graphmind
A local-first code-intelligence tool that turns a codebase into a queryable knowledge graph for AI assistants, combining an AST-based structural graph, semantic embeddings, and persistent cross-session memory to cut token usage versus raw search.
graphmind builds a function-level knowledge graph of a codebase using tree-sitter across more than 30 languages, adds semantic embeddings for vector search over symbols using local ONNX models or a hosted embedding provider, and keeps a persistent, declarative store of decisions, patterns, and conventions that survives across sessions rather than being rediscovered by the AI assistant each time. A fourth layer links dependencies and relationships across multiple registered repositories.
The project reports that its structured search returns ranked results in under 300 tokens for a query that a raw recursive grep search over a roughly 100,000-line codebase would need more than 1.5 million tokens to answer. It works with Claude Code, Cursor, Windsurf, Cline, Zed, Continue, and any other MCP-compatible AI assistant.
It runs entirely locally with no cloud dependency, open ports, or telemetry, and installs through a macOS desktop app with guided onboarding, a shell script, or Homebrew on macOS or Linux; a Linux or Windows install is CLI-only.