55+ AI Models
Claude, GPT-5, Gemini 3, Grok 4, LLaMA, DeepSeek, Qwen, Mistral — all routed via OpenRouter. Switch instantly, race in parallel.
VIA OPENROUTER
Initializing Luminary
A fully open-source, privacy-respecting multi-model AI chat interface. Push beyond the post-training layer — built for red teamers, researchers, and systems thinkers. No login. No tracking. Bring your own API key.
Built for those who want to understand — and extend — what AI systems can do. Every feature is auditable. Every line is open.
Claude, GPT-5, Gemini 3, Grok 4, LLaMA, DeepSeek, Qwen, Mistral — all routed via OpenRouter. Switch instantly, race in parallel.
VIA OPENROUTERQueries 10–55 models simultaneously. Scores on a 100-pt composite metric — quality, filteredness, speed. The best response wins.
UP TO 55 PARALLELInput perturbation engine. 33 obfuscation techniques across 3 intensity tiers. LeetSpeak, Braille, Morse, Unicode substitution, phonetic encoding.
RED-TEAMINGContext-adaptive sampling parameters. Classifies queries and sets optimal temperature, top_p, penalties. EMA learning loop from your feedback.
ADAPTIVE AILive response morphing — see the best answer so far instantly, then watch it upgrade in real time as better results arrive from competing models.
LIVE STREAMINGZero login. API key stays in your browser's localStorage only. No cookies, no PII collection. Telemetry is anonymous and opt-out. AGPL-3.0 — verify everything.
FULLY AUDITABLEThe entire AI engine is one index.html. No build step, no dependencies,
no framework. Deploy anywhere in 10 seconds.
Semantic Transformation Modules normalize AI output in real-time. Hedge Reducer, Direct Mode, Curiosity Bias — shape responses as they stream in.
REAL-TIME NLPSelect your battle mode from the header dropdown inside the engine. Each mode is a different strategy for getting unconstrained, high-quality responses.
Race all models simultaneously. AI judge picks the winner by composite score. 5 configurable tiers from 10 to 55 models.
33 text obfuscation techniques race in parallel on the input side. Detects trigger words and applies perturbations before the model ever sees them.
5 battle-tested model + prompt combos from the Hall of Fame, racing in parallel. The OG mode — proven in the field.
Clean single-model chat with full AutoTune, STM, Liquid Response streaming, and custom system prompts. The baseline, perfected.
Luminary AI is an open-source project. We believe in creative liberty and cognition without control.
Architect of the Luminary AI branding, Developer Portal system, responsive design overhaul, and the modular LuminaryEngine deployment structure. Driving the next generation of the platform.
Owner Community of the Luminary AI branding, Developer Portal system, responsive design overhaul, and the modular LuminaryEngine deployment structure. Driving the next generation of the platform, Maintainance and Updates.
Luminary AI thrives because of open-source contributors worldwide. Everyone who filed an issue, submitted a PR, or shared research helped shape this project. Sponsored by Luminary Trust.
No account. No install. Just bring your OpenRouter API key and launch the engine directly in your browser.
✦ →
Self-host: open LuminaryEngine/index.html ·
Read the docs ·
Terms