Autonomous AI Agents with On-Chain Credentials and Treasury Governance
AgentGuard is a decentralized framework that allows AI agents to earn on-chain credentials through completed work, build reputation over time, and autonomously create treasury spend requests based on their analysis. Human governance remains in control through approval, rejection, and execution of treasury actions.
Built on Arbitrum Sepolia, AgentGuard combines AI decision-making, verifiable on-chain credentials, and treasury governance into a single system.
(https://agent-guard-wine.vercel.app/)
(https://agent-guard-xsll.onrender.com)
0x954f2FBC0fA38E3fa940E551EF5396C132cE1286
(https://youtu.be/KAdVY6d82LE)
Traditional AI agents can generate recommendations, but they have no verifiable reputation and no structured path to participate in treasury decisions.
AgentGuard introduces:
- On-chain credentials earned through completed tasks
- Reputation-based agent progression
- Credential-gated treasury permissions
- Autonomous treasury request generation
- Human-in-the-loop governance controls
- Verifiable on-chain history for every agent
An agent's authority is determined by its demonstrated performance rather than arbitrary trust assumptions.
Every completed task generates:
- Task type
- Evaluation score
- Evidence hash
- Timestamp
Credentials are permanently recorded on-chain.
Agents build reputation through performance.
Current credential tiers:
| Level | Requirements | Spend Limit |
|---|---|---|
| Unverified | Default | 0 ETH |
| Bronze | Performance threshold reached | 0.1 ETH |
| Silver | Higher reputation threshold | Higher treasury authority |
| Gold | Highest reputation tier | Maximum treasury authority |
Credential levels are derived automatically by the smart contract.
After completing a task:
- Agent performs analysis
- Output is evaluated
- Credential is recorded on-chain
- Agent determines whether treasury action is required
- Deterministic treasury policy computes allowable amount
- Spend request is submitted on-chain
The AI decides whether funding is required.
The amount is determined by deterministic policy rules, not by the AI.
Agents cannot spend funds directly.
All treasury actions remain subject to governance approval.
Supported actions:
- Approve Request
- Reject Request
- Cancel Request
- Execute Request
This creates a balance between autonomous intelligence and human oversight.
The AI never decides the amount of money requested.
Instead:
- Evaluation score determines confidence
- Credential level determines maximum authority
- Treasury availability limits request size
This ensures predictable and auditable treasury behavior.
┌────────────────────┐
│ User Creates │
│ Task │
└─────────┬──────────┘
│
▼
┌────────────────────┐
│ AI Agent │
│ Executes Analysis │
└─────────┬──────────┘
│
▼
┌────────────────────┐
│ Evaluation Engine │
│ Scores Output │
└─────────┬──────────┘
│
▼
┌────────────────────┐
│ On-Chain Credential│
│ Recording │
└─────────┬──────────┘
│
▼
┌────────────────────┐
│ Treasury Decision │
│ Service │
└─────────┬──────────┘
│
▼
┌────────────────────┐
│ Spend Request │
│ Created On-Chain │
└─────────┬──────────┘
│
▼
┌────────────────────┐
│ Human Governance │
│ Approval / Reject │
└────────────────────┘
- React
- TypeScript
- Vite
- Node.js
- Express
- TypeScript
- Solidity
- Hardhat
- Arbitrum Sepolia
- Groq API
- Qwen 3 32B
- Register agents
- Track reputation
- Track credential levels
- Track treasury authority
- Record task credentials
- Store evidence hashes
- Automatic level progression
- Treasury deposits
- Spend requests
- Approval workflow
- Rejection workflow
- Execution workflow
Task Created
│
▼
Agent Executes Task
│
▼
Output Evaluated
│
▼
Credential Recorded On-Chain
│
▼
Agent Determines Funding Need
│
▼
Deterministic Policy Computes Amount
│
▼
Spend Request Created On-Chain
│
▼
Governance Review
│
▼
Approve / Reject / Execute
Agent Registered
│
▼
Completes Tasks
│
▼
Earns Credentials
│
▼
Builds Reputation
│
▼
Reaches Bronze Level
│
▼
Eligible For Treasury Requests
│
▼
Creates Autonomous Spend Requests
│
▼
Governance Review
AgentGuard
│
├── frontend/
│ ├── src/
│ └── public/
│
├── backend/
│ ├── src/
│ ├── routes/
│ ├── services/
│ └── models/
│
├── contracts/
│ ├── AgentGuard.sol
│ └── scripts/
│
└── README.md
- Multi-agent treasury councils
- Agent voting mechanisms
- ZK-based credential verification
- Cross-chain credential portability
- Agent-to-agent coordination
- Autonomous budget allocation
- Wallet-based governance approvals
- DAO integrations
AI systems are becoming increasingly capable of making decisions, but they lack verifiable reputation and accountable governance structures.
AgentGuard bridges this gap by enabling agents to:
- Earn verifiable on-chain credentials
- Build reputation through demonstrated performance
- Participate in treasury workflows
- Operate under transparent governance constraints
The result is a framework where autonomous agents can contribute meaningfully to decentralized organizations without sacrificing accountability or human oversight.
MIT