Skip to main content
BACK TO BLOG

Why We Built OnlyFence: AI Agents Need Wallets. Wallets Need Guardrails.

· 5 min read
Seal Labs Team
Seal Labs Team

Why We Built OnlyFence

The agent revolution is here. Tools like Cowork and OpenClaw let AI agents analyze markets, form strategies, and act — all without you lifting a finger. Now point that at DeFi: an agent that can swap tokens at the right moment, supply collateral when rates spike, or rebalance a portfolio while you sleep.

That's real edge. The kind that used to require a quant team and a Bloomberg terminal.

But here's the catch no one talks about enough.

The Wallet Problem​

For an agent to do anything on-chain, it needs your wallet. Your private keys. Your funds.

And the moment you hand those over, you're trusting that:

  • The agent won't make trades you didn't approve
  • No one can manipulate it through prompt injection
  • A single hallucination won't drain your entire balance
  • The platform holding your keys won't get breached

That's a lot of trust for a system that, by design, makes autonomous decisions.

One bad prompt. One exploited endpoint. One compromised plugin. That's all it takes. We've seen it happen — agents tricked into signing malicious transactions, wallets drained in seconds, users left with nothing.

This isn't a theoretical risk. It's the #1 reason most serious traders won't let agents near their funds.

We Built OnlyFence to Fix This​

OnlyFence is a free, open-source DeFi toolkit where the agent is treated as a threat actor by default.

Not because agents are evil. Because good security assumes they could be.

The idea is simple: your agent gets full DeFi capabilities — swap, lend, borrow, manage positions — but every action passes through rules you define. If it breaks a rule, the action is blocked. No exceptions.

Agent wants to swap 10,000 USDC → SUI

  ✓ Token approved?     SUI is on the allowlist
  ✓ Under trade limit?  $200 max per trade → BLOCKED

  Action denied. Funds safe. You get notified.

Think of it as a company credit card with spending limits — not handing over the company bank account.

What We Got Right (and Why It Matters to You)​

1. Safety as the Default, Not an Add-on​

Most tools give you power first and let you figure out safety later. We flipped it.

Every agent interaction goes through a local policy engine — no server, no API call, no latency. Rules are checked in microseconds:

  • Token allowlist — the agent can only touch tokens you approve
  • Per-trade limits — cap every single transaction (default: $200)
  • Daily volume caps — limit total exposure over 24 hours (default: $500)
  • Simulation before execution — every trade is dry-run on-chain before signing
  • Full audit trail — every action logged, approved or rejected

Your keys never leave your machine. They're encrypted locally with your password. No cloud. No backend. No middleman.

2. A Real DeFi Toolkit, Not Just a Wrapper​

Here's what separates OnlyFence from a simple transaction signer: it's a complete DeFi layer.

Your agent doesn't need to know how Cetus works, or how to route through DeepBook, or how to calculate optimal lending rates on AlphaLend. OnlyFence handles it:

  • Swap across 5+ DEXes with best-price routing via 7K Aggregator
  • Lend and earn — supply assets to lending protocols
  • Borrow against collateral — leveraged strategies, managed safely
  • Check prices and balances — real-time data, no third-party API keys

One interface. All of DeFi. The agent sends a simple command, OnlyFence does the heavy lifting.

# Your agent just says:
fence swap SUI USDC 100

# OnlyFence handles: routing, simulation, limit checks, signing, confirmation

3. Multi-Chain from Day One​

We didn't build for one chain and bolt on others later. The architecture is chain-agnostic from the core — every module outside the chain layer is designed to work across networks.

Today: Sui is live with full support. Next: EVM chains (Ethereum, Base, Arbitrum) and Solana are in the pipeline.

One toolkit. One config. Multiple chains.

4. Built for Agents. Works for Everyone.​

OnlyFence ships with built-in skills for Claude, ChatGPT, and custom agents — so integration is minutes, not days.

But we know not everyone runs agents the same way:

You are...Your setup
Trying agents for the first timeOne command install on Mac, Linux, or Windows
A developer building a trading botJSON CLI output, pipe it into anything
Running a production agent fleetDocker container with security hardening
Deploying at scaleKubernetes with Vault or AWS Secrets Manager

Every output is structured JSON. Every action is auditable. Whether you're experimenting on your laptop or running 50 agents in production, the toolkit scales with you.

Who Is This For?​

If you're exploring AI agents for on-chain opportunities — OnlyFence lets you experiment without risking your funds. Set tight limits, start small, watch the audit log. Expand as you gain confidence.

If you're a developer building agent workflows — stop writing wallet management, transaction signing, and safety checks from scratch. OnlyFence gives you a tested, open-source foundation to build on.

If you're running agents in production — you need guardrails that don't add latency, keys that never leave your infrastructure, and a complete audit trail. That's what this is.

The Bottom Line​

AI agents on-chain are not a question of if — it's when. The edge they provide is too significant to ignore. But the current model of "give the agent your keys and hope for the best" is broken.

OnlyFence is the guardrail layer between your agent and your wallet. Open source. Local-first. Free.

The agent trades. You set the rules. Your funds stay safe.

Get started in 30 seconds:

curl -fsSL https://raw.githubusercontent.com/seallabs/onlyfence/main/install.sh | sh

Read the docs → | View on GitHub →

READ NEXT