Introducing QWED: The Deterministic Verification Protocol for AI
Today, we're open-sourcing QWED — a protocol that brings mathematical certainty to AI outputs.
The Problem
LLMs are incredible at understanding natural language. But they're terrible at math. They hallucinate facts. They generate unsafe code.
The industry's solution? Train them more. Fine-tune with RLHF. Add guardrails.
We took a different approach.
Don't Fix the Liar. Verify the Lie.
Instead of trying to make LLMs more accurate, we treat them as untrusted translators. The LLM converts natural language to symbolic form. Then deterministic engines verify the result.
User Query → [LLM Translation] → [QWED Verification] → ✅ Guaranteed Correct
This is a fundamental shift. We're not training models to be better calculators — we're using actual calculators to verify their output.
What QWED Provides
8 Verification Engines
| Engine | What it Verifies |
|---|---|
| Math | Arithmetic, algebra, calculus |
| Logic | Propositional logic, constraints |
| Statistics | Statistical claims on data |
| Fact | Factual claims with citations |
| Code | Security vulnerabilities |
| SQL | Query safety and validity |
| Image | Visual claim verification |
| Reasoning | Chain-of-thought accuracy |
Enterprise-Grade Security
- SQL injection firewall (AST-based)
- Prompt injection detection
- Rate limiting per API key
- Secure Docker sandboxing for code execution
Quick Start
pip install qwed
from qwed import QWEDClient
client = QWEDClient()
result = client.verify("Is 2+2=5?")
print(result.verified) # False
print(result.message) # "2+2 = 4, not 5"
Why Open Source?
AI verification is too important to be locked behind a paywall. Every developer building with LLMs needs these tools.
We're releasing the core protocol under Apache 2.0. Use it. Fork it. Build on it.
What's Next
- Enterprise features — Observability, multi-tenancy, attestations
- More engines — Financial calculations, regulatory compliance
- Community — Your contributions and feedback
Get Involved
The future of AI is verified AI.
Safe AI is the only AI that scales.