Alchemist

Ask your local data what it knows. A local-first desktop app for macOS and Windows that opens your databases, MemPalace, and config files — and lets you ask questions in plain English.

---

✨ What it does

Alchemist turns your local data into a conversation. Open a SQLite database, point it at your MemPalace, or drop in a YAML file — then just ask questions.

Drop in a database, ask a question, get answers. No cloud, no accounts, no data leaves your machine.

You: "How many users signed up last month?"
Alchemist: SELECT COUNT(*) FROM users WHERE created_at >= date('now', '-30 days')
           → 1,247 users

Features

🗄️ SQLite	Open any .db or .sqlite file — read-only, safe, can't accidentally modify your data
🧠 MemPalace	Connect to your palace, browse wings and rooms, search by meaning
💬 Natural language	Just ask. Alchemist generates safe SQL from your question
🔍 SQL preview	See the SQL before it runs. Edit it, cancel it, or approve it
📊 Charts	Results as bar, line, or pie charts — auto-detects numeric columns
✨ Saved spells	Name and reuse your best queries across sessions
🤖 Your model, your rules	Ollama (local), OpenAI, Google Gemini, DeepSeek — or any OpenAI-compatible provider
🔒 Private by design	No telemetry, no accounts, no cloud dependency. Your keys stay on your machine
💻 Mac & Windows	One app, both platforms — same witchy experience on macOS 13+ and Windows 10/11

---

🚀 Quick start

Download

Grab the latest build for your computer from Releases — no account, no sign-up, nothing to configure. Just download and open.

🪟 Windows 10 / 11 Download Alchemist_x64-setup.exe, double-click it, and click through the installer. Alchemist then lives in your Start menu, ready to go.

🍎 macOS 13+ (Ventura or later — Apple Silicon or Intel) Download the .dmg, open it, and drag Alchemist into your Applications folder.

That's the whole setup. Open the app and you're ready for the next step.

Setup a provider

Alchemist needs an AI model to generate queries from your questions.

Local (free, private):

1. Install Ollama
2. Pull a model: ollama pull llama3.2
3. Open Alchemist — Ollama is auto-detected

Cloud (bring your own key):

• OpenAI, Google AI Studio, DeepSeek — or any OpenAI-compatible endpoint (Groq, Together, vLLM, etc.)
• Add your API key in Settings → pick a provider

Your first query

1. Open Alchemist — it auto-loads MemPalace if you have one
2. Or drag a database — any .db file onto the vault screen
3. Type a question — "show me my tables" or "how many records?"
4. Review the SQL — Alchemist shows you what it's about to run
5. Hit Run — see results as a table or chart
6. Save it — click "Save as Spell" to reuse later

---

🛠️ Development

Curious enough to run it from source — or build your own copy? You'll need Node.js and Rust installed first (on Windows, that also means the Microsoft C++ Build Tools — the Rust installer points you to them).

macOS / Linux:

git clone https://github.com/witchdaddylabs/alchemist.git
cd alchemist
npm install
npm run tauri dev

Windows: we've bundled a couple of double-click-friendly helper scripts so you don't have to memorise any commands:

• run-windows.bat — launches the app. It checks you've got everything, installs what's missing, and opens Alchemist.
• build-windows.ps1 — builds your very own installer at src-tauri/target/release/bundle/nsis/.

There is also a report-only duplicate audit helper for local folder trees:

npm run audit:duplicates -- --root "/path/to/folder"

Stack

Layer	What
Desktop	Tauri v2 — lightweight native shell
Frontend	React 19, TypeScript, Vite
UI	shadcn/ui (dark mode, purple neon accents)
State	Zustand
Backend	Rust — SQLite, ChromaDB, MemPalace parsing
SQL safety	sqlparser-rs — SELECT-only, single statement, auto-LIMIT
AI providers	Ollama, OpenAI, Gemini, DeepSeek, custom OpenAI-compatible
Auto-update	tauri-plugin-updater — checks GitHub Releases on launch

Project layout

src/                          # React frontend
├── features/vault/           # File picker, recent sources
├── features/workspace/       # Schema sidebar, chat, inspector panels
├── features/results/         # Tables, charts, export
├── features/spells/          # Saved query library
├── features/settings/        # Provider config, API keys, privacy
└── state/                    # Zustand stores

src-tauri/src/                # Rust backend
├── db.rs                     # Read-only SQLite
├── chroma.rs                 # ChromaDB vector search
├── palace.rs                 # MemPalace YAML parser
├── llm.rs                    # Prompt builder, query generation
├── providers/                # Ollama, OpenAI, Gemini clients
├── secret_store.rs           # API key storage (macOS Keychain + local fallback)
├── paths.rs                  # Cross-platform home/data paths
├── validate.rs               # sqlparser-rs safety checks
├── models.rs                 # Shared type contracts
├── errors.rs                 # Error types
└── commands/app.rs           # 17 Tauri commands

---

🔐 Security

• Read-only by design — SQLite opened with SQLITE_OPEN_READONLY
• SQL validation — every query parsed by sqlparser-rs: blocks writes, multi-statements, dangerous pragmas
• Keys stay local — on macOS your API keys live in the system Keychain; on Windows they're kept in your private user folder (%USERPROFILE%\.alchemist). Either way, they never leave your machine
• No telemetry — zero analytics, crash reporting, or network calls to WDL servers
• Open source — MIT licensed. Build it yourself if you want

---

🧩 What's next?

Alchemist is fully functional and shippable. A few things we're thinking about:

Feature	Why it's cool
Pagination	Load-more for results over 1,000 rows
Multi-source queries	Cross-reference SQLite + ChromaDB in one chat
Plugin architecture	Postgres, MySQL, CSV support as plugins
Native export dialog	macOS save dialog for CSV/JSON exports

Want to help? PRs welcome!

---

_{Built by Witch Daddy Labs in Melbourne, Australia.
MIT License · Copyright (c) 2026 Billy @ Witch Daddy Labs}