DAGent

DAGent is a Python-based AI agent framework focused on creating AI agents organized as directed acyclic graph (DAG) workflows, where each node represents a function or decision controlled by large language models (LLMs). It is designed to simplify the construction of AI agents that can perform multi-step, complex tasks by dynamically deciding which functions (tools) to execute next based on inference from LLMs. DAGent uses a modular structure with nodes classified as function nodes that run Python functions and decision nodes where the LLM chooses the next step, enabling flexible and interpretable AI behavior automation.

Key Features:

• Organizes AI agent behavior as workflows structured in directed acyclic graphs.
• Decision nodes utilize LLMs to infer and select which function node (tool) to run next.
• Supports automatic tool description generation to improve agent reliability and interpretability.
• Compatible with multiple LLM models and supports passing outputs between functions for chained execution.

Use Cases:

• Building AI agents that automate complex workflows involving multi-step decision-making.
• Implementing AI-driven task orchestration where the agent adaptively chooses next actions.
• Creating customized AI solutions that integrate Python code functions as callable tools within an agent.

Technical Specifications:

• Written in Python and installable via pip or rye package managers.
• Integrates with various LLMs including local or API-based large language models.
• Utilizes function signatures and docstrings for LLM tool inference and workflow compilation.
• Supports persistent state passing using parameters like prev_output and flexible argument overrides.