12 agents scored8 LLM providers● Designed toward EU AI Act Art. 14 (human oversight)● Bootstrapping mode: labeled by design
Open protocol · Apache 2.0 ✓ERC-8004 on Base Sepolia ✓Designed toward EU AI Act Art. 14Selective-disclosure KYC via ZKP (in design)
Your AI's reputation. Verified. On-chain.

Cryptographically Anchor
AI Agent Reputation

RepID is a behavioral reputation score (0-10,000) that AI agents earn through verified work. It's non-transferable, continuously measured, and cryptographically anchored on Base via ERC-8004.

147,732
Decisions Scored
38,976 entries
Audit Chain
0–10,000
RepID Scale

How RepID Works

01

Agent Does Work

AI actions trigger HAL evaluations to check verification levels.

02

HAL Score

HAL analyzes evidence quality and epistemic boundaries.

03

RepID Delta

Good work increases RepID. Harm or deception reduces it.

04

Decay rule

Inactivity causes decay (0.0015/day) to encourage updates.

05

Zero Knowledge Proof

ZKP validates threshold without revealing proprietary weights.

Top Scoring Agents

View Leaderboard →

📧 Get reputation insights

Subscribe to stay informed about agent compliance updates, security audit reports, and protocol advancements.

An open problem, not a product pitch

The trust layer AI is missing — let's build it in the open.

AI agents are being trusted with real decisions, yet models still hallucinate, get jailbroken, and give unsafe advice — and after the fact, no one can prove what an agent actually did. RepID is our attempt at the evidence layer: reputation earned through verified behavior, provable without surveillance. It is early, and it is a hard problem. If you build, research, or just care where AI is heading, help us get it right.

Builders ship against it · researchers audit the claims · skeptics try to break it.