Stop wrestling with context. Let Ander read your code, understand your architecture, and generate a blueprint that makes Claude, Cursor, or ChatGPT build exactly what you want — on the first attempt.
Pick a folder → Select files → Describe what you want → Get a perfect blueprint → Paste into any AI
You open Claude / Cursor / ChatGPT. You want it to add a feature. So you start copying files, explaining your folder structure, describing your patterns, listing your dependencies...
20 minutes later, you're still writing the prompt. And when the AI finally responds, it hallucinated half your architecture.
Sound familiar?
Ander scans your codebase, understands your conventions, and generates a zero-code architectural blueprint — a prompt so detailed and precise that any AI coding assistant can implement your feature correctly on the first try.
No code in the output. Just pure intent, specification, and design. The receiving AI doesn't copy-paste — it understands.
┌─────────┐ ┌──────────┐ ┌──────────┐ ┌───────────┐ ┌──────────────┐
│ 1. Pick │───▶│ 2. Scan │───▶│ 3. Curate│───▶│ 4. Describe│───▶│ 5. Generate │
│ Folder │ │ Files │ │ Context │ │ Feature │ │ Blueprint │
└─────────┘ └──────────┘ └──────────┘ └───────────┘ └──────────────┘
Respects AI-powered "Add OAuth Copy & paste
.gitignore smart select with Google" into any AI
Then paste the blueprint into Claude, Cursor, ChatGPT, Copilot — whatever you use. It just works.
| Feature | Description | |
|---|---|---|
| 🔍 | Smart Project Scanner | Walks your codebase respecting .gitignore. Skips binaries, lock files, and node_modules. Prioritizes source code over configs. |
| 🔒 | Security-Aware | Auto-excludes .env, secrets, private keys, and certificates. Your sensitive files never touch the AI. |
| 🧠 | AI-Powered File Selection | Click "Smart Select" and let the AI recommend which files matter for your feature. You stay in control. |
| ⚡ | Real-Time Streaming | Watch your blueprint generate token-by-token. No waiting for batch responses. |
| 🔄 | Dual AI Providers | Switch between Gemini 2.5 Pro (1M context) and Claude (Opus / Sonnet / Haiku) with one click. |
| 🧩 | Extended Thinking | Opus uses 32K thinking tokens, Sonnet uses 16K — deeper reasoning for complex architectures. |
| 💬 | Follow-up & Refinement | Ask clarifying questions or refine the blueprint without starting over. Full conversation context preserved. |
| 📋 | One-Click Copy | Copy the blueprint to clipboard or send it directly to a Claude Code window. |
| 📁 | Conversation History | Every generation saved to local SQLite. Search, filter, and revisit past blueprints. |
| 🖥️ | Desktop-Native | No server. No SaaS. No rate limits. Runs locally with your own API quota. |
| 📱 | Responsive UI | Works on any screen — desktop, tablet, or a tiny laptop. |
- Feature Implementation — "Add dark mode to the settings page"
- New Project — "Design a REST API for a task manager"
- Bug Fix — "This error happens when I click submit" (paste the stack trace)
- Follow-up — "What about error handling for the OAuth flow?"
- Refinement — "Make it use Redis instead of in-memory cache"
- Rust (latest stable)
- Node.js 18+ and pnpm
- Tauri v2 system dependencies
git clone https://github.com/riftzen-bit/ander.git
cd ander
pnpm install
pnpm startThat's it. The app opens, you authenticate with Google (for Gemini) or paste a Claude API key, and you're generating blueprints.
pnpm tauri build| Layer | Technology | Why |
|---|---|---|
| Desktop Framework | Tauri v2 | Tiny binary, native performance, secure IPC |
| Backend | Rust (2021 edition) | Memory safety, async I/O with Tokio, fast file scanning |
| Frontend | React 19 + TypeScript 5.8 | Modern hooks, strict types, component architecture |
| Build Tool | Vite 7 | Instant HMR, fast builds |
| Styling | Tailwind CSS v4 | Utility-first, responsive, dark mode |
| AI (Primary) | Gemini 2.5 Pro | 1M token context via Google Code Assist API |
| AI (Secondary) | Claude (Opus / Sonnet / Haiku) | Extended thinking, flexible model selection |
| Auth | Google OAuth 2.0 + Claude API keys | Same OAuth as Gemini CLI, auto-detect Claude credentials |
| Database | SQLite (rusqlite, bundled) | Local history, zero config |
| File Watching | notify + debouncer | Live project updates with loop prevention |
ander/
├── src/ # React frontend
│ ├── components/wizard/ # 5-step wizard UI
│ ├── hooks/ # Auth, scanning, generation, context tracking
│ ├── lib/ # Wizard state & types
│ └── App.tsx # Root component
├── src-tauri/ # Rust backend
│ ├── src/
│ │ ├── scanner.rs # .gitignore-aware file walker
│ │ ├── gemini.rs # Gemini API integration
│ │ ├── anthropic.rs # Claude API integration
│ │ ├── auth.rs # OAuth + token management
│ │ ├── prompts.rs # Shared system prompts
│ │ ├── db.rs # SQLite history
│ │ └── watcher.rs # File system watcher
│ └── tauri.conf.json # App configuration
├── package.json
└── Cargo.toml
Ander doesn't just dump your files into a prompt. It instructs the AI to think like a staff engineer reverse-engineering an unfamiliar codebase:
- Analyze — Understand the architecture, patterns, and conventions from the selected files
- Design — Plan the implementation respecting existing patterns, naming conventions, and project structure
- Specify — Output a precise blueprint: file paths, function signatures, data flow, edge cases, error handling
- No code — The blueprint describes what and why, never how. The receiving AI writes the code.
This separation of concerns is what makes the output reliable. The generating AI focuses on understanding; the implementing AI focuses on building.
Contributions welcome! Whether it's a bug fix, new feature, or documentation improvement:
- Fork the repository
- Create a feature branch (
git checkout -b feat/amazing-feature) - Commit your changes (
git commit -m 'feat: add amazing feature') - Push to the branch (
git push origin feat/amazing-feature) - Open a Pull Request
Please follow Conventional Commits for commit messages.
Report Bug · Request Feature · Discussions
MIT License · Built with Rust, React, and an unreasonable hatred for copy-pasting files into ChatGPT.