On-chain Policy Enforcement for AI Agents

Guardrails for
Autonomous AI

Deploy smart contract policies that enforce spending limits, protocol whitelists, and instant kill switches — so your AI agents operate safely on-chain.

Launch Dashboard Learn More

100%

On-chain

<1s

Kill switch

0

Trust required

The Problem

AI agents are operating without guardrails

Autonomous AI agents are making on-chain transactions — swapping tokens, entering DeFi positions, and moving funds — with zero enforceable constraints. One hallucination, one prompt injection, or one logic bug can drain a wallet in seconds.

Existing solutions rely on off-chain checks that agents can bypass. There's no on-chain enforcement layer that the agent itself cannot circumvent.

No spending limits on agent wallets
Prompt injection can override soft constraints
No way to instantly halt a rogue agent
Off-chain guards are easily bypassed
How It Works

Four steps to safe autonomy

01

Deploy a Policy Contract

The owner creates an on-chain policy specifying daily spending limits, single transaction caps, allowed protocols, allowed tokens, and slippage bounds.

02

Agent Routes Through Policy

The AI agent sends every transaction through the policy contract. The contract validates each action against the configured rules before execution.

03

Monitor in Real Time

Every executed and blocked transaction emits on-chain events. The dashboard provides a live activity feed with full transparency.

04

Kill Switch When Needed

The owner can instantly deactivate any agent with a single transaction. The kill switch is enforced at the smart contract level — the agent cannot override it.

Why Agent Safety Layer

Trust the contract, not the agent

AI agents are powerful but unpredictable. Smart contract policies give you mathematical guarantees — not just promises.

Tamper-proof Enforcement

Policies live as immutable smart contracts. The AI agent cannot modify, bypass, or disable its own constraints — only the owner can.

Instant Emergency Response

Kill switch activates in a single transaction. No waiting for off-chain services to respond. Immediate, verifiable, on-chain halt.

Full Transparency

Every action is recorded on-chain with events. Audit trails are permanent and verifiable by anyone — no black box behavior.

Composable Architecture

Works with any EVM chain and any agent framework. Plug in to LangChain, AutoGPT, or custom agents without changing their core logic.

Granular Controls

Set per-agent policies with daily spend limits, per-transaction caps, protocol whitelists, token restrictions, and slippage bounds.

Chain Agnostic

Deployed on Base with plans for multi-chain support. The same policy patterns work on Ethereum, Arbitrum, Optimism, and beyond.

Features

Everything you need for safe agents

Per-agent policy contracts

One-click kill switch

Daily spend limits

Single transaction caps

Protocol whitelists

Token restrictions

Real-time activity feed

AI-generated weekly digests

TypeScript SDK

Slippage & leverage bounds

On-chain event logging

Multi-agent management

Architecture

How the pieces fit together

Smart Contracts

  • PolicyManager — factory & registry
  • AgentPolicy — per-agent enforcement
  • On-chain event emission
  • Kill switch & resume functions

TypeScript SDK

  • Typed contract interactions
  • Policy creation helpers
  • Event parsing utilities
  • Framework-agnostic design

Dashboard Frontend

  • Real-time activity monitoring
  • Policy management UI
  • One-click kill switch
  • AI-powered weekly digests

Transaction Flow

AI Agent→Policy Contract→Validate Rules→Execute / Block→Emit Events

Ready to secure your AI agents?

Deploy your first policy in minutes. The Agent Safety Layer is open source and free to use.

Open Dashboard Documentation