use case. Supported LLMs include:

     • anthropic claude 3 haiku

     • anthropic claude 3 sonnet

     • gpt-4o

     • gpt-4o mini

     • gpt 4 turbo

     • gpt 3.5

     • cohere command r

     • cohere commanr r plus

2. Compute Resources

   • Access to decentralized compute networks optimized for AI tasks.

   • Compatible with GPU-powered clusters for training and inferencing, ensuring cost-effective and scalable performance.

3. Training Environments

   • Pre-configured environments for fine-tuning and optimizing agents.

   • Supports transfer learning to reduce training time and cost by leveraging pre-trained models.

4. Data Integration

   • APIs for incorporating datasets from on-chain and off-chain sources.

   • Secure storage for uploading custom data to train or improve agent performance.

5. Modular Expansion

   • Developers can extend basic functionalities by integrating their own LLMs, APIs, or data processing pipelines.

Developer Workflow

1. Model Selection

   • Choose the most suitable LLM for the use case (e.g., GPT-4 for complex tasks, PaLM for multi-modal inputs).

2. Data Preparation

   • Load and preprocess datasets via integrated APIs or manual uploads.

   • Use pre-built data connectors for blockchain data, market feeds, and social metrics.

3. Agent Training

   • Fine-tune agents in a dedicated training environment using labeled data or transfer learning.

4. Testing and Debugging

   • Simulate agent tasks and refine logic using the SDK’s debugging tools.

5. Deployment

   • Deploy trained agents to Hybrid’s Layer 2 or other supported networks with integrated compute and storage options.