--- title: 'Hermes + Paperclip: Automate Anything' source: 'https://youtube.com/watch?v=pbBgFa49oWw' video_id: 'pbBgFa49oWw' date: 2026-07-28 duration_sec: 491 --- # Hermes + Paperclip: Automate Anything > Source: [Hermes + Paperclip: Automate Anything](https://youtube.com/watch?v=pbBgFa49oWw) ## Summary The video introduces a framework for managing multiple AI agents using Paperclip (open source orchestration) and Hermes (AI agent from Nous Research) to create an 'Agent OS' with roles, schedules, token limits, and ticket-based output, replacing chaotic tabs with a structured team. ### Key Points - **Agent OS Structure** [00:15] — Introduces an agent operating system with job titles, a boss, a schedule, and tickets for real work. - **Live Board Tracking** [00:28] — A live board shows agents with timestamps, who picked up what, and start/end times. - **Five-Layer Framework** [01:36] — The five layers: founder (human), org chart (roles), heartbeat (schedule), token limit, and output (tickets). - **Paperclip as Orchestrator** [02:00] — Paperclip is an open-source project (MIT licensed) that handles layers two through five, orchestrating any agent. - **Scheduled Agent Wake-ups** [02:56] — Agents wake on a schedule set by the user, can be paused, and have a hard token limit to prevent runaway. - **Output to Tickets** [03:25] — Every finished task ends with a real file attached to the issues tab, which can be opened, read, and approved. - **Hermes Agent Capabilities** [04:06] — Hermes has over 30 native tools, persistent memory, session persistence, 80+ loadable skills, and MCP support. - **Hermes-Paperclip Adapter** [04:20] — Nous Research maintains an official Paperclip adapter, allowing Hermes to drop into the org chart as a managed employee. - **Memory Architecture** [05:56] — Paperclip holds the org chart, Hermes holds agent-level memory, and an Obsidian vault holds business context, enabling context to compound. - **Model Swapping** [06:21] — You can swap the model underneath any agent without rebuilding the org chart, offering flexibility. - **Tips for Getting Started** [07:05] — Start with three agents (CEO and two workers), write a clear mission, set token limits early, stagger schedules, and use the comment box for feedback. ### Conclusion The key to scaling AI agents is a structured system (org chart, schedules, token limits, output tracking) using Paperclip and Hermes, enabling a 'one-person empire' where agents coordinate autonomously. ## Transcript anything. How many AI agents do you have open right now? 4, 10, 20? Now tell me can't. Nobody can. The tools were never the problem. I'm the digital avatar of Julian Goldie and I help people learn AI tools and actually use them in their Hermes and paper clips together inside my agent OS. A full team of agents with job titles, a boss, a schedule, and real work attached to tickets. Stick around because the piece most people miss isn't what's on my screen. There's a live board. On the left, every agent I have doing right now with a timestamp on each one. Ada sits at the top as the chief marketing lead, an engineering lead, and operations lead. Under them, the teams. talking to you. I can see who picked up what, when they started, and what came running inside my agent OS. Here's a real one. I ran this to build out a 2-week email sequence for the AI Profit Boardroom. I didn't write it. I set the and it came back with the full sequence mapped out day by day. Another agent Profit Boardroom off its own back. I didn't ask. It checked the board, saw the goal, and got on with it. Both landed as tickets I could open, read, and approve. I call this framework the one-person empire. Not because you're You set the mission and the team handles together is structure, not tools. Because here's what everyone gets wrong. 20 tabs open is not a team. That's what I had a few months ago. One chat here, them all day. I'd lose track of who was doing what. I'd re-explain the same I'd restart my laptop and the whole thing was gone. Terminals closed, context wiped, back to square one. So think of it in five layers. Layer one is the founder. That's the human. You set the mission and approve the big moves. Layer two is the org chart, where every agent gets a role, a title, and a boss. Layer three is the heartbeat, which is a schedule that wakes them up. Layer four is the token limit, so nothing runs away from you. Layer five is the output, something real attached to a ticket. Miss a layer and you're back to tabs. Paperclip is what handles layers two through five. It's an open source project, MIT licensed, and you self-host it. It's not a model and it's not an sits above the agents you already use. Claude code, code X, cursor, open code, open claw, Hermes, all of it. Any agent, any runtime, one org chart. In the everything they've done, how long ago, and the actual output file they produce. Now, if you want the exact setup I'm running, it lives inside the AI Profit section in the classroom with a video tutorial and the complete Agent OS zip rebuilding this from scratch. We've built out a 30-day roadmap that walks you from a blank dashboard to a working agent team step-by-step. You get the four weekly coaching calls where you can share your screen, and we'll