Lutum Veritas

Open Source Deep Research Engine
"The search for a truth can never be worth more than the search to question it."

Features • Installation • Quick Start • How It Works • Tech Stack • License

Perplexity, OpenAI and Google deliver summaries. I wanted truth.

So I stopped waiting and built it myself. The Camoufox scraper cuts through Cloudflare, Bloomberg and paywalls with 0% detection. The recursive pipeline passes context forward – each research point knows what the previous ones discovered. Claim Audits force the model into self-reflection instead of blind assertions.

The result: 203,000 characters of academic depth for a single query. Cost: under 20 cents. That's orders of magnitude cheaper than OpenAI o3 and qualitatively in a different league.

This isn't an "alternative" to existing tools. This is proof that a solo dev with the right architecture can beat billion-dollar corporations at what should be their core competency: deep, verifiable knowledge.

The bar for Deep Research is set right here.

— Martin Gehrken, January 30, 2026

Demo

Full Deep Research workflow: Query → Clarification → Plan → Research → Final Report

🔬 Benchmark Results

Independent comparison of Lutum Veritas vs. ChatGPT Deep Research vs. Perplexity Pro vs. Gemini Advanced:

📊 View Full Benchmark Report (German) 🌐 Auto-Translated Version (EN) (Google Translate)

• 4-way comparison with identical scientific query
• Quantitative metrics: output length, sources, costs, duration
• Quality review by competing AIs
• 16-agent fact-check protocol for hallucinations

TL;DR: Lutum delivered 103k characters with 90 sources for $0.19. ChatGPT: 12k chars, 25 sources, fabricated citations. Gemini: 24k chars, data minimization detected. Perplexity: 21k chars, $20/mo subscription.

🏆 Thank You, Community!

The First 4 Days

Released: January 30, 2026 at 09:00 Current: February 3, 2026 at 10:00

What you've accomplished:

📊 Traffic Stats (First 4 Days):

• ⭐ 46 Stars (+24 in 24h, +109%)
• 🍴 6 Forks (+3 in 24h, +100%)
• 🔥 390 Clones (+101 in 24h, +35%)
• 👀 1,016 Views (+673 in 24h, +196%)
• 👥 630 Unique Visitors (+490 in 24h, +350%)
• 🎯 7.3% Conversion Rate (Stars/Visitors - Industry avg: 1-3%)
• 🌍 Featured on: Hacker News, ComputerBase.de, Hardwareluxx, Product Hunt, DeepLearning.AI Community

You Made This Possible

In just 3 days, you've helped prove something important:

A solo developer with an idea can stand toe-to-toe with billion-dollar companies. You don't need permission to build something great. You need passion, code, and a community that believes.

Every star, every clone, every "this is exactly what I needed" message keeps this project going.

You're not just users. You're proof that Research Without Permission isn't just a tagline - it's a movement.

Thank you for standing against the giants. 🚀

Want to join the fight? ⭐ Star the repo · 🐛 Report issues · 💬 Share with your network · 🔨 Contribute on GitHub

What is Lutum Veritas?

Lutum Veritas is a self-hosted Deep Research Engine that transforms any question into a comprehensive research document. Unlike Perplexity, ChatGPT, or Google's AI Overview, you bring your own API key and everything runs locally on your machine - Windows, macOS, or Linux.

Why Use This?

Features

🔬 Deep Research Pipeline

Your Question
     ↓
┌─────────────────────────────────────────────────────┐
│  1. CLARIFICATION                                   │
│     AI asks smart follow-up questions               │
├─────────────────────────────────────────────────────┤
│  2. RESEARCH PLAN                                   │
│     Creates structured investigation points         │
├─────────────────────────────────────────────────────┤
│  3. DEEP RESEARCH (per point)                       │
│     Think → Search → Pick URLs → Scrape → Dossier   │
├─────────────────────────────────────────────────────┤
│  4. FINAL SYNTHESIS                                 │
│     Cross-reference all findings into one document  │
└─────────────────────────────────────────────────────┘
     ↓
📄 Comprehensive Report (5,000-10,000+ words)

🎓 Academic Mode

Hierarchical research with autonomous areas:

• Parallel Processing: Research areas independently
• Meta-Synthesis: Find cross-connections between areas
• Toulmin Argumentation: Structured academic reasoning
• Evidence Grading: Rate source quality (Level I-VII)
• Claim Audit Tables: Confidence ratings for every claim
• 200,000+ character outputs: Full academic depth, no shortcuts

🎯 Ask Mode - NEW in v1.3.0

Quick answers. Verified facts. No hallucinations.

