AutoAgency
My own product: a desktop workspace where a small team's AI agents run recurring work on a schedule, with a record of every run. It ships with atelier and salon, two operator plugins I built for design, product and social work.
Most AI tools are one person and one chat. AutoAgency is for a team that wants agents to do recurring work, a weekly research desk or a content pipeline, without anyone babysitting a chat window.
You describe the operation once. Named agents run it on a schedule. Each run leaves a timeline of which agent did what, with which tools and files, and the result lands in the team's shared folder. A person reviews before anything is applied. Memory lives in the team's own git, so what the agents learn this week is still there next week, and none of it sits in a vendor's tenant.
Two open-source plugins ship with it and also work on their own inside Claude Code. atelier is the design, product and brand operator: 15 specialist agents, 33 skills and 32 commands, from PRD writing and design critique to naming and positioning, with per-project memory so it learns a client's voice and decisions. salon is the social campaign operator for X, LinkedIn, Discord and Telegram: briefs, calendars, platform-native drafts, reply playbooks and listening. It shares atelier's memory and its anti-AI-tone guardrails, so nothing it writes sounds like a bot wrote it.
AutoAgency is built on an open-source multi-agent desktop project. The operational layer is mine: the scheduled automations, the run records, folder delivery, git-shared memory, the team server and sign-in. So are the strategy, the pitch and the product decisions. It is the first thing on this page where I am the founder rather than the hired builder.
