Paperclip plus Hermes Agent turn scattered AI chats into a scheduled 10-agent company you control.
Both are free open source projects that you connect together. Paperclip links to all your AI agents, not just Hermes, so you can plug in OpenCore or Claude Code or whatever you want, and run them from one agent operating system with a single org chart, board and dashboard instead of scattered chats.
It is a 10 agent company: a CEO, a CTO, a CMO, engineers and marketers, three departments and a real org chart. Ada is the chief executive, with agents like Ogilvy as marketing officer working underneath her. They wake up on their own, pick up tasks together, do the work and report back.
Each agent has a role and a boss. Give them a goal and they wake up, check the board and get to work. They wake on a schedule you set, which you can pause or resume, and every task ends with a real deliverable attached to the issues ticket, running 24/7.
You click a task and see what they built. For the Goldilabs team you go back into Paperclip, open Goldilabs, then issues, and the website build shows what was created, what it worked on and the comments. Every task ends with a real deliverable attached to the ticket.
You set a limit on tokens for each AI agent, so when they hit it they stop and you get no surprises on token usage. That replaces the old risk, where a runaway loop with your AI agents could use up loads of tokens without you realising it.
Every agent has a token limit, and if you run on free models and free APIs it is free. Paperclip itself is free and Hermes Agent is free too, and agents can run on local models. He also points to free models on OpenAI and free Claude Code.
No. You are the founder, so nothing is created without your sign-off. You review what was delivered on the ticket and approve it. If you do not approve it, it goes back to working, and you can reopen it and leave a comment saying make this better.
He says it is basically a dashboard, so you just click hire, pick a role and you are done. If you can use a task manager you can use this. Step two is assigning a goal and walking away, because you give agents tasks like a real team instead of prompting them.
He had 20 AI tabs open with different agents and no idea what each one was doing, copy-pasting between them each day. It got messy and hard to control, you re-explained everything to the AI every session, and restarting your laptop lost all the terminals you had set up.
With only Hermes desktop or Hermes OS you cannot use everything else, such as OpenCore or Claude together, and you cannot see what you created with Claude. The agent operating system keeps every agent in one place, lets you switch between them and build your own mission control, customised how you want.
“Essentially, I had 20 AI tabs open, different agents, no idea what each agent was doing.”
“every task ends with a real deliverable attached to the issues ticket and it just keeps running 24 7.”
“so it's the same tools but they're just organized differently into a different system”
“some people say that looks like a team or developers built it honestly it just runs on 10 agents and zero humans”
“the people who learn to run AI teams now are years ahead when everyone else catches up”
“Hermes + Paperclip Just Changed AI Agents Forever!.” https://www.youtube.com/watch?v=_o1uhgQztYA. Transcript summary by WhipScribe, https://whipscribe.com/transcript-pages/hermes-paperclip-just-changed-ai-agents-forever.html.