LangGraph is a library for building stateful, multi-actor applications powered by language models. It represents workflows as directed graphs whose nodes run agent functions and whose edges control execution. section 15.1
Instead of passing data directly between components, each agent reads and updates a shared state object. A retrieval agent can add documents, after which an answer-generation agent uses the question and retrieved material. section 15.1
A knowledge-graph question-answering pipeline can organize its work into specialized agents:
- Detect visualization intent and store the output type and reasoning. section 15.1.5
Conditional post-query routing
A post-query edge retries failed execution while retries remain, ends after three attempts, summarizes graph or map results, and ends directly for other output types.
A Streamlit integration exposes the pipeline interactively. Temporary placeholders show agent progress, while MessageHistory preserves questions, answers, intermediate state, reasoning, tables, and maps for display. section 15.2.2
