Hey Product Hunt — we’re the makers of Loopgraph. We kept seeing the same failure mode in AI automation: we connect an agent directly to a trigger, give it tools, and hope the prompt contains enough policy to make the right decision. That can automate a task. It does not give a company an operating model. Real company work crosses departments, systems, owners, approval boundaries, and definitions of “done.” Ten alerts may describe one unresolved problem. One event may contain several problems. The right response may be to act, wait, request context, ask for approval, or deliberately abstain. We built Loopgraph to make that structure explicit. Loopgraph turns recurring company work into versioned, visual loops. Hermes Agent receives business events and proposes where each problem belongs. Loopgraph checks the registered LoopSpecs, routing contracts, readiness, policy, deduplication, approvals, and exact prepared action before anything is allowed to run. The short version is: Hermes proposes. Loopgraph commits. Today, the supported local flow lets you: • interview a department and propose its recurring loops • materialize accepted loops as versioned specs • visualize Hermes → Department → Multiple Loops • rehearse event routing with fixtures • simulate runs and inspect reproducible traces • keep risky actions behind explicit approval boundaries Loopgraph is MIT-licensed, local-first, and in active early development. Live execution is still experimental; the safest path today is to design, rehearse, simulate, and review locally. We’re launching now because we want the operating model to be challenged in public before the graph becomes larger. The question we would genuinely love your answer to: What recurring company process would you trust an agent to recommend—but not execute yet? We’ll be here throughout the launch to answer questions, help people run the local setup, and hear where this model breaks.