Manage online stores and campaigns with an open-source agent

The author built an open-source agent that turns voice notes into reports, checks orders and ad spend every five minutes, and suggests one next task across crowdfunding, Amazon and Shopify businesses.

What They Did

The author built the workflow around Nous Research’s Hermes Agent with a third-party LLM gateway and a three-tier model router: a cheap model for routine work, a smarter one for ambiguous tasks, and a local fallback for outages or private jobs. They fed it voice notes or messy text from two crowdfunding campaigns, an Amazon store, and a Shopify site; a cron job then checked orders, ad spend, inbox flags, and a todo list every five minutes, kept one memory file updated, and pushed back one next task, plus weekly and monthly recaps.

How It Went

After six months, the author says the core loop stuck where Notion, Todoist, and lightweight CRMs had failed, and the routing setup was about 60–70% cheaper than using a top model every time while being more reliable. It still mislabels tasks, misses details, and some reports or automations end up useless.

Worth Knowing

Their main lesson was to start with just three automations that worked—daily report, next-task push, and weekly recap—after spending two weeks over-automating and making the system brittle.