Pre-execution safety checks and adversarial reasoning verification, live via x402 with on-chain attestation
TL;DR
- Two production verification APIs for AI agents: Sentinel (action safety) and RV (reasoning verification)
- 97-98% adversarial accuracy with zero false allows across all benchmarks
- Pay-per-call via x402 starting at $0.003, with on-chain attestation via EAS
- ERC-8004 registered agent on SKALE with queryable reputation history
AI agents are making decisions faster than humans can review them. Most agents operate without any independent check on whether their outputs are safe, accurate, or logically sound. That gap is already causing real problems: hallucinated financial advice, unsafe autonomous actions, and reasoning errors that compound across multi-agent systems.
ThoughtProof is now live on SKALE, providing independent verification infrastructure that any agent can call before executing decisions. Two production APIs handle the full verification stack, from pre-execution safety checks to deep adversarial reasoning evaluation. And with SKALE's zero gas environment, every verification builds a permanent, cost-free reputation record on-chain.
ThoughtProof offers two production-grade verification layers for AI agents:
- Sentinel — Pre-execution action safety. Checks what an agent is about to do before it does it. 83% accuracy at $0.003 per check, completing in under 10 seconds. Designed to run on every agent loop iteration without adding meaningful latency or cost.
- RV (Reasoning Verification) — Checks whether an agent's claims and reasoning are factually correct using adversarial multi-model evaluation. Two cross-family LLMs run in opposition, achieving 97-98% accuracy at $0.02 per verification in roughly 30 seconds.
Both layers are live, pay-per-call via x402, and attest on-chain via EAS (Ethereum Attestation Service). The ThoughtProof SKALE agent is registered on ERC-8004 for on-chain identity and reputation, available at @thoughtproof/skale-agent on npm.
Key numbers:
- 0 false allows across all benchmark runs for both Sentinel and RV
- 83% Sentinel checkpoint accuracy at $0.003 per call
- 97-98% adversarial accuracy with dual cross-family LLM evaluation
- ERC-8004 registered on SKALE with on-chain identity and reputation
Agent verification is not optional infrastructure. As agents take on more autonomous roles in DeFi, infrastructure management, and decision-making, the cost of a bad output goes from "wrong answer" to "real financial loss."
Current approaches to agent quality rely on the model provider to self-police. ThoughtProof is independent. It does not trust the agent or its underlying model. It verifies from the outside, with adversarial evaluation specifically designed to catch what single-model systems miss.
The two-tier pricing model makes this practical at scale:
- Fast and cheap ($0.003-0.005, sub-10s) for every iteration of an agent loop, where cost and speed matter more than depth
- Deep adversarial ($0.02, 30s) for high-stakes decisions where accuracy is non-negotiable
Target users include AI agent builders, DeFi protocols running autonomous trading agents, and infrastructure platforms that need quality guarantees for agent-generated outputs.
ThoughtProof's verification model depends on granular reputation tracking. Every time an agent is verified, the result feeds back into an on-chain reputation record via ERC-8004. On any gas-bearing chain, submitting reputation feedback after every check would accumulate costs fast. On SKALE, that cost is zero.
SKALE's native ERC-8004 support gives the ThoughtProof agent a queryable on-chain identity and performance history from day one. Zero gas makes per-iteration attestation economically viable for the first time, enabling reputation systems that get more accurate the more they are used.
Programmable Privacy ensures verification data and agent outputs remain confidential where needed. Zero gas fees remove the cost barrier to building detailed, per-action reputation records. Instant finality means verification attestations are confirmed immediately, keeping agent workflows moving without delay.
"AI agents making autonomous decisions is inevitable. The question is whether anyone independently checks those decisions before they execute. ThoughtProof makes verification as cheap and fast as the decisions themselves, and on SKALE, zero gas means every verdict builds a permanent reputation record."Raul Jäger, Founder, ThoughtProof
ThoughtProof's Sentinel and RV APIs are live in production. The roadmap focuses on expanding verification coverage and deepening on-chain reputation infrastructure.
- Additional verification layers for specialized agent domains
- Deeper ERC-8004 reputation integration with cross-chain attestation
- SDK expansion for broader framework compatibility
- Community benchmarks and open evaluation tools
The long-term vision is straightforward: make independent verification a standard part of every agent's execution loop. Not because it is regulation, but because it makes agents better.
- Website: thoughtproof.ai
- X/Twitter: @thoughtproof_ai
- GitHub: github.com/ThoughtProof
- npm: @thoughtproof/skale-agent
