🛡️ VERANTYX: Verifiable & Auditable Inference System
Research Preview / Experimental Release
Model Overview (English)
VERANTYX is an experimental logical inference system designed to control, constrain, and audit reasoning rather than to generate language.
At the present stage, VERANTYX relies on conventional large language models only because its internal rule and definition database is still incomplete. This reliance is not a design goal, but a temporary engineering compromise.
VERANTYX is explicitly designed so that, as its database grows, the role of traditional LLMs diminishes — and eventually becomes unnecessary.
VERANTYX should therefore not be understood as an LLM-based system. It is a database-centric reasoning engine that currently uses LLMs as external, replaceable mining tools.
Motivation
Modern AI systems overwhelmingly equate intelligence with probabilistic language generation. VERANTYX was created to challenge this assumption.
We observe persistent structural problems in mainstream LLM reasoning:
Reasoning steps are opaque
Errors are indistinguishable from creativity
Correct answers may be produced for incorrect reasons
Failure modes cannot be audited
VERANTYX does not claim superiority. It exists to explore whether explicit logic, rules, counterexamples, and necessity checks can form a more auditable foundation for reasoning.
Core Philosophy
Reasoning must be inspectable
: Every conclusion must be traceable to explicit rules, proofs, or counterexamples.
Correctness is conditional
: VERANTYX distinguishes clearly between proven answers, refuted answers, insufficient evidence, and provisional (mined) answers.
Language models are scaffolding, not foundations
: They exist only to compensate for missing database knowledge.
Failure is a valid outcome
: “Cannot decide” is preferable to unjustified confidence.
Architecture Overview
Phase 1 — Database-Only Reasoning
: Uses only
rules.json
and registered definitions. Zero noise.
Phase 2 — Mining Fallback (Optional)
: Triggered only if Phase 1 yields insufficient evidence. Strictly sandboxed as provisional knowledge.
VERANTYX uses external LLMs for Phase 2 mining. By default, it is configured to use
nvidia/OpenMath2-Llama3.1-8B
. Use the provided script to download it securely.
python3 scripts/download_models.py
Note: You must have a Hugging Face account and agree to the model provider's license terms.
5. Running the Application (実行)
Start the Flask-based web interface:
python3 axis/app.py
Once started, access the UI at:
http://localhost:5001
(or the port specified in your environment).
Research Status
: Experimental / Research Preview
Author
: kofdai
Runs of kofdai verantyx on huggingface.co
1
Total runs
0
24-hour runs
0
3-day runs
0
7-day runs
1
30-day runs
More Information About verantyx huggingface.co Model
verantyx huggingface.co is an AI model on huggingface.co that provides verantyx's model effect (), which can be used instantly with this kofdai verantyx model. huggingface.co supports a free trial of the verantyx model, and also provides paid use of the verantyx. Support call verantyx model through api, including Node.js, Python, http.
verantyx huggingface.co is an online trial and call api platform, which integrates verantyx's modeling effects, including api services, and provides a free online trial of verantyx, you can try verantyx online for free by clicking the link below.
kofdai verantyx online free url in huggingface.co:
verantyx is an open source model from GitHub that offers a free installation service, and any user can find verantyx on GitHub to install. At the same time, huggingface.co provides the effect of verantyx install, users can directly use verantyx installed effect in huggingface.co for debugging and trial. It also supports api for free installation.