Turn your codebase into a queryable knowledge graph locally
Run a single command in your project folder, and scattered code, documentation, and PDFs are mapped into a single graph. The tool does not read file contents directly; instead, it follows connections between concepts to find answers. It creates the working environment developers need to grasp complex structures at a glance.
Walking actual graphs without vector stores
The tool builds an actual graph of nodes and edges rather than a text-similarity-based vector index. Every connection has a clear source. Connections read directly from the source are marked as EXTRACTED, while those inferred by the tool are marked as INFERRED, allowing you to distinguish reliability. When you ask a question, you get answers by tracing the path between two concepts or by explaining a specific concept.
Local code analysis
Code files are parsed deterministically locally using tree-sitter. No LLM is involved in this process, and no data leaves your machine. It supports approximately 40 or more programming languages and identifies relationships such as function calls, imports, and inheritance. Documents, PDFs, images, and videos can optionally undergo semantic analysis using an AI model, but if you only handle code, it works fully offline without an API key.
Installation and first run
You need Python 3.10 or higher and uv or pipx. Install the CLI with the uv tool install graphifyy command, then run graphify install to automatically detect your current AI assistant and register the skill. The outputs include graph.html, which opens in a browser, GRAPH_REPORT.md, which summarizes key concepts, and graph.json, which contains the full graph data.
Benchmark results and caveats
In the LOCOMO benchmark, it recorded a recall@10 of 0.497, with LLM credit consumption confirmed at 0. This is higher than mem0 (0.048) or supermemory (0.149). However, if you want semantic analysis of documents or media files, you need to configure the API key for that backend. If you only want code-only analysis, use the --code-only option. The official PyPI package is graphifyy (double y), so be careful not to confuse it with other similarly named packages.