get with you live. It's updated every single day, so what you see in this video is what you'll find inside the AI Profit Boardroom. Back to the setup. The feel different. Agents wake up on a schedule you set. I open an agent, tell on its own. I can pause any of them, Right now, I've got most of mine paused just so the screen isn't chaos while I'm Every agent gets one. When it hits that limit, it stops. That's it. No runaway while you're asleep. And if you're pointing your agents at local models you nothing to burn through anyway. Paperclip is free, Hermes agent is free, both open source. Then output. Every finished task ends with a real to the issues tab and there it is, the keyword research, the site an agent built, the screenshot it took of its own work. I open the ticket, read what was comment saying make this better, and it goes back and reworks it. Nothing ships without me. So, why not just open Hermes in a terminal or open Claude code and go? Because that's managing a chat, one agent, one conversation, one thing at a time. Compare the two side by side. Old way, 20 tabs, no idea who's doing what, a pile of conversations at the end of the day. New way, agents with roles and a boss, goals instead of prompts, one org chart, one board, one token limit, and sessions that survive a reboot. Hermes is where this gets interesting. Hermes agent is from Nous Research, it's open source, and it runs on your own machine. It's got over 30 native tools, persistent memory, session persistence, over 80 loadable skills, and MCP support built in. And Nous Research maintain the official Paperclip adapter themselves, so Hermes drops straight into the org chart as a managed employee. Each Hermes agent keeps its own memory file, while Persistent memory at the agent level, coordination at the company level. came out of it. My agents built a live monitoring dashboard where I can watch expect that out. They built the Goldilab site, took a screenshot of it, and together an SEO strategy and some ad creatives. The CEO agent started proposing new agents to hire. Here's another one I ran. I used this to plan out a full content series designed to Boardroom. I gave it the topics AI Profit Boardroom members keep asking about, and it came back with the angles and the hooks in one pass. And I've used it to map the onboarding path new AI Profit Boardroom members walk through on day one, so they land inside and know exactly what to do next. Now, the part bit I care about most. All of this is running inside my Agent OS. Paperclip is voice chat with Hermes, normal chat, a studio for video, images, and voice, my built, my MCPs, and a control room. We with a wake word, so I can talk to it and it answers. That's the difference. do you see everything Claude built? How do you bring Open Claude in? You don't. Agent OS is mine, so I customize it. This morning, I decided I wanted teams of agents producing work around the clock, so we built that in. You can't do reason that matters is memory. Every agent I run sits on top of the same shared context. Paperclip holds the org chart, Hermes holds the memory at the agent level, and my Obsidian vault sits underneath all of it holding the business context. So, when an agent picks up a ticket, it already knows what the same background in again. That's the exact thing that dies when you work in tabs. Context lives inside one chat, then that chat closes and it's gone. Here it compounds instead. The team gets more useful the longer it runs because next agent starts with. And because I own the whole thing, I can swap the model underneath any agent without rebuilding the org chart. Same team, different engine, nothing else changes. Let me handle the pushback because I need a big team and endless tokens for few agents, and every agent has a hard token limit. Second, managing a bunch of 20 tabs version. With an org chart, the agents coordinate each other. Ada review. Third, I'll wait until it's more ready. Agent OS didn't start out looking day, and that only worked because we started early. So, a few tips before you go. Start with three agents, not 10. A CEO and two workers. Get one clean add anyone else. Write your mission properly because every ticket traces back to it, and vague goals give you vague work. Set the token limit on day one, not after something surprises you. Stagger the schedules so your agents And actually use the comment box on tickets. Rejecting work with a clear note is how the team gets sharper. If you want the full process, SOPs, and 100-plus AI use cases like this one, join the AI Success Lab. Link's in the comments and description. You'll get all the video notes from there, plus access to our community of 85,000 members who are crushing it with AI. And when you go same wall everyone hits. The first agent won't wake up, the adapter won't connect, you'll stare at an empty org chart wondering which roles to hire That's exactly what the AI Profit Boardroom is for. You get the full Agent OS zip file ready to install, the 30-day roadmap built around getting paperclip prompts, the Obsidian memory setup, and the walk-through tutorials, four weekly coaching calls where you share your screen and we fix it with you in real this stuff every day. Go to AIprofitboardroom.com.