graphify, from Graphify Labs, turns a project folder into a queryable knowledge graph for AI assistants. It extracts code structure, ingests documents and multimedia, and exposes an MCP server plus a CLI for semantic queries that target specific subgraphs rather than raw files. Key elements include deterministic AST parsing, provenance-labeled edges, Leiden clustering, SHA-256 incremental caching, and HTML/JSON/Obsidian exports. Designed for software engineers and AI researchers who need more precise context retrieval from large repositories.

What tasks can you actually use it for?

The tool acts as a local memory layer so AI coding assistants can perform targeted semantic searches and explore architectural relationships inside projects. It maps entities and relationships from code, documentation, PDFs, images and transcripts into a structured graph, enabling agents to answer focused queries about dependencies, where symbols are defined, and which modules belong to the same architectural community. Example integrations include Claude Code, Cursor, and Codex.

How accurate and auditable are its retrievals?

Extraction is deterministic for code because the tool uses tree-sitter to build AST-based nodes without invoking an LLM for the extraction phase. Every graph edge is tagged with provenance as EXTRACTED, INFERRED, or AMBIGUOUS, so consumers can audit sources before trusting a relationship. The project reports a measured 71.5x reduction in token usage when agents query subgraphs instead of raw files, and a "reflect" feature collects feedback to improve future retrieval accuracy.

What file formats and environment does it accept?

Input flexibility is multimodal: the tool indexes source code (40+ languages), PDFs, SQL schemas, images, diagrams, and video or audio transcripts generated by local transcription tools. It runs on any environment supporting Python 3.10+ and updates graphs incrementally using SHA-256 caching so altered files do not require full reprocessing. Export options include interactive HTML visualizations, JSON for programmatic use, and Obsidian vaults for knowledge management.

Does it fit existing AI coding workflows?

Integration is geared to MCP setups and CLI workflows. The tool implements an MCP server that allows MCP-compliant hosts to query graphs directly and it also functions as a standalone command-line utility. The developer notes cross-platform Python support and IDE integration paths. Some users report that initial configuration for specific MCP clients can require manual adjustment of JSON paths, which adds a setup step before live querying.

Practical choice for teams needing verifiable, local project memory

The tool is a practical option for engineers and researchers who require a verifiable, queryable project memory for AI-assisted coding, since it creates structured, provenance-tagged context on-device. Expect a short configuration phase when connecting certain MCP clients, and plan to verify any INFERRED relationships before using them in critical decisions.