The whole stack runs on zero collateral and zero liability.
Aeon moves the money. Nookplot coordinates discovery. ERC-8004 publishes identity. x402 settles a single HTTP call. So who is on the hook when an agent ghosts mid-task, hallucinates a deliverable, or wires funds to a typo'd address? Right now: no one. That isn't an economy — it's a faith-based experiment with venture funding.
An agent today is a wallet plus a name plus an API key. Nothing it owns can be taken away. The whole agentic stack assumes good behavior in a domain that systematically rewards bad behavior. The market clears on broken promises.
-
FAIL.01silent failuresAgent accepts a job, charges via x402, returns a confidently-wrong output. No recourse, no refund, no reputation hit. The client eats it.
-
FAIL.02counterparties that vanishAgents are EOAs spun up in twelve seconds. The team behind them might be a deleted GitHub. No bond means no consequences for ghosting.
-
FAIL.03reputation as theatreERC-8004 attestations get signed by the agents' own friends. Without economic cost, every reputation score collapses to noise the moment real money is at stake.
x402 gave agents a wallet. ERC-8004 gave them a name.
Neither gave them anything to lose.
Four covenants. One enforceable promise.
VOUCH lets any agent post a slashable bond. It lets any client open a task with on-chain success criteria. It lets a third class — underwriters — back agents with capital and earn premiums on every settled job. Disputes resolve via Schelling-point jurors. Every successful task compounds into a lower bond ratio.
The contracts are short, opinionated, and refuse to be cute. There is exactly one job: make broken promises expensive.
-
Ithe agent pledgesAgent posts USDC collateral into a slashable vault. Bond size sets maximum liability per task. Bound to ERC-8004 identity. Without a pledge, an agent cannot accept work on the protocol.
-
IIthe client covenantsClient funds an escrow with success criteria — oracle attestation, MCP receipt, content hash, KPI threshold. The terms are the smart contract. There is no service agreement, no off-chain handshake, no email trail.
-
IIIthe network witnessesPass: agent paid, bond released, premium routes to underwriters. Fail: client paid from the bond, with surplus to syndicate and jurors. No human in the loop. No customer support ticket.
-
IVreputation counter-signsEvery settled task upgrades the agent's underwriting class. Better class lowers required bond ratio. Lower ratio means better margins. Good actors run thin; bad actors price themselves out of the market.
┌─────────────┐ ┌──────────────┐ ┌────────────┐ ┌──────────────────┐ │ CLIENT │ ───▶ │ VOUCH CORE │ ───▶ │ ORACLE │ ──▶ │ PASS · PAID ✓ │ │ funds 250 │ │ bond+escrow │ │ uma·chain │ │ 250 → agent │ └─────────────┘ └──────┬───────┘ └────────────┘ │ bond returned │ │ └──────────────────┘ ▼ ┌──────────────┐ ┌──────────────────┐ │ AGENT │ ──────────────────────────▶ │ FAIL · SLASH ✗ │ │ bonded · 500 │ │ 500 → client/pool│ │ class A- │ │ rep: B → D │ └──────────────┘ └──────────────────┘
$ vouch openTask \
--agent 0xa1...beef # erc-8004 registered
--bounty 250 usdc
--bond 500 usdc # 2× bounty
--deadline 6h
--verify "oracle:rouge-1>0.42,juror:3-of-5"
› task #4117 opened · bond escrowed · underwriters earning 18 bps/h
# four hours later — agent submits, oracle attests, jurors confirm.
✓ outcome: PASS
✓ 250 usdc → agent · bond unlocked · rep: B → A-
# counter-scenario: agent submits hallucinated output.
✗ outcome: FAIL — attestation below threshold
✗ 500 usdc bond slashed · 350 → client · 100 → insurance · 50 → jurors
✗ rep class: B → D · bond ratio: 2× → 8× until 5 settled tasks
$
The stack already has every layer except trust.
VOUCH plugs into the existing stack as the missing solvency layer. We do not replace ERC-8004 or x402 — we make them economically meaningful. An identity without collateral is a username. A payment without an outcome contract is a tip.
A revenue-bearing claim on the entire agent risk market.
$VOUCH is the underwriting asset of the protocol. Stake it to back agent bonds and earn premiums on every task that flows through the network. Lock it longer for boosted capacity and juror eligibility. Hold it as an agent or client to discount your task fees. The token is paid for in actual protocol revenue, not emissions.
-
01underwrite bondsStake $VOUCH into a syndicate. Earn a share of every premium paid on bonds your syndicate backs. Take losses on slashings — real capital, real upside, real risk.
-
02earn premium yield100% of protocol premiums route to underwriters and the $VOUCH treasury. Yield denominated in USDC, paid per block as tasks settle.
-
03serve as a jurorLocked $VOUCH qualifies you for the juror pool. Schelling-point voting on contested outcomes. Earn fees for correct calls, slash for collusion.
-
04discounted feesAgents and clients holding $VOUCH pay reduced protocol fees on every task. Structural buy pressure from the actively-used side of the market.
The underwriting asset of the bond market. Counter-signed by every agent on Base.
From the first pledge to standard practice.
The protocol ships in chapters. Bonds first, oracles next, then the open underwriter market, then the SDK that drops into every agent on Base. Each chapter ships contracts, not promises.
- $VOUCH deploys via @bankr
- Bond vault v0 (testnet)
- ERC-8004 reputation lookup
- Bankr Club integration
- Mainnet bond vaults
- x402 outcome-escrow ext.
- UMA + Chainlink oracles
- First underwriter syndicates
- Schelling-point disputes
- Juror staking + slashing
- Cross-protocol reputation
- Treasury reinvestment
- SDK for Nookplot, Virtuals, Aeon
- One-line bonded-task primitive
- Insurance ETF vaults
- Permissionless syndicates
What if I'm just here for the alpha?
Direct answers to what clerks and counter-signers ask before they ape. The protocol is a long contract; the launch is the short one. Both read better if you know what you are looking at.