TrustLoopGuard: Control AI agents before irreversible actions

Hey Product Hunt, I built TrustLoopGuard after testing AI agents in production and seeing the same gap repeatedly: teams can evaluate what an agent says, but the dangerous moment is when it acts. An agent can propose a refund, send an email, update a customer record, call an API, deploy code, or delete data. The important question is whether that exact action should be allowed, blocked, rewritten, or sent for approval before the side effect happens. TrustLoopGuard sits at that boundary. Your app sends the proposed output or tool call plus its context. The runtime checks policy, authority, values, provenance, and approval rules, then returns a typed decision with a trace. Money-bearing actions add durable authorization state plus decision and execution receipts. It is open source and self-hostable, with TypeScript, Python, and Rust SDKs, a provider-compatible gateway, and an MCP/tool-call path. I am launching this early because the use case is still being discovered. I want to learn which irreversible action is hardest for teams to trust today.