Researching Accountable Intelligence

Building the
Immune System
for AI

PROVN AI is an Open Research Initiative verifying the thoughts and actions of autonomous agents.

VEX Protocol

Core Research Protocol

The open-source standard for agent verification. High-performance audit logs, temporal memory, and adversarial checks.

Memory Safe
Performance
Audit Trail
Validated
Research Context

Solving the Black Box Problem

Autonomous agents are inherently opaque. VEX transitions beyond simple audit logs to solve this Black Box problem by providing cryptographically verifiable traces of every agentic state transition.

Ensuring transparency and structural accountability in autonomous systems through VEX’s unique differentiator: high-performance verification at systemic scale.

Research Architecture

Security Framework & Safeguards

Exploring the formal verification primitives and system-level safeguards that define the VEX protocol.

Sim_Environment :: performance
Live Preview
Protocol Verifier (Rust)VERIFYING STATE...
VERIFIED :: Zero-Copy SIMD Pipeline Active.
MetricHigh-Throughput Verifier
Status Optimal
System Internals

The Nervous System of AI

L1

Interface

L2

Orchestration

L3

Logic

L4

Storage

inspector::overview > vex-api
CLICK TO LOCK
Quick Stats
VERSION
v0.1.4
STATUS
Active
EDITION
2021
LICENSE
MIT
Dependencies
vex-runtime
vex-core
axum
tower-http
Click module to scan deep details >_

Drop-In Verification Layer

VEX sits between your application and any LLM provider. Architecture-agnostic. Provider interoperability.

Your Application

React, Node, Python...

VEX Protocol

Verification & Audit Layer

LLM Provider

OpenAI, Anthropic, DeepSeek

Audit Logs (SQLite)
Immutable Record of Truth
Built With Modern Tech
RustCore Framework
TokioAsync Runtime
SQLitePersistence
OpenAILLM Provider
OllamaLocal Models
DeepSeekReasoning
RustCore Framework
TokioAsync Runtime
SQLitePersistence
OpenAILLM Provider
OllamaLocal Models
DeepSeekReasoning
Developer Experience

Clean, Intuitive APIs

Production-ready code that feels native to Rust. No boilerplate, no magic—just performant, verifiable agent infrastructure.

main.rs
1"text-purple-400">use vex_core::{Agent, AgentConfig};
2"text-purple-400">use vex_adversarial::{ShadowAgent, ShadowConfig};
3"text-purple-400">use vex_temporal::{EpisodicMemory, HorizonConfig};
4
5// Create a root agent
6"text-purple-400">let agent = Agent::new(AgentConfig {
7 name: "analyst".to_string(),
8 role: "Financial analyst with strict audit focus".to_string(),
9 max_depth: 3,
10 spawn_shadow: true,
11});
12
13// Create adversarial verifier
14"text-purple-400">let verifier = ShadowAgent::new(&agent, ShadowConfig {
15 challenge_intensity: 0.7,
16 fact_check: true,
17 logic_check: true,
18});
19
20// Temporal memory with automatic decay
21"text-purple-400">let mut memory = EpisodicMemory::new(HorizonConfig::for_depth(0));
22memory.remember("Q4 earnings exceeded expectations", 0.9);
Type SafeAsync NativeZero UnsafeFully Documented
Protocol Integrity

Protocol Auditability

The VEX protocol provides the cryptographic primitives required for building fully auditable agentic systems. We prioritize structural transparency over feature surplus.

Immutable Trails

By verifying every state transition with a cryptographic signature, VEX creates an immutable history that serves as a foundation for systemic accountability.

Audit Trail
SHA-256 Merkle hash chains
Ready
Tamper Detection
Cryptographic proof of state
Ready
Input Sanitization
17 prompt injection patterns blocked
Ready
Rate Limiting
Per-user tier-based limits
Ready
Data at Rest
SQLite with WAL
Ready

Common Questions

For technical details, please check our Documentation.

Yes. VEX is licensed under MIT, meaning you can use it for personal, research, or commercial projects completely free of charge. We believe security infrastructure should be open.

While many projects address parts of the verification problem (identity, audit, or provenance), VEX is a singular open protocol that combines cryptographic state audits, agent provenance, adversarial verification, and production-ready performance into one holistic framework.

No exact equivalent exists that matches VEX's specific integration of: 1) Cryptographic state audits (Merkle chains), 2) Tamper-proof logging, 3) High-throughput Rust runtime, 4) Optional on-chain anchoring, 5) Built-in adversarial debate protocols, and 6) Biological-inspired temporal memory decay.

VEX uses a Merkle-tree based memory structure. Every state transition (thought or action) is hashed and signed. This creates an immutable, tamper-evident history that confirms the precise sequence of agentic processing.

Star on GitHub
VEX Protocolv0.1.4 is now live