ClawGuard

Most crypto transactions are either not screened at all or reviewed manually. As payment size grows, so does anxiety. Existing simulation and security tools catch obvious issues, but they don’t understand your behavior, history, or intent.

ClawGuard is a personalized wallet assistant built with an OpenClaw agent that learns how you transact onchain. It understands patterns like amounts, timing, frequency, recipients, and assets, and uses them to rate transaction risk in real time. When screening is on, routine USDC payouts can be approved instantly, while unusual or suspicious transactions are blocked and sent for review.

For risky actions, ClawGuard opens a review flow where users can ask questions and approve or reject transactions directly from Telegram or WhatsApp. The agent can also read invoices, analyze intent from chat history, assess risk, and proactively propose transactions for execution with no manual setup required. Screening can be turned off anytime for full auto-execution.

ClawGuard is built for speed without sacrificing security. It works for individuals, DAOs, DeFi protocols, and institutional treasuries that need safe, reliable payments at scale. Over time, the agent builds a behavioral profile that enables faster approvals, smarter protection, and less friction in everyday onchain payments.

Built on an industry-standard smart account with multisig flows, ClawGuard uses Circle Wallets for seamless social-style onboarding, with the agent operating through a designated server wallet.

ClawGuard is built with three main components working together.

1. User-facing app & smart account setup The frontend handles onchain interactions and transaction creation. Users onboard easily using Circle user-controlled wallets. Each user gets a smart account with multisig support, powered by ERC-4337 for gas sponsorship and transaction bundling. The backend queues pending UserOperations and coordinates signature collection for non-screened transactions.
2. OpenClaw AI screening & agent logic When a transaction is proposed with screening enabled, it is sent to an OpenClaw agent running models like Qwen3-235B and Sonnet 4.5 via OpenRouter updated with the ClawGuard skills. The agent continuously evaluates transactions against learned behavioral patterns and hard guardrails. If the transaction is safe, the agent partially signs the UserOperation and proceeds with execution. If it’s risky, the transaction is blocked and alerts are sent via Telegram or WhatsApp. Clearly malicious transactions are fully blocked. OpenClaw’s skill system powers autonomous decision-making using custom logic for transaction analysis.
3. Onchain execution on Arc Transaction signatures are queued and combined once final approval is available. The completed UserOperation is submitted to a bundler for gasless execution on Arc. ClawGuard uses an ERC-4337 module contract deployed on Arc, with Viem and Permissionless.js handling smart account interactions and execution.