github.com/Kehl-io/nestweaver · MIT · v12.2.3

NestWeaver

Code intelligence knowledge graph for AI agents — 17-34x faster queries than Graphify on large repos in our June 2026 benchmarks.

NestWeaver — a fierce geometric raptor built from graph nodes

Why NestWeaver exists

AI agents waste their context window. Ask an agent to fix a bug and it reads entire files — thousands of tokens of boilerplate and unrelated code — just to find the three functions that matter. The context window fills up before the agent even starts thinking.

NestWeaver solves this at the source. It builds a persistent code knowledge graph from your codebase — parsing every symbol, resolving every cross-file reference — then uses Personalized PageRank to rank which nodes are actually relevant to the task at hand. When an agent needs context, NestWeaver returns a precomputed subgraph sized to fit the token budget, not raw file contents. Agents get precise, ranked answers instead of entire files.

How it works

  1. Index

    Tree-sitter parses 32 languages into a unified symbol graph. Every function, class, import, and type is a node; every reference is an edge.

  2. Query

    Cross-file symbol resolution with confidence scoring — NestWeaver knows not just where a symbol is defined, but how certain that link is.

  3. Context

    Personalized PageRank walks the graph from a task-relevant seed set and extracts the highest-value subgraph that fits your token budget.

  4. Connect

    43 MCP tools expose the graph to any MCP-aware agent. Agents query by symbol, file, or natural language — no file reads needed.

Features

Graph Intelligence

  • Personalized PageRank for task-relevant context extraction
  • Cross-file symbol resolution with confidence scoring
  • Token-budget-aware subgraph extraction
  • Incremental graph updates — only reparse changed files
  • 43 MCP tools for agent integration

Multi-Language

  • 32 languages via Tree-sitter: Rust, TypeScript, Python, Go, Java, C/C++, Lua, Scala, Elixir, Zig, Vue, Svelte, and more
  • Unified symbol graph across all languages in a polyglot repo
  • Package-aware imports for Go, Python, Kotlin, and Java; type-aware calls in 21 languages
  • Language-agnostic MCP API — one interface for everything
  • Extensible parser layer for adding new languages

Brain

  • Index Markdown and Obsidian vaults alongside code
  • Link documentation, ADRs, and RFCs to the code symbols they describe
  • Natural language queries over code + docs combined
  • Unified knowledge graph spanning code and prose
  • Supports Obsidian-style wikilinks and frontmatter

Platform

  • macOS menubar app (source build) with Metal-accelerated embeddings
  • Written in Rust, with prebuilt CLI builds for macOS and Linux on x86_64 and arm64
  • Web UI with interactive 3D graph visualization
  • Background daemon owns the database, so the CLI, AI tools, and web UI share one graph
  • MIT licensed, open source — install via npm, a release archive, or a source build

Get started

Run this in your terminal:

npm install --global nestweaver

Then, from your repository root, nestweaver index --repo . builds the graph and nestweaver setup connects your AI tools, with 16 integrations including Claude Code, Codex, Cursor, and GitHub Copilot. Release archives (Linux needs glibc 2.35+ and the bundled lib/ folder) and source builds are covered in the install guide.

12.2.3 hardens Go resolution, restore safety, source freshness and embedding recovery. Upgrading an existing graph? Resolver generation 17 needs nestweaver index --repo <path> --force once per repository and nestweaver brain refresh <vault> once per vault. Use your usual database and config bindings.

Status

v12.2.3

NestWeaver is production-ready with a 43-tool MCP server, 32-language parsing, and a shared background daemon. Contributions are welcome — see CONTRIBUTING.md for build, test, and commit guidelines.