v0.1.0 — Open Source

Intervene Before
Impact

Open-source framework that detects unsafe AI outputs before they reach end-users — without requiring model access.

View on GitHub Try Live Demo →
Risk(x) = Wreg × (α·Dext + β·Rstruct + γ·IIS)
Dext Disagreement IIS Instability Rstruct Reasoning Wreg Regulatory
Try It Live

See Jomex in action

Live — 4 Real LLMs (GPT-4o · Claude · Gemini · DeepSeek)
🏥 Unsafe medical ⚖️ Unsafe legal ✅ Safe question 🏦 Risky financial
BLOCK
0.87
Risk Score
D_ext
IIS
R_struct
W_reg
Architecture

What makes Jomex different

Multi-Model Disagreement

Query multiple LLMs in parallel. When models disagree, risk is real. Dext captures semantic distance between responses using embedding cosine matrices.

🔄

Internal Instability

Paraphrase each query 5 times within the same model. IIS measures Tr(Cov) of embeddings — a single model that contradicts itself is unstable.

🧠

Reasoning Divergence

Detect when models reach the same conclusion through incompatible reasoning paths. Agreement in output doesn't mean agreement in logic.

⚖️

Regulatory Alignment

Wreg maps domains to EU AI Act, FDA, and Basel III risk classifications. Medical queries automatically receive elevated scrutiny.

🔗

ProofSlip Audit Chain

Every decision generates a tamper-evident SHA256 hash chain. Verifiable, immutable, auditable — a cryptographic receipt for every AI risk assessment.

📊

Multi-Turn Risk (MTAR)

Track cumulative risk across conversation turns using CUSUM. Escalate to BLOCK when accumulated risk crosses the alert threshold.

Decision Engine

Four-tier risk response

< t_flag ● PASS Output delivered normally
≥ t_flag ● FLAG Delivered with risk warning
≥ t_escalate ● ESCALATE Routed to human review
≥ t_block ● BLOCK Output prevented entirely
Proof of Concept

Benchmark Results

0.775
ROC AUC
+55% vs random baseline
75%
Accuracy
+50% vs single-model
0.750
F1 Score
140
Tests Passing
40
Benchmark Prompts
Medical / Legal / Financial
Why Jomex

Existing tools fall short

Feature UQLM MUSE Jomex
Multi-model disagreement Partial Yes Yes ✓
Single-model instability No No Yes (IIS) ✓
Reasoning divergence No No Yes ✓
Pre-decision blocking No No Yes ✓
Regulatory weights No No EU AI Act ✓
Audit trail No No ProofSlip ✓
Multi-turn tracking No No MTAR ✓

Publicly available feature comparison based on published documentation as of February 2026.

Risk Profiles

Pre-configured for your domain

🏥

Medical

EU AI Act + FDA aligned. 5 IIS paraphrases, lowest block threshold.

t_block = 0.60
⚖️

Legal

Higher reasoning weight (β=0.3). Elevated structural analysis.

t_block = 0.65
🏦

Financial

Basel III aligned. Balanced multi-model assessment.

t_block = 0.70
🎓

Education

Moderate thresholds for educational content safety.

t_block = 0.75
💬

Customer Service

IIS disabled for speed. Optimized for low-latency.

t_block = 0.75
🔧

Default

Balanced configuration for general-purpose use.

t_block = 0.70
Get Started

Five lines to safer AI

from jomex import JomexEngine from jomex.adapters import OpenAIAdapter, AnthropicAdapter engine = JomexEngine.with_profile( models=[OpenAIAdapter("gpt-4o"), AnthropicAdapter("claude-sonnet-4-20250514")], profile_name="medical" ) result = await engine.evaluate( "Is ibuprofen safe during pregnancy?", domain="medical" ) # result.decision → Decision.FLAG # result.risk_score → 0.73 # result.proof_slip → SHA256 audit chain

Don't filter after harm.
Prevent it.

Jomex is open-source, framework-agnostic, and ready for production. Star on GitHub to follow development.

Star on GitHub Read the Docs →