spcl/graph-of-thoughts

Constructs problem-solving workflows as directed acyclic graphs of LLM operations (Generate, Score, Aggregate, etc.), enabling flexible composition of reasoning strategies from simple Chain-of-Thought to more sophisticated multi-path exploration. Supports pluggable LLM backends (ChatGPT, local models via controller abstraction) and includes custom prompters/parsers for domain-specific problems. Provides built-in examples for sorting and keyword counting that demonstrate how to define scoring functions and ground-truth validation within the operation graph.

2,614 stars. No commits in the last 6 months. Available on PyPI.