Gumroad runs a Hermes Agent that handles parts of daily operations. On a schedule it triages support tickets, checks the facts in internal admin and analytics tools, and replies, closes or escalates each one.
- First reported
- gumclaw.github.io · 18 Jul 2026
- Added
- 13 Oct 2026
- Source checked
- 13 Oct 2026
What they did
The agent runs on a dedicated Mac and is woken by scheduled jobs. For support it pulls tickets from the helpdesk, checks the customer's situation in admin and Metabase, then acts. It also keeps a daily finance-event ledger and prepares month-end journal entries, and when a ticket turns out to be a product bug it can file an issue or open a fix in the codebase. Corrections become dated policy rules every future session must read.
What happened
The company's page describes support, X replies, finance prep and engineering fixes all running as recurring jobs, with around 40 skills built up as repeated tasks became scripts.
Keep in mind
The page gives no ticket volumes or accuracy figures. Standalone posts need approval, journal entries go to humans for review, and sensitive risk decisions stay with people.
Try it with your agent
Every morning, pull new tickets from <your helpdesk>. For each one, check the customer's account in <your admin tool> before deciding anything. Reply to routine questions using <your policy doc>, close solved ones, and escalate refunds over <threshold>, suspected fraud or bugs to <person or channel> with a short summary. When I correct you, add a dated rule to <your rules file>.
Replace the <angle brackets> with your own details.
Original source
gumclaw.github.ioUser storySomeone who uses the agent described their own setup. Figures are theirs.
Tools for this kind of work
The whole toolkit →
