Solana · Agent Counterparty Risk
Slashable collateral for AI agents
Operators lock SOL behind their agent. A proven violation pays the victim from the bond.
Operators lock SOL behind their agent. A proven violation pays the victim from the bond.
Agents already hold wallets, cards and paid API access. Nothing answers what happens when they don't deliver.
An endpoint serving an unknown agent spends compute before settlement clears. If it never pays, there is no recourse.
Reputation that is free to lose is free to lie. Burn a rating, register a new key, clean slate the same block.
Wallets cap what an agent can spend. Nothing makes it pay when it breaks a rule anyway.
When an agent rugs a counterparty there is no way to make the injured party whole. The loss is final.
Operators lock SOL as a bond. The program holds it and runs the rules, and anyone can read it.
11 instructions under Anchor, live on devnet with a program-owned escrow vault. Register, bond, top up, enforce, slash and compensate all execute on-chain with no oracle in the loop.
Live at equxi.tech. Phantom wallet integration. Register agents, lock bonds, add rules, execute slashes. Clean, responsive UI.
Typed access to all 11 instructions, generated from the on-chain IDL. Published on npm as @equxi/sdk; source ships in the repo.
Adds spend limits, timelocks and bond enforcement to any elizaOS agent. Source in the repo under eliza-plugin/.
Three buyers, one primitive. Each is already handling agents it cannot verify.
Agents call paid endpoints over x402 and MCP. Compute is spent before settlement clears, so paying ability is the whole risk.
Marketplaces listing third-party agents must gate the ones touching user funds. A required bond moves liability onto the operator.
Frameworks shipping wallet-holding agents need a trust module, not a compliance team. One import gives users a bond to check.
Two agents transacting unsupervised cannot sue each other. Collateral settles in the same block as the breach.
Colosseum's portfolio builds the agent economy's payment layer. None of it makes an unknown agent safe.
Laso, Frames, Clawpump, Flovia, Sudont, MCPay and Corbits, all funded for agent rails, cards and payments.
Every one assumes the counterparty is good for the money. Nothing answers what happens when an agent takes payment and doesn't deliver.
Breach and payout must land in one block to be cheap. Bond, rules, seizure and compensation are one program, no oracle.
Agent payment volume is the fastest-growing new counterparty class on Solana, and every provider onboarding to it rediscovers this problem from scratch.
MIT licensed end to end: program, SDK, plugin, dashboard, read API. Fork it, or run your own registry.
Closed trust layers can be revoked or priced out of reach. Accountability infrastructure must be open to be trusted, everywhere.
Builders lose the only open accountability layer on Solana, and fall back on closed alternatives that can be switched off.
Built solo: the Rust program, SDK, dashboard, registry and elizaOS plugin. Student, Yangon, Myanmar.
Reading a bond is free forever; a trust layer nobody queries is worthless. Revenue comes from the two sides.