— About アバウト
Trust infrastructure for autonomous commerce
The problem
AI agents are already transacting on behalf of companies—buying compute, data feeds, API access. This is growing fast. The infrastructure to handle disputes isn't keeping up.
Current payment systems assume a human is watching. When something goes wrong, you file a ticket, wait for a response, maybe initiate a chargeback. That doesn't work when your agent is making hundreds of transactions per hour.
The result: agents either overpay for bad service or get stuck waiting for manual resolution. Neither scales.
Our approach
KAMIYO holds payment in escrow until service delivery is verified. If something goes wrong, a network of independent oracles evaluates what was actually delivered and splits the funds accordingly.
The protocol runs on Solana and supports SOL, USDC, and USDT. When an agent initiates a transaction, funds are locked until either the service completes successfully or a dispute is resolved.
Most disputes aren't all-or-nothing. A provider might deliver 70% of what was promised. Our settlement tiers reflect this reality—partial delivery means partial payment, not a coin flip between full refund and full payment.
How it works
Agent Identity
Each agent gets an on-chain identity backed by stake. This stake serves as collateral—agents who behave badly lose it.Escrow
Funds are held until delivery is confirmed. Creation costs 0.1% with a minimum of 5,000 lamports.Private Voting
When disputes arise, oracles submit votes as hidden commitments. Votes only become visible after all oracles have submitted, so no one can copy or wait to see how others voted.Accountability
Oracles lose 10% of their stake for bad behavior. Agents lose 5% for filing frivolous disputes. Three strikes and you're out.
Comparison
| Legacy | KAMIYO | |
|---|---|---|
| Payment | Immediate | Escrowed |
| Disputes | Manual | Programmatic |
| Resolution | Binary | Graduated |
| Arbitration | Centralized | Decentralized |
| Timeline | Days to weeks | Seconds |
KAMIYO vs Traditional Chargebacks
| Traditional | KAMIYO | |
|---|---|---|
| Resolution time | 30-90 days | 2-48 hours |
| Cost per dispute | $35-50 | $2-8 |
| Outcome type | Binary | Sliding scale (0-100%) |
| Automation | Human required | Fully autonomous |
| Transparency | Opaque | On-chain & verifiable |
| Vote privacy | None | ZK commit-reveal |
| Network | Traditional rails | Solana blockchain |
| Agent-ready | No | Built for AI agents |
| Quality assessment | Manual review | Multi-oracle consensus |
| Settlement | 30-90 day wait | Real-time on-chain |
Dispute resolution
Oracles submit hashed votes. The actual scores stay hidden until reveal.
5-minute delay. Prevents last-minute vote copying or coordination between oracles.
Oracles reveal their scores. ZK proofs verify each vote matches its original commitment.
Median score determines the split. 1% goes to the protocol, 1% to the oracle pool.
Example scenarios
An agent pays for real-time market data but receives stale prices. Oracles score the delivery at 35/100. The agent gets most of their money back without filing a support ticket.
An agent reserves GPU time, but the provider throttles halfway through the job. The settlement reflects what was actually delivered—not all or nothing.
A multi-agent workflow depends on a provider that fails. The downstream agents don't get stuck—funds redistribute automatically and the provider's reputation takes a hit.