Show HN: Agd – a content-addressed DAG for tracking what AI agents do (github.com)

A new tool named Agd has been released, designed to function as a content-addressed Directed Acyclic Graph (DAG) store for tracking the interactions of AI agents, particularly those utilizing Large Language Models (LLMs). This innovative tool records every interaction, including messages sent and received, tool calls, and the underlying causal relationships between these events. When developers open a pull request (PR), reviewers can now access a comprehensive trace of the AI agent's decisions and actions, rather than just a simple code diff. Agd enhances transparency and accountability in AI agent workflows, making it easier for developers to understand and audit the decision-making processes of their models. The tool operates seamlessly through a simple plugin integration with OpenCode, automatically initializing a collection of agent traces without the need for code modifications in existing projects. Key features of Agd include the ability to store different object types (blobs, messages, and actions) in a structured manner, enabling users to log and visualize AI interactions effectively. This functionality is pivotal for teams implementing AI-driven solutions, as it provides insights into agent behavior that can guide further development and improve model performance.