Ask Mode workflow: Question → C1-C6 stages → Verified Answer with Citations

The new Deep Question mode bridges the gap between chat and Deep Research. It's the tool you keep open when your question isn't "big enough" for a 20-minute deep dive, but you need more than an unverified chat response based on outdated, biased training data.

The difference:

• Regular Chat: No verification. No live search. Answers from stale training data.
• Ask Mode: Every answer is researched, sourced, and self-verified against a second round of sources.

When you need a real answer on the first try: this is your mode.

Features

• 6-Stage Pipeline: Intent → Knowledge → Search → Scrape → Answer → Verify → Fact-Check (~70-90s)
• Dual-Scraping Phases: First scrape for answer, second scrape for verification
• Citation System: Inline citations [1], [2] for sources + [V1], [V2] for verification
• Claim Auditing: Every claim is fact-checked against additional sources
• Auto-Language Detection: Responds in same language as your question
• Separate Sessions: Ask sessions stored separately from Deep Research

Cost

Cost? A joke. ~400 queries for $1.

• 0.24 cents per answer
• 416 verified answers for $1
• Model: google/gemini-2.5-flash-lite-preview-09-2025

💻 Desktop App Features

🛡️ Zero Detection Scraping

Powered by Camoufox - a hardened Firefox fork that bypasses:

• Cloudflare
• DataDome
• PerimeterX
• Bloomberg, TCGPlayer, and most anti-bot systems

Installation

Option A: Download Windows Installer (Easiest for Windows)

Platform: Windows only Requirements: Python 3.11+ installed (python.org)

1. Download Lutum Veritas_1.2.4_x64-setup.exe from Releases
2. Run the installer
   • If Python is not found, the installer will prompt you to install it
   • Dependencies are installed automatically via pip
3. Launch Lutum Veritas from your Start Menu
4. Select your API Provider in Settings (OpenRouter, OpenAI, Anthropic, Google Gemini, or HuggingFace)
5. Enter your API Key
6. Start researching!

Note: The backend starts automatically when you open the app. No separate process to manage.

Option B: Install via uv/uvx (Cross-Platform)

Platform: Windows, macOS, Linux Requirements: uv package manager

This is the recommended method for macOS and Linux users, and also works great on Windows if you prefer command-line tools.

Install and Run:

# Option 1: Install as a persistent tool
uv tool install git+https://github.com/IamLumae/lutum-veritas.git

# Then run anytime with:
lutum-veritas

# Option 2: Run directly without installation (ephemeral)
uvx --from git+https://github.com/IamLumae/lutum-veritas.git lutum-veritas

Both commands will:

• Install all dependencies automatically
• Start the backend server on http://localhost:8420
• Open your browser to the web interface

Tip: Use uv tool install if you plan to use Lutum regularly. Use uvx for one-time runs or testing.

Option C: Build from Source (Developers)

Platform: All platforms Requirements:

• Python 3.11+
• Node.js 18+
• Rust (for Tauri)

# Clone
git clone https://github.com/IamLumae/lutum-veritas.git
cd lutum-veritas

# Backend
cd lutum-backend
pip install -r requirements.txt
python main.py

# Frontend (new terminal)
cd lutum-desktop
npm install
npm run tauri dev

Quick Start

1. Launch App - Open Lutum Veritas (backend starts automatically)
2. Select Provider - Settings → Choose OpenRouter, OpenAI, Anthropic, Gemini, or HuggingFace
3. Enter API Key - Enter your API key for the selected provider
4. Ask Anything - Type your research question
5. Answer Clarifications - Help the AI understand your needs
6. Review Plan - Approve or modify the research plan
7. Click "Let's Go" - Watch the magic happen
8. Export - Download your research as MD or PDF

