Arbor

Graph‑Native Intelligence for Codebases

Know what breaks before you break it.

Status (March 2026)

• Latest release: v1.7.0
• Main development trunk: main
• Maintenance line: release/v1.6 (v1.6.3)

Arbor v1.7.0 is the active release as of March 25, 2026.

Highlights

• Accurate Token Counting — tiktoken (cl100k_base) replaces heuristic estimates for precise LLM context budgets
• Fuzzy Symbol Suggestions — Typo tolerance with Jaro-Winkler matching: arbor refactor autth → "Did you mean: auth?"
• Enhanced MCP/AI Integration — Rich JSON output with confidence, roles, and edge explanations for Claude/Cursor
• Git-Aware Risk Workflows — arbor diff, arbor check, and arbor open for refactor confidence
• Incremental Refresh — arbor index --changed-only for faster re-index during active branches
• Better Python UX — Empty __init__.py handled silently (no false warnings)
• Polyglot Parser Expansion — New fallback parsing support for Kotlin, Swift, Ruby, PHP, and Shell families

What is Arbor?

Arbor is a local‑first impact analysis engine for large codebases. Instead of treating code as text, Arbor parses your project into a semantic dependency graph. This lets you trace real execution paths—callers, callees, imports, inheritance, and cross‑file relationships—so you can confidently understand the consequences of change.

Unlike keyword search or vector‑based RAG systems, Arbor answers questions like:

“If I change this function, what actually breaks?”

with structural certainty, not probabilistic guesses.

---

Example: Blast Radius Detection

Before refactoring detect_language, inspect its true impact:

$ arbor refactor detect_language

Analyzing detect_language...

Confidence: High | Role: Core Logic
• 15 callers, 3 dependencies
• Well-connected with manageable impact

> 18 nodes affected (4 direct, 14 transitive)

Immediate Impact:
  • parse_file (function)
  • get_parser (function)

Recommendation: Proceed with caution. Verify affected callers.

This is execution‑aware analysis, not text matching.

---

Graphical Interface

Arbor ships with a native GUI for interactive impact analysis.

arbor gui

GUI Capabilities

• Symbol Search – Instantly locate functions, classes, and methods
• Impact Visualization – Explore direct and transitive dependencies
• Privacy‑Safe – File paths are hidden by default for clean screenshots
• Export – Copy results as Markdown for PRs and design docs

The CLI and GUI share the same analysis engine—no feature gaps.

---

Quick Start

1. Install Arbor (CLI + GUI):

   cargo install arbor-graph-cli

   Or use package managers:

   # Homebrew (macOS/Linux)
   brew install Anandb71/tap/arbor
   
   # Scoop (Windows)
   scoop bucket add arbor https://github.com/Anandb71/arbor
   scoop install arbor
   
   # npm (any platform)
   npx @arbor-graph/cli
   
   # Docker
   docker pull ghcr.io/anandb71/arbor:latest

   Or use one-command installers (no Rust toolchain required):

   • macOS/Linux: curl -fsSL https://raw.githubusercontent.com/Anandb71/arbor/main/scripts/install.sh | bash
   • Windows (PowerShell): irm https://raw.githubusercontent.com/Anandb71/arbor/main/scripts/install.ps1 | iex

   See Installation Guide for version pinning and manual assets.

2. One-shot setup + first index:

cd your-project
arbor setup

3. Run Impact Analysis:

   arbor refactor <symbol-name>

For git-aware workflows:

arbor diff
arbor check --max-blast-radius 30
arbor open <symbol>

4. Launch the GUI:

   arbor gui

You can run Arbor from any nested subdirectory; it automatically resolves to your project root.

📘 See the Quickstart Guide for advanced workflows.

---

Release Channels & Branches

To keep maintenance and feature work clean:

• main → ongoing development
• release/v1.7 → current release line (v1.7.0)
• release/v1.6 → maintenance-only fixes for 1.6.x (v1.6.3 current)
• release/v1.5 → legacy maintenance/backport branch

This avoids shipping new features into older maintenance branches and keeps backports explicit.

Automated distribution (not GitHub-only)

Tagged releases (vX.Y.Z) are automated across channels:

• GitHub Release assets (CLI binaries for Linux, macOS, Windows)
• crates.io packages
• GHCR container images
• VS Code Marketplace extension
• Open VSX extension
• Homebrew formula
• Scoop manifest
• npm wrapper package

Maintainer setup and verification steps are in docs/RELEASING.md.

---

MCP Directory Listing

Arbor is listed on Glama MCP Directory:

• Glama: https://glama.ai/mcp/servers/@Anandb71/arbor

MCP Health Badges

Claude Code MCP Quick Install

claude mcp add --transport stdio --scope project arbor -- arbor bridge

Then verify with:

claude mcp list

Universal MCP Integration Kit

Arbor ships ready-to-copy MCP templates and bootstrap scripts for multi-client onboarding:

• Templates: templates/mcp/
• macOS/Linux bootstrap: scripts/setup-mcp.sh
• Windows bootstrap: scripts/setup-mcp.ps1

# Generate project-scoped config for Claude Code + Cursor + VS Code
./scripts/setup-mcp.sh --client all --target-dir .

# Windows equivalent
./scripts/setup-mcp.ps1 -Client all -TargetDir .

This creates:

• .mcp.json (Claude Code project scope)
• .cursor/mcp.json
• .vscode/mcp.json

For enterprise-managed setups, see docs/MCP_INTEGRATION.md (policy + managed registration notes).

