# goalflow

A LangGraph-based production framework that combines visual workflow graphs with code-first agent loops, and transpiles Dify's exported workflow format into runnable, version-controllable Python.

goalflow is a framework for building LLM applications on top of LangGraph in two complementary ways: visual-first workflows designed in Dify's drag-and-drop editor and transpiled by one command into a runnable LangGraph Python file with no continued dependency on Dify's runtime, and code-first agents built with a bundled SDK supporting ReAct, deep, and custom agent loops with middleware, model routing, failover, skills and observability. Because a workflow graph node can host an agent loop and an agent can call sub-workflows as tools, the two approaches can be combined.

It includes more than 20 built-in workflow nodes covering LLM calls, code execution, HTTP requests, conditionals, classification and iteration, a pluggable data adapter that defaults to the Dify wire protocol but also supports an OpenAI-compatible one, markdown-based skills matched to queries and injected into prompts, and a document ingestion pipeline for retrieval backed by Milvus. In load testing, a two-replica deployment on 2 vCPU and 4 GB RAM per replica sustained 100 concurrent conversations without measurable regression in time to first token.

It is written in Python.

Website: https://github.com/wanmol/goal-flow