Flomny: A Multi-Agent System for Natural Language-Driven Workflow Automation

While multiple programming agents are beingput on market but they are currently meant forstandalone tasks, not for engineering softwarethat has numerous moving parts being integrated.Our methodology emphasizes modularity, robust-ness, and traceability. The system begins withprompt validation and task decomposition usingLLMs, followed by graph-based orchestrationof agents. This work presents Flomny, a multi-agentic system that automates software develop-ment workflows and their integration. Integra-tion Agents leverage Pinecone for vector-basedretrieval and produce code grounded in documen-tation rather than relying solely on the LLM’sinternal knowledge. Additionally, Flomny inte-grates the Model Context Protocol (MCP) to en-able advanced agent-based execution capabilities,allowing for dynamic discovery and utilization ofexternal tools while maintaining security bound-aries and user control. Static code validation pre-cedes optional execution in isolated subprocesses,enhancing both safety and reliability. Through adetailed case study involving Custom Servers andDiscord integrations, we demonstrate Flomny’sability to automatically generate and execute afour-step workflow derived entirely from a natu-ral language prompt. The results show successfultask decomposition, accurate API calls, and seam-less execution with real-time updates. By combin-ing the interpretability of LLMs, the structure ofgraph-based multi-agent systems, the precision ofretrieval- augmented code generation, and the flex-ibility of MCP integration, Flomny paves the wayfor more accessible, scalable, and human-centricworkflow automation.