Aleph Hackathon

WWall sits between any MCP-compatible AI agent (Claude Desktop, Claude Code, etc.) and its wallet. The agent never sends money directly — every payment request goes through a deterministic policy engine first, which checks it against rules like spending caps, allowlists, and confirmation thresholds. The verdict is always one of three outcomes: allowed, rejected, or escalated to a human for confirmation. Critically, the guard doesn't evaluate the agent's intent or reasoning — only the facts of the call (recipient, amount, frequency) — so it can't be talked around, regardless of how the agent arrived at the request.