APEX v1.5.3 EL v2.1
Activation Generator
Configure the operational domain parameters to generate your custom MFF-EX universal prompt.
New here? Read the guide first →
Choose your path
How do you want to activate MFF?
RECOMMENDED
Use the PWA
Activate L1–L7, choose your AI model, manage sessions and read verified answers — all in one app. More powerful, faster, more precise than the manual prompt.
Discover the PWA →MANUAL GENERATOR
Manual Activation
Configure the parameters below and generate your custom MFF-EX prompt. Copy, paste into any AI, and the protocol kicks in. Free, no account required.
Continue below ↓Step 1 — Operational Data
Session Config
Certainty Standards
Step 2 — Active Shields
Modular Levels
L1 — Forced Citation FIXED
No 🟢 CERTAIN without a source in appendix.
COMPATIBLE AI ECOSYSTEM
Choose where to activate the Framework from this selection of platforms
⚠️ DISCLAIMER AND TERMS OF USE
By generating and using the MFF-EX prompt, the user declares to have understood and accepted the following binding conditions:
- 1. Liability Disclaimer. MarcoFLY Framework is an experimental engineering protocol under active development. The author disclaims any liability for damages, malfunctions, data loss, or erroneous decisions derived from its use. The framework is provided as-is, without explicit or implicit warranties of fitness for a specific purpose.
- 2. Operating Costs and Rate Limits. Applying the framework significantly increases token consumption in input and output. Possible consequences: premature exhaustion of free plan limits, early rate limit hits, unexpected costs on pay-per-use platforms (API). The user is responsible for monitoring their subscription plan.
- 3. Latency Degradation. The cognitive load imposed on the model to enforce epistemic rigor, certainty labeling, and structured formatting physiologically increases response generation times. The framework prioritizes verified quality over speed: the trade-off is deliberate.
- 4. AI Laziness (Hallucinations). Language models remain subject to systemic hallucinations even under protocol. The model may simulate verifications not actually performed, fabricate sources, degrade performance in prolonged contexts (Amnesia or context decay), or formally adhere to the protocol while violating it in content (Methodological Compliance). The protocol reduces, not eliminates, these risks.
- 5. Human-in-the-Loop Validation. The certainty labels — including 🟢 CERTAIN — are probabilistic estimates from the model, not absolute truths. Final validation of information, operational decisions, and any consequent action rests solely with the human user.
- 6. Architectural Compatibility. Optimized exclusively for frontier-class LLMs (GPT-5, Claude 4.x, Gemini 2.5+, Mistral Large equivalents). Smaller, distilled, or quantized models may exhibit logical loops, loss of labeling, or protocol collapse after few iterations.
⚠️ Summary of the trade-off. The framework consumes more tokens and more time per response. In return, it provides structured outputs, labeled for reliability and verified at the root — reducing follow-up prompts and manual corrections. You consume more, but you get more, and faster. ⚠️