zhiliscope/signalgraph

SignalGraph is a Python tool that reads plain text or Markdown documents and pulls out explicit relationships between things mentioned in the text, turning scattered sentences into a connected knowledge graph. Given input like Microsoft invested in OpenAI, OpenAI created GPT, GPT powers ChatGPT, it produces a set of linked facts that can be explored, queried, and exported instead of staying as separate isolated statements. Unlike tools that rely on a language model to guess relationships, SignalGraph uses fixed, rule-based logic to find a specific list of relationship types, including released, created, developed, invested_in, uses, works_with, part_of, related_to, and powers. Every extracted relationship keeps the original sentence it came from, along with which sentence number it appeared in and the first four-digit year found in that sentence, used as a rough timestamp. Because the rules are fixed and repeatable rather than model generated, the same input always produces the same output, and results can be traced back to exactly the sentence that produced them. Once entities and relationships are extracted, they are stored in an in-memory directed graph that supports looking up entities by name, ID, or type, finding the shortest path between two entities, listing incoming or outgoing neighbors, and building smaller connected subgraphs limited to a chosen depth. Basic statistics about the graph, such as how many entities and relationships exist and which entities are isolated, can also be calculated. Results can be exported to JSON for reuse in other programs or to Markdown for a readable report, and previously exported JSON can be reloaded and validated. SignalGraph is pure Python with no third-party runtime dependencies and no calls to external AI models or services, so it runs directly from a cloned copy of the repository with Python 3.9 or newer. It is used through a command line interface with subcommands for analyzing text or files, finding paths and neighbors in a saved graph, printing statistics, and exporting subgraphs. The project is marked as an early, alpha stage release and is licensed under MIT.