AgentGuardrail

Smart Accounts for Your AI Agents

Deploy autonomous AI agents with built-in spending limits, policy enforcement, and complete auditability. AgentGuardrail uses ERC-4337 smart contracts to guarantee your agents can't exceed their constraints.

Whether you're building trading bots, market makers, or orchestration agents, AgentGuardrail provides the cryptographic safety layer your agents need.

Why Your Agents Need Guardrails

The Problem

  • •Private Key Exposure: Agents with your wallet's private key can drain your entire portfolio if compromised
  • •No Spending Limits: Agents can execute transactions of any size, with any recipient, at any time
  • •Unauditable Actions: Off-chain policies can be bypassed; no cryptographic proof of enforcement
  • •Integration Fragility: Different agents, different safety standards - no unified control

AgentGuardrail Solution

  • Isolated Signers: Agents get their own keypairs, completely separate from your wallet
  • On-Chain Enforcement: Spending limits enforced by smart contracts, not APIs
  • Cryptographic Proof: Every transaction logged on-chain with block number and tx hash
  • Unified Control: One platform for all your agents, consistent safety guarantees

How It Works in 3 Steps

1

Generate Bot Signer

Create a dedicated cryptographic keypair for your agent. This keypair has zero access to your personal wallet.

Even if the agent code is compromised, the attacker can't access your funds. The signer is isolated by design.

2

Deploy Smart Account

Deploy an ERC-4337 smart account for your agent. The smart account is owned by you, controlled by the agent's signer.

Works with all ERC-4337-compatible applications. Supports batching and arbitrary transactions.

3

Set Guardrails

Define a JSON policy specifying what your agent can do: allowed actions, tokens, protocols, spending limits.

Enforced on-chain. Invalid transactions cannot execute, period.

Key Features

On-Chain Enforcement

Policies enforced by smart contracts, not APIs. Guaranteed by the blockchain.

Bot Signer Isolation

Agents get their own keypairs. Your personal wallet stays safe, always.

Smart Account Deployment

Deploy ERC-4337 smart accounts with built-in spending limits and policy validation.

Complete Audit Trail

Every action logged on-chain. Full visibility, cryptographic proof.

Multi-Agent Management

Deploy multiple agents, each with separate policies and smart accounts.

Framework Agnostic

Works with LangChain, Claude, ChatGPT, Crew AI, or any custom agent framework.

JSON Policies

Define constraints in simple JSON. Actions, tokens, protocols, spending limits.

Real-Time Monitoring

Dashboard and API for live transaction logs, policy violations, metrics.

Common Use Cases

Trading & DeFi Bots

Deploy autonomous trading bots that execute swaps, provide liquidity, or rebalance portfolios, all within predefined spending limits.

Max $10K/tx, $100K/day

Market Makers (AMMs)

Market-making agents that manage liquidity positions across multiple pairs with guaranteed constraints.

Allowed tokens: USDC, WETH, DAI

Orchestration Agents

Coordinate actions across protocols (borrow, swap, stake) without ability to deviate from approved flows.

Allowed protocols: Uniswap, Aave, Curve

Enterprise Operations

Automate business workflows (payments, transfers, accounting) with guaranteed compliance and auditability.

Time-windowed policies, audit logs

For Developers

Integration Flow

1.Your agent framework (LangChain, Claude, GPT) makes decisions
2.Call AgentGuardrail API to validate action against policy (pre-flight)
3.If valid, agent signer calls the smart account
4.Smart contract enforces policy again (final safety layer)
5.Transaction executes with proof of compliance on-chain

API Endpoints

POST /api/v1/validate
Pre-flight validation of single action
POST /api/v1/validate/batch
Validate multiple actions in sequence
POST /api/v1/validate/simulate
Simulate without committing
GET /api/v1/audit
Query complete audit log

Example: Trading Agent Policy

ETH Trading Agent

A policy that lets your agent swap USDC for WETH on Uniswap within strict spending limits

{
  "name": "ETH Trader",
  "description": "Autonomous agent that swaps USDC for WETH on Uniswap",
  "enforceOnChain": true,
  "policies": {
    "allowedActions": ["swap", "transfer"],
    "allowedTokens": [
      "0xA0b8...USDC",
      "0xC02a...WETH"
    ],
    "allowedProtocols": ["uniswap-v3"],
    "constraints": {
      "maxValuePerTransaction": "2.5",
      "maxDailyVolume": "25.0",
      "minSlippageProtection": "0.5%"
    },
    "timeWindow": {
      "validFrom": "2024-03-01T00:00:00Z",
      "validUntil": "2024-03-31T23:59:59Z"
    }
  }
}

When this agent tries to execute a 2.51 ETH swap, the smart contract automatically rejects it before it hits Uniswap. If the agent is compromised, it still can't exceed the limits. They are enforced by the blockchain.

Why AgentGuardrail Wins

FeatureAgentGuardrailMulti-SigRate LimitingOff-Chain Validator
On-Chain Enforcement✅❌❌❌
No Private Key Sharing✅⚠️✅✅
Detailed Policies✅❌⚠️✅
Cryptographic Proof✅⚠️❌❌
Multi-Agent Support✅✅⚠️⚠️
Open Source✅⚠️⚠️❌

Getting Started

1

Create an Account

Sign in with your wallet using SIWE. No email, no password.

2

Generate a Bot Signer

Create a dedicated keypair for your agent. Save it securely.

3

Deploy a Smart Account

Deploy on Base mainnet or Sepolia testnet. A one-time $10 creation fee is charged at deployment.

4

Write Your First Policy

Define constraints in JSON. Use templates for common use cases.

5

Integrate Your Agent

Add AgentGuardrail API calls to your LangChain, Claude, or custom agent code.

6

Monitor & Adjust

Watch the audit log in real-time. Update policies as your agent evolves.

Frequently Asked Questions

What if my agent gets hacked?

The agent's keypair can be compromised, but the smart account's constraints are enforced by the blockchain. Even a compromised agent cannot exceed the policy limits. Your personal wallet remains entirely separate and safe.

How do I update my policy?

Policies can be updated in real-time from the dashboard. The new policy takes effect immediately for new transactions. All changes are logged on-chain.

Does this work with my LLM?

Yes. AgentGuardrail is LLM-agnostic. Whether you're using LangChain, Anthropic Claude, OpenAI ChatGPT, or custom code, you can integrate via our REST API.

Is it ready for production?

Yes. The smart contracts are audited, the platform is battle-tested, and policies are enforced on-chain. Start on Sepolia testnet first to validate your setup.

What networks do you support?

Currently Base mainnet (Chain ID 8453) and Sepolia testnet. More EVM-compatible chains coming soon.

How much does it cost?

Smart account deployment has a one-time $10 USD fee (paid in ETH at deployment). Outbound transfers are charged 10 basis points (0.10%), capped at $100 per transaction. Inbound deposits, policy management, and validation API calls are free.

Start Building Safe AI Agents Today

Deploy autonomous agents with guaranteed safety. On-chain enforcement. Cryptographic proof. Complete auditability.