Skip to main content
QWED v4.0.0 Sentinel Edition is now live — Agentic Security Guards, Process Determinism, and 147 commits of hardening. See what’s new →

​
What is QWED?

QWED (Query With Evidence & Determinism) is a model-agnostic trust boundary for AI systems.
“Don’t fix the liar. Verify the lie.” — QWED verifies outputs, processes, and tool interactions before they enter production. It doesn’t reduce hallucinations — it makes them irrelevant.
Where QWED fits first: Verify AI-generated math, logic, SQL, code, and schemas before execution. Protect RAG pipelines against prompt injection. Inspect agent tool calls before they reach external systems. Enforce deterministic process steps in high-stakes workflows.

​
Quick Start

pip install qwed

# Verify math
qwed verify "Is 2+2=5?"
# → ❌ CORRECTED: The answer is 4, not 5.

# Verify logic
qwed verify-logic "(AND (GT x 5) (LT y 10))"
# → ✅ SAT: {x=6, y=9}

​
Why QWED?

LLMs hallucinate math

QWED uses SymPy for symbolic verification — algebraic proof, not pattern matching.

LLMs break logic

Z3 SAT solver provides formal satisfiability checks with model generation.

LLMs generate unsafe code

AST analysis + pattern detection catches vulnerabilities before execution.

LLMs produce SQL injection

Query parsing + schema validation catches injections and malformed queries.

​
11 Verification Engines

Math

Symbolic algebra with SymPy

Logic

SAT/SMT solving with Z3

Code

AST security analysis

SQL

Query validation & injection detection

Fact

NLI-based fact checking

Stats

Statistical claim verification

Reasoning

Chain-of-Thought verification

Image

Visual content verification

Graph

Knowledge graph fact checking

Schema

JSON/API schema validation

Taint

Data flow taint analysis

​
Model Agnostic = Your Choice

QWED works with ANY LLM — OpenAI, Anthropic, Gemini, Llama (via Ollama), or any local model. Same verification quality, your choice of cost.
1

Choose your LLM

Use any provider — cloud or local. QWED stays neutral. No vendor lock-in.
2

Send queries through QWED

QWED routes LLM responses through the appropriate symbolic engine for verification.
3

Get verified results

Receive deterministic, provably-correct responses with optional cryptographic attestations.

​
Ecosystem & SDKs

Python SDK

pip install qwed — Full-featured SDK with CLI.

TypeScript SDK

npm install @qwed-ai/sdk — Browser and Node.js support.

Go SDK

Lightweight Go module for backend services.

Rust SDK

cargo add qwed — Zero-cost abstractions.

LangChain

Drop-in verification guard

LlamaIndex

Query engine integration

CrewAI

Multi-agent verification

​
🆕 What’s New in v4.0.0: Sentinel Edition

The v4.0.0 Sentinel Release introduces Agentic Security Guards, Process Determinism, and critical security hardening. 147 commits — the largest update in QWED history.

Agentic Security Guards (Phase 17)

  • RAGGuard — Detects prompt injection and data poisoning in RAG pipelines.
  • ExfiltrationGuard — Prevents data exfiltration through agent tool calls.
  • MCP Poison Guard — Detects poisoned MCP tool definitions.
  • SovereigntyGuard — Data residency and local routing enforcement.
  • ToxicFlowGuard — Stateful toxic tool-chaining detection.
  • SelfInitiatedCoTGuard — Reasoning integrity verification.
  • ProcessVerifier — IRAC/milestone-based verification with decimal scoring, budget-aware timeouts, and structured compliance reporting.
  • Replaced all eval() with AST-compiled execution.
  • Patched sandbox escape, SymPy injection, and protocol bypass vulnerabilities.
  • Resolved CVE-2026-24049 (Critical), 19 Snyk findings, and CodeQL alerts.

Full Changelog

See the complete release history including v3.0.1 Ironclad and v2.4.1 Reasoning Engine.

​
Next Steps

Installation

Install QWED via pip, Docker, or from source.

Quick Start

Verify your first LLM output in 5 minutes.

SDK Documentation

Choose your language — Python, TypeScript, Go, or Rust.

Specifications

Read the formal protocol specification.