Hedera Agent Kit has released version 4 of its SDK, introducing a major architectural overhaul focused on modularity, security, and fine-grained control over AI agent behavior. The update is designed to support developers building decentralized applications on Hedera Hashgraph by breaking the previous monolithic structure into smaller, composable packages while adding new policy and lifecycle control systems for agent execution.
The new version replaces the earlier single-package design with modular components that allow developers to install only the specific tools they need. This shift reduces unnecessary dependencies and improves flexibility for teams building specialized applications such as payments, identity systems, or AI-driven automation workflows. The core SDK now focuses on shared types and plugin architecture, while framework-specific integrations such as LangChain, Vercel AI SDK, ElizaOS, and MCP are separated into dedicated packages.
A key architectural change is the introduction of a new tool structure based on a BaseTool abstraction. This replaces the previous plain object model and enables structured lifecycle stages for each tool execution. Instead of a single execution function, tools are now broken into clearly defined phases, allowing the system to intercept, validate, and modify behavior at different points in the execution process.
Policies, Hooks, and Modular Plugins Strengthen AI Agent Governance
One of the most significant upgrades in version 4 is the introduction of a hooks and policies system that gives developers granular control over AI agent behavior. This system allows developers to define rules that govern how agents interact with blockchain infrastructure, including when they can execute transactions, what parameters they are allowed to use, and how outputs are validated before final execution.
The lifecycle includes four key control points: pre-tool execution, post-parameter normalization, post-core action, and post-tool execution. Each stage allows developers to enforce constraints, modify inputs, inspect transaction data, or log activity after execution. This design is intended to reduce risks such as unauthorized transactions, incorrect parameter usage, or unintended on-chain actions.
Related: Hedera Launches AI Studio Agent Bounty Campaign for Onchain AI Commerce
Alongside this, the plugin system has been made fully explicit, meaning developers must intentionally select and import only the tools required for their application. Built-in plugins such as account, token, consensus, and EVM tools are now organized under a structured namespace, ensuring that agents operate only within clearly defined capabilities. Third-party integrations, including Chainlink, Pyth, and identity verification tools, further extend functionality for specialized use cases such as pricing feeds and compliance workflows.
The update also includes a shift in SDK governance, with the underlying JavaScript SDK now maintained under the Hiero namespace instead of its previous structure. Developers are required to explicitly install the correct SDK version as a peer dependency, improving version control and reducing hidden dependency conflicts across applications.
Overall, Hedera Agent Kit V4 represents a shift toward production-grade AI agent infrastructure, combining modular design, deterministic execution controls, and enterprise-ready plugin systems. The update is aimed at enabling developers to build more secure, auditable, and customizable AI-driven applications while maintaining strict control over blockchain interactions in real-world environments.
