SolanaGuard

> The problem -> AI agents are increasingly given their own Solana wallets so they can act autonomously — make payments, pay per request via x402, trade, and interact with DeFi. The moment an agent holds a funded wallet, a hallucination, a jailbreak, a prompt injection, or a plain bug becomes an unbounded financial loss. Solana has no native guardrail layer between an agent's decision and an on-chain transaction. Owners can't cap per-transaction or daily spend, can't allowlist which programs an agent may touch, can't require approval over a threshold, and have no way to freeze a misbehaving agent in real time. As agentic payments go mainstream, this missing safety primitive is the thing standing between "agents with wallets" and "agents you can actually trust with money." > The solution -> SolanaGuard is a policy firewall that sits in front of an AI agent's Solana wallet. An owner connects their wallet, cryptographically proves ownership, registers an agent, and defines a policy: max spend per transaction, daily limit, allowed/blocked program IDs, manual-approval thresholds, and risk thresholds. Every transaction intent is evaluated and returns allowed / warning / blocked with a risk score and the exact rule that fired - all written to an immutable audit trail. An action-bound emergency kill switch freezes the agent instantly.