---

GitHub Marketplace Action

Arbor now includes a reusable GitHub Action at the repo root (action.yml) so teams can run impact analysis directly in CI.

Quick usage

name: Arbor Check
on: [pull_request]

jobs:
  arbor:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5
      - uses: Anandb71/arbor@release/v1.7
        with:
          command: check . --max-blast-radius 30

Inputs:

• command (default: status .) — Arbor subcommand + args (without leading arbor)
• arbor-version (default: latest) — install from crates.io at a pinned version
• install-from-source (default: false) — build from this repository source

To publish this action in GitHub Marketplace: create a release tag (for example v1), add marketplace listing metadata in the release flow, and submit from the repository's Marketplace listing page.

---

Documentation Hub

• Quickstart: docs/QUICKSTART.md
• Installation: docs/INSTALL.md
• Architecture: docs/ARCHITECTURE.md
• MCP Integration: docs/MCP_INTEGRATION.md
• Protocol Specification: docs/PROTOCOL.md
• Roadmap: docs/ROADMAP.md
• Release Runbook (all channels): docs/RELEASING.md
• Release Notes (v1.6): docs/RELEASE_NOTES_v1.6.0.md
• Release Notes (v1.6.1.1): docs/RELEASE_NOTES_v1.6.1.1.md
• Release Notes (v1.6.2): docs/RELEASE_NOTES_v1.6.2.md
• Release Notes (v1.7.0): docs/RELEASE_NOTES_v1.7.0.md

---

Why Arbor?

Most AI coding tools treat code as unstructured text, relying on vector similarity. This approach is fast—but imprecise.

Arbor builds a graph.

Every function, class, and module is a node. Every call, import, and reference is an edge. When you ask a question, Arbor follows the graph—the same way your program executes.

Traditional RAG:              Arbor Graph Analysis:

"auth" → 47 results          AuthController
(keyword similarity)           ├── calls → TokenMiddleware
                               ├── queries → UserRepository
                               └── emits → AuthEvent

The result: deterministic, explainable answers.

---

Core Features

Symbol Resolution

Arbor resolves symbols across files with full qualification:

• Imports and re‑exports
• Inheritance and interfaces
• Overloads and namespaces

User in auth.ts is never confused with User in types.ts.

---

Supported Languages

Language	Status	Parser Coverage
Rust	✅	Functions, Structs, Traits, Impls, Macros
TypeScript	✅	Classes, Interfaces, Types, Imports, JSX
JavaScript	✅	Functions, Classes, Vars, Imports
Python	✅	Classes, Functions, Imports, Decorators
Go	✅	Structs, Interfaces, Funcs, Methods
Java	✅	Classes, Interfaces, Methods, Fields
C	✅	Structs, Functions, Enums, Typedefs
C++	✅	Classes, Namespaces, Templates
C#	✅	Classes, Methods, Properties, Interfaces
Dart	✅	Classes, Mixins, Widgets
Kotlin	✅	Functions, Classes, Interfaces (fallback parser)
Swift	✅	Functions, Types, Protocols (fallback parser)
Ruby	✅	Methods, Classes, Modules (fallback parser)
PHP	✅	Functions, Classes, Traits (fallback parser)
Shell	✅	Function definitions (fallback parser)

Python note: Decorators, __init__.py, and @dataclass are statically analyzed. Dynamic dispatch is flagged with reduced confidence.

---

Build from Source

git clone https://github.com/Anandb71/arbor.git
cd arbor
cargo build --release
cargo test

Frictionless setup notes

• No external graph database is required. Arbor is local-first and uses embedded storage.
• From a fresh clone, contributors can start with just:
  • cargo build
  • cargo test

If you prefer containerized setup, run:

docker compose up --build arbor-dev

Linux GUI Dependencies

sudo apt-get install -y pkg-config libx11-dev libxcb-shape0-dev libxcb-xfixes0-dev \
  libxkbcommon-dev libgtk-3-dev libfontconfig1-dev libasound2-dev libssl-dev cmake

---

Troubleshooting

Symbol not found?

• .gitignore – Arbor respects it (arbor status --files)
• File type – Ensure the extension is supported
• Empty files – Skipped (except __init__.py)
• Dynamic calls – eval / runtime reflection may not resolve
• Case sensitivity – Use arbor query <partial> to search

Empty graph?

Run arbor status to verify file detection and parser health.

Need environment diagnostics?

Run arbor doctor (or arbor check-health) to verify ports, project structure, and integration readiness.

Repo suddenly huge (multi-GB)?

Rust and Flutter build artifacts can grow quickly during iterative testing.

• Windows PowerShell: ./scripts/clean.ps1
• macOS/Linux: ./scripts/clean.sh
• Deeper cleanup (also removes local Arbor/Flutter cache artifacts):
  • PowerShell: ./scripts/clean.ps1 -Deep
  • Bash: ./scripts/clean.sh --deep

This is safe for source code; it only removes generated artifacts that can be rebuilt.

---

Security Model

Arbor is Local‑First by design:

• No data exfiltration
• Fully offline
• No API keys
• Fully open source

Your code never leaves your machine.

---

License

MIT License. See LICENSE for details.

⭐ Star Arbor on GitHub

---

Contributors

_{6 contributors | View all}