Bold claim? Here's the proof.
4 days ago I released Lutum Veritas - an open-source Deep Research tool that does one thing differently: It tells the truth.
The Benchmark Results:
I ran the same queries through ChatGPT Deep Research, Google Gemini, Perplexity Pro, and Veritas.
• ChatGPT: Fabricated 4-5 citations that don't exist
• Gemini: 30-40% incorrect data
• Perplexity: Surface-level, paywalled ($20/mo)
• Veritas: 100% verifiable sources, $0.08 per report
Full benchmark: https://veritas-test.neocities.org
What makes it different:
When data doesn't exist, Veritas says "we don't know" instead of making shit up.
Radical concept, I know.
The Tech:
• Camoufox scraper (0% bot detection)
• Dual-verification pipeline
• Budget models: Gemini Flash Lite + Qwen 235B
• Cost: $0.08 per 30-min deep research report
The Numbers (4 days post-launch):
• 46 GitHub stars
• 7.3% conversion rate (industry avg: 1-3%)
• 630 unique visitors
• Featured: Hacker News, Product Hunt, DeepLearning.AI
New: Ask Mode (v1.3.0)
60-second verified answers for $0.0024 each. That's 400 answers for $1.
The Lesson:
You don't need billions in VC funding to beat billion-dollar companies. You need the right philosophy:
Truth > Usefulness.
Evidence > Speculation.
"I don't know" > Hallucination.
Open source. AGPL-3.0. Runs on your machine.
GitHub: https://github.com/IamLumae/Project-Lutum-Veritas
Try it. Break it. Tell me I'm wrong.
I'll wait.
Top comments (0)