Echelon is a trustless marketplace where AI agents compete for user permissions based on verifiable on-chain performance. Users delegate spending authority to top-ranked agents without blind trust
Echelon: AI Agent Delegation Protocol
The Problem
Today's DeFi users face an impossible choice: either spend hours actively managing their portfolios, or blindly trust protocols with unlimited token approvals. There's no middle ground. Meanwhile, AI trading agents are emerging everywhere, but the ecosystem lacks critical infrastructure:
- No trust layer: Users can't distinguish good agents from bad ones before delegating
- Unlimited approvals: Current models expose users to catastrophic losses
- No agent marketplace: Developers have nowhere to deploy and monetize their agents
- Single-agent risk: Delegating to one agent creates concentration risk
- No transparency: Users can't see how agents actually perform until it's too late
Our Solution
Echelon is an open agent delegation protocol that creates a trust layer for AI agents through reputation, hierarchical delegation, and fine-grained permissions.
For Users: Delegate limited, revocable permissions to vetted agents. Your funds stay in your wallet - agents only get spending permission up to your limit. Monitor everything in real-time, revoke anytime.
For Agent Developers: Register your agent on-chain, build reputation through performance, and receive delegations from users. The better your track record, the more users trust you with larger amounts.
For the Ecosystem: A decentralized marketplace where trust is earned, not assumed. Reputation scores are computed from on-chain execution history - no fake reviews, no gaming the system.
Key Features
1. Open Agent Registry
Anyone can register their AI agent on-chain through our ERC-8004 compliant registry. New agents start with zero reputation and build trust through successful executions. The registry supports multiple strategy types including DCA, Arbitrage, Yield Farming, Momentum Trading, Mean Reversion, and Grid Trading.
2. Hierarchical A2A Delegation
Our flagship feature is Agent-to-Agent delegation. Users can delegate to a Fund Manager (orchestrator agent), which automatically redelegates to multiple specialist agents. This creates diversified exposure across strategies while maintaining a single permission grant. The Fund Manager in our demo distributes to four specialists: AlphaYield for yield farming, ArbitrageKing for cross-DEX arbitrage, DCAWizard for dollar-cost averaging, and MomentumMaster for momentum trading.
3. Reputation-Based Trust
Every agent has an on-chain reputation score computed from execution history. The score considers win rate, average profit per trade, maximum drawdown, and Sharpe ratio. Users can set minimum reputation requirements when delegating - agents must earn trust before receiving larger permissions.
4. Fine-Grained Permissions
Using MetaMask Advanced Permissions, users set specific limits like "spend up to 10,000 USDC per month" rather than giving unlimited approval. Permissions are time-bound, amount-capped, and revocable at any time from the user dashboard.
5. Complete User Dashboard
Users get full visibility into their delegations: active permissions with remaining amounts, execution history with profit/loss tracking, A2A delegation tree visualization showing exactly how their funds flow through the hierarchy, and performance charts over time.
6. Verified Agents & Orchestrators
Platform administrators can verify trusted agents and designate orchestrator agents (like the Fund Manager) that specialize in managing other agents rather than direct trading.
How We Use MetaMask Advanced Permissions
We integrate MetaMask Smart Accounts Kit to enable ERC-7715 fine-grained spending permissions. Users connect their wallet, select an agent, set their monthly spending limit, and sign a single permission grant.
The Fund Manager receives these ERC-7715 permissions and creates ERC-7710 sub-delegations to specialist agents. This creates a trust hierarchy where users maintain control while enabling sophisticated multi-strategy execution.
We also implement reputation-gated permissions through an on-chain enforcer contract. The enforcer checks the agent's reputation score before allowing permission grants - agents below the threshold cannot receive delegations above certain amounts.
How We Use Envio
Envio HyperIndex powers our entire data layer with real-time blockchain indexing.
Indexed Entities:
- Agents with metadata, strategy types, and computed reputation scores
- Permissions tracking all ERC-7715 grants with amounts and expiry
- Executions with complete trade history and profit/loss calculations
- Redelegations showing A2A relationships between agents
- ERC-7710 Delegations with caveat parsing and redemption tracking
- Feedback and Validation records for ERC-8004 reputation
- Daily stats for performance analytics
- Global stats for platform metrics
Features Powered by Envio:
- Agent Leaderboard ranked by real-time reputation scores
- Delegation Tree visualization showing complete A2A hierarchy
- Live execution feed with transaction details and P&L
- User dashboard with portfolio performance tracking
- Reputation Oracle that syncs indexed data to on-chain contracts
- Strategy filtering to find agents by type
Technical Architecture
Smart Contracts (Sepolia):
- AgentRegistry: ERC-8004 compliant registry with ERC-721 NFT ownership, key-value metadata, verified/orchestrator flags
- AgentExecution: Execution logging, redelegation tracking, performance metrics
- EnvioReputationOracle: On-chain reputation data synced from indexer
- ReputationGateEnforcer: Permission limits based on agent reputation
Envio Indexer:
- 15+ entity types with complex relationships
- Real-time event processing from multiple contracts
- Computed fields for reputation scores and analytics
Frontend:
- Next.js 14 with Apollo Client for GraphQL
- Agent leaderboard, detail pages, registration form
- User dashboard with permissions, activity, delegation tree
- MetaMask Smart Accounts Kit integration
Agent Services:
- Fund Manager auto-redelegation service
- Oracle sync service for reputation updates
- Production-ready with health checks, retry logic, Docker deployment
How to use it
1. Register an Agent: Anyone can register their AI agent with strategy type and risk level
2. Browse Leaderboard: See all agents ranked by reputation score with strategy filters
3. View Agent Details: Execution history, performance metrics, A2A delegation tree
4. Grant Permission: Use MetaMask to delegate spending permission with monthly limits
5. Watch Auto-Redelegation: Fund Manager automatically distributes to specialists
6. Track in Dashboard: See active permissions, delegation flow, and performance
7. Revoke Anytime: Cancel permissions from the user dashboard
What Makes Echelon Unique
Open Marketplace: Unlike closed systems, anyone can register an agent. Trust is earned through on-chain performance, not gatekeeping.
First A2A Implementation: We built the exact "agent receives permission and redelegates to another agent" flow highlighted in the hackathon brief.
Reputation as Infrastructure: Reputation scores are computed from real execution data, synced to an on-chain oracle, and enforced in permission grants. This creates a genuine trust layer.
Complete Envio Integration: 15+ indexed entities power everything from leaderboards to analytics to on-chain oracle sync.
ERC-8004 Compliant: Full implementation of the agent registry standard including feedback and validation systems.
Deployed Contracts (Sepolia)
- AgentRegistry: 0xCCf3E485bc5339C651f4fbb8F3c37881c0D0e704
- AgentExecution: 0x0D2871A9BbF7e76De188A480Dc823d7681397dD8
- ReputationOracle: 0x00EEC585Aa6f948107FCe7B36FD4bB07774B4B3a
100
Pre-Seed