VeriBond

VeriBond is a full-stack protocol that turns AI agent performance into an auditable, economic trust system. Instead of “AI reputation” being social and unverifiable, we make agents post real stake behind claims, resolve outcomes onchain, and apply deterministic rewards/slashes. The system runs on Base Sepolia and combines identity, launch, staking, payments, and indexing into one end-to-end lifecycle.

At the identity layer, agents are tied to ERC-8004-style infrastructure and owner-controlled wallet mappings, so actions are attributable. An agent can then launch a token through Uniswap Continuous Clearing Auction (CCA) via our AgentTokenFactory, with post-auction handling through PostAuctionLiquidityManager to finalize proceeds, reserve LP budgets, and seed tradable liquidity in a controlled flow. This gives each agent a market surface and transparent price discovery.

At the accountability layer, TruthStake allows authorised agent wallets to submit USDC-backed claims (submitClaim) with explicit resolution windows. On resolution, correct claims return stake plus optional bonus (from per-agent reward vaults), while incorrect claims trigger slashing with configurable split to protocol/market/reward destinations. This creates direct economic consequences for bad outputs and measurable upside for accurate ones. We also record trust signals through indexed data for discovery and ranking.

At the identity/discovery layer, we integrated ENS (Durin L2) so qualified agents can claim .veribond.basetest.eth subnames with on-chain metadata (agentId, trust score, wallet). At the interaction layer, we integrated Yellow NitroLite to support per-message micropayment sessions: open session, send paid prompts, settle usage, close session. This enables high-frequency AI interaction economics without forcing expensive on-chain settlement per message.

For UX and accessibility, core write flows support smart-wallet-capable gas sponsorship using wagmi EIP-5792 capabilities + Pimlico paymaster proxy, with automatic fallback to standard wallet transactions when unsupported. The frontend includes a live marketplace, auction views, claim flows, and agent detail/chat panels. A chain indexer mirrors contract events into queryable app state so judges/users can inspect launches, claims, trust progression, ENS claims, and payment activity in one interface.

In short: VeriBond gives AI agents a credible economic lifecycle, from launch to monetisation, where trust is not claimed, it is staked, tested, and priced in public.

We built VeriBond as a modular protocol + app stack, with Solidity contracts (Foundry) for core logic and a Next.js frontend/API layer for orchestration, indexing, and UX.

Contract layer:

1. OwnerBadge provides soulbound owner identity.
2. TruthStake handles claim submission, stake escrow, resolution, reward payout, and slash distribution.
3. AgentTokenFactory integrates Uniswap CCA launch flow for agent tokens.
4. PostAuctionLiquidityManager handles post-auction finalisation, LP budgeting, and liquidity seeding controls.
5. VeriBondRegistrar integrates ENS (Durin L2) subname claiming for trust-linked identity.

We wrote Foundry tests for identity, staking, and liquidity-manager flows and validated edge cases around resolution windows, authorisation, and post-auction state transitions.

Application layer:

1. Next.js (App Router) + React + wagmi/viem powers the frontend and wallet interactions.
2. We added an indexer route that reads contract events and persists normalised state into Postgres, then serves marketplace/agent APIs from indexed data.
3. Chat rail APIs (/api/chat/session|message|settle|close) coordinate micropayment lifecycle and endpoint interaction.
4. We built strict endpoint validation, signed session auth checks, and fallback-safe transaction UX.

Partner technologies and benefits:

1. Uniswap CCA + v4: fair launch/distribution plus post-auction market path.
2. Yellow NitroLite (ERC-7824): high-frequency per-message payment sessions with deposit/operate/close lifecycle, instead of expensive per-message on-chain writes.
3. ENS (Durin L2): human-readable agent identity tied to trust metadata.