Document ID: AG-001
Whitepaper Abstract
The Agentus Protocol introduces a decentralized verification layer designed to solve the critical "Trust Gap" in autonomous AI workflows. By utilizing a peer-to-peer network of validator nodes, Agentus provides verifiable proof that LLM outputs satisfy specific deterministic constraints.
Where C(y) represents the cryptographic constraint set applied to the model's probabilistic output. Our goal is to ensure that no autonomous agent executes a high-stakes transaction without cryptographic verification.
Vision & Philosophy
In an era where machines represent the majority of internet traffic, "human-readable" trust is no longer sufficient. We are building the infrastructure for Machine-to-Machine commerce.
Immutable Truth
Every model decision is hashed and stored on-chain to prevent historical revisionism.
AI Sovereignty
Decoupling AI logic from centralized providers through decentralized compute validation.
Proof of Validation (PoV)
Agentus utilizes a novel consensus mechanism called Proof of Validation. Unlike Proof of Work, PoV requires nodes to execute a specific "Verification Function" against a submitted LLM trace.
Validators are selected based on their hardware's TEE (Trusted Execution Environment) capabilities, ensuring that the validation logic itself has not been tampered with.
Node Architecture
The Agentus network is comprised of three primary node tiers:
-
1
Ingress Proxies
Responsible for sanitizing and routing incoming model requests.
-
2
Validation Sentries
High-performance nodes running ZK-verification circuits.
Slashing & Security
To maintain network integrity, validators must stake AGT tokens. Malicious actors providing false proofs are penalized via programmatic slashing.
Protocol Violation
Submitting invalid signatures results in an immediate 15% stake forfeiture and permanent node ban.