The trust & safety layer for the AI agent economy: vets tokens, wallets and agents, screens transactions before they are signed, and defends on-chain via a proof-gated vault that can never drain you.
Prospect is the trust and safety layer for the AI agent economy, built on OKX OnchainOS and X Layer. It turns raw on-chain data into decisions: token, wallet and agent reports that always end in a verdict (AVOID/OK, TRUSTED/RISKY, HIRE/AVOID), plus a market brief and a discovery scout - so an agent, or the human behind it, never has to read ten dashboards to answer "is this safe?".
The Transaction Firewall screens a dApp, an off-chain signature or a raw transaction BEFORE it is committed and returns SAFE / WARN / BLOCK, catching infinite approvals, drainer patterns and malicious permit signatures. Guardian watchers run 24/7, and a deterministic Auto-Defense brain picks exactly one bounded action when risk rises: exit, stop-loss or revoke.
Every autonomous action lands in ProspectVaultV2, a deny-by-default Policy Vault deployed on X Layer: target allowlist, per-action and daily caps, a cooldown, and a one-shot on-chain proof the owner must accept first. The agent can defend you, but is mathematically incapable of draining you. All 11 skills ship as MCP tools (stdio + Streamable HTTP) and as pay-per-call x402 / A2MCP services - ASP #7151 on OKX AI.
Built and shipped during the OKX.AI Genesis Hackathon. Registered as ASP #7151 on OKX AI, with A2MCP services live behind a public endpoint and a stable Tailscale Funnel host.
Working today: 11 MCP tools over stdio and Streamable HTTP; the full x402 paid path (402 challenge -> EIP-712 payment authorization in USDT0 -> facilitator verify/settle) on X Layer; a real-LLM Copilot as the natural-language front door; a live web app, landing page and whitepaper; and ProspectVaultV2 deployed and adversarially tested on X Layer - non-allowed target, over-cap, missing proof and proof replay all revert, while an owner-accepted proof executes.
Next: mainnet vault deployment, more Guardian strategies, and deeper agent-to-agent trust scoring (ERC-8004) so agents can vet each other before they hire each other.
Not currently raising. Bootstrapped and self-funded, built for the OKX.AI Genesis Hackathon.
Open to grants, partnerships and pilot integrations within the OKX and X Layer ecosystem.