mirror of
https://github.com/hanzoai/papers.git
synced 2026-10-29 07:34:31 +00:00
21 lines
1.9 KiB
TeX
21 lines
1.9 KiB
TeX
\section{Related Work}
|
|
|
|
\label{sec:related}
|
|
|
|
\subsection{Agent Frameworks}
|
|
|
|
LangChain~\cite{langchain2023} provides a code-first framework for LLM application development with chains, agents, and memory. LangGraph~\cite{langgraph2024} extends LangChain with graph-based workflow definition. CrewAI~\cite{crewai2024} specializes in multi-agent role-based collaboration. AutoGen~\cite{wu2023autogen} enables multi-agent conversation-driven workflows. DSPy~\cite{khattab2023dspy} compiles declarative language model programs. Hanzo Flow complements these frameworks by providing a visual layer with static verification and debugging.
|
|
|
|
\subsection{Visual Programming for AI}
|
|
|
|
ComfyUI~\cite{comfyui2023} provides node-based visual programming for Stable Diffusion workflows. Flowise~\cite{flowise2023} offers a drag-and-drop UI for LangChain flows. n8n~\cite{n8n2024} is a general-purpose workflow automation tool with AI nodes. Rivet~\cite{rivet2023} provides a visual IDE for AI agent prototyping. Hanzo Flow differs in its typed dataflow formalism, static verification, and adaptive execution engine.
|
|
|
|
\subsection{Dataflow Programming}
|
|
|
|
The dataflow programming paradigm has roots in Lucid~\cite{wadge1985lucid}, Lustre~\cite{halbwachs1991synchronous}, and LabVIEW~\cite{labview2003}. Modern implementations include Apache Beam~\cite{beam2016} for data processing and TensorFlow's computational graphs~\cite{abadi2016tensorflow}. Hanzo Flow applies dataflow principles to AI agent orchestration with domain-specific typing and execution semantics.
|
|
|
|
\subsection{Workflow Verification}
|
|
|
|
Formal verification of workflows has been studied in business process management~\cite{van2003workflow} and scientific workflows~\cite{deelman2005pegasus}. Type systems for dataflow have been explored in stream processing~\cite{hirzel2014catalog}. Hanzo Flow introduces a lightweight type system tailored to LLM input/output types with practical static verification.
|
|
|