How It Works

Architecture

┌─────────────────────────────────────────────────────────────┐
│                    LUTUM VERITAS DESKTOP                     │
│  ┌───────────────────────────────────────────────────────┐  │
│  │              Tauri Shell (Rust + WebView)              │  │
│  │         Auto-starts Python backend on launch           │  │
│  │  ┌─────────────────────────────────────────────────┐  │  │
│  │  │           React Frontend (TypeScript)           │  │  │
│  │  │  • Chat Interface     • Session Management      │  │  │
│  │  │  • Live Status        • Markdown Rendering      │  │  │
│  │  └─────────────────────────────────────────────────┘  │  │
│  └───────────────────────────────────────────────────────┘  │
│                          ↕ HTTP                              │
│  ┌───────────────────────────────────────────────────────┐  │
│  │              FastAPI Backend (Python)                  │  │
│  │  • Research Orchestrator    • LLM Integration         │  │
│  │  • Session Persistence      • SSE Streaming           │  │
│  │  ┌─────────────────────────────────────────────────┐  │  │
│  │  │         Camoufox Scraper (Firefox Fork)         │  │  │
│  │  │              0% Bot Detection Rate               │  │  │
│  │  └─────────────────────────────────────────────────┘  │  │
│  └───────────────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────────────┘

LLM Pipeline

Supported Providers: OpenRouter (200+ models), OpenAI, Anthropic, Google Gemini, HuggingFace - bring your own API key.

Tech Stack

Project Structure

lutum-veritas/
├── lutum/                      # Core Python library
│   ├── researcher/
│   │   └── prompts/            # LLM prompts (Think, Pick, Dossier, Synthesis)
│   └── scrapers/
│       └── camoufox_scraper.py # Zero-detection web scraper
├── lutum-backend/              # FastAPI server
│   └── routes/
│       └── research.py         # Research pipeline orchestrator
├── lutum-desktop/              # Tauri desktop app
│   ├── src/
│   │   ├── components/         # React components
│   │   ├── hooks/              # useBackend API hook
│   │   └── stores/             # Session state management
│   └── src-tauri/
│       ├── src/lib.rs          # Auto-start backend logic
│       └── nsis-hooks.nsh      # Installer: Python check + pip install
├── LICENSE                     # AGPL-3.0
├── NOTICE                      # Copyright & commercial licensing
└── README.md                   # You are here

API Reference

Endpoints

SSE Events (Deep Research)

// Status updates
{"type": "status", "message": "Searching Google..."}

// Sources found
{"type": "sources", "urls": ["https://...", "https://..."]}

// Point completed
{"type": "point_complete", "point_title": "...", "key_learnings": "..."}

// Synthesis starting
{"type": "synthesis_start", "dossier_count": 5, "total_sources": 45}

// Research complete
{"type": "done", "data": {"final_document": "...", "source_registry": {...}}}

Cost Comparison

Real benchmark: 513k input tokens, 55k output tokens

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing)
5. Open a Pull Request

Development Setup

# Backend (with hot reload)
cd lutum-backend
uvicorn main:app --reload --port 8420

# Frontend (with hot reload)
cd lutum-desktop
npm run tauri dev

License

Lutum Veritas is licensed under the GNU Affero General Public License v3.0.

This means:

• ✅ Free to use, modify, and distribute
• ✅ Commercial use allowed
• ⚠️ Must disclose source code (including SaaS)
• ⚠️ Modifications must use same license

Commercial Licensing

Need to use Lutum Veritas without AGPL obligations? Commercial licenses are available.

Contact: iamlumae@gmail.com

Security

v1.3.0 Installer:

• VirusTotal: 0/65 detections ✅ Clean
• SHA256: 96faa40b63150632a96486086a2a778a4ec8a19b31dd06907d5178bb961fc287

Acknowledgments

• Camoufox - The magic behind zero-detection scraping
• Tauri - Lightweight desktop app framework
• OpenRouter - Unified LLM API access

Built with obsessive attention to detail
Because truth shouldn't be locked behind paywalls

@IamLumae