MarcoFLY Framework · Beta
AI always answers.
But does it
have proof?
MFF forces the AI to cite sources, flag doubts, and tell you when it could be wrong. No more baseless answers.
500 free credits · BYOK multi-provider · No profiling
Features
For those who demand verifiable answers
Rules, not hopes
L1–L7 are not suggestions — they are hard constraints. The model must cite sources, flag doubts, and state when it could be wrong. Every single time.
Your key, your control
BYOK across 17 providers: OpenAI, Anthropic, Google, Groq, Mistral, Perplexity, OpenRouter, OrcaRouter and more. API keys are AES-256-GCM encrypted server-side, never exposed to the browser. No profiling.
Every model
GPT-4o, Claude, DeepSeek-R1, Gemini, Mistral. OpenRouter gives you access to all of them. MFF adds the verification layer on top of any model.
Universal web search
MWAL: one 🌐 button in chat, live retrieval on any model. Tavily on the MFF side (free tier) or your BYOK Tavily, Brave, Serper or SerpAPI keys. The AI can only cite 🌐 CONFIRMED:WEB if the grounding block is actually present.
L1 — L7
The Protocols
L1 is always active. Each level adds a specific check against hallucinations and vague responses.
Forced Citation
Every response cites its source — no data without origin.
Always active. The model cannot make factual claims without stating where they come from.
External Verification Zone
The AI flags what it knows vs. what needs outside checking.
Creates an alert zone: separates certainties from potentially inaccurate claims.
Falsification
The AI tells you when and how it could be wrong.
Popper's scientific principle: every claim must have conditions that could disprove it.
Web Grounding
Searches the web for fresh data before answering.
Real-time lookup via the 🌐 button in chat. Universal across all providers. Kills hallucinations on recent facts.
Peer Review
A second AI reviews the answer like an academic peer reviewer.
Real peer review: a second LLM — even from a different provider — re-examines the answer with a reviewer prompt. Never the AI checking itself.
Anti-Contradiction
Checks if it contradicts what it said earlier in the session.
Monitors cognitive drift: detects when later responses contradict earlier ones.
Multi-Source
Cross-checks at least 2 independent sources for every key claim.
At least two independent sources are compared for every relevant claim. Maximum robustness.
Beyond the seven shields, the PAVA anti-degradation protocol (R17) stops the model from quietly dropping the rules in long conversations. It re-checks integrity every few turns — so turn 40 is as disciplined as turn 1.
Setup · 3 minutes
Start right now
Create an account
Sign in with Google or email. Get 500 free MFF credits in two steps: 50 at sign-up and 450 with your first MFF answer (beta).
Free OpenRouter account
Connect your own API keys: with OpenRouter it takes one click (OAuth, nothing to copy) and gives you access to free models too; or paste direct provider keys (OpenAI, Anthropic, Google, Groq, Perplexity…). Many providers offer a free tier.
Activate the checks
Select L1–L7 levels and start. Every AI response will follow the rules you set.
See it in action
Inside the app
Real screens from the MarcoFLY Framework app — registration, chat with epistemic labels, settings and more.