01What Is Agent OS and How It Works
Agent OS is a system that brings all your custom workflows, agents, and memory files into one place. It lets your AI agents work like a real team. They can share memory and talk to each other around the clock.
In this episode, Julian Goldie answers community questions about setting up and using Agent OS. He explains that Agent OS works with whatever agents and memory you already have. You do not need to delete anything or start from scratch.
02Main Features of Agent OS
Agent OS has several key features that make it powerful:
- Plug-and-play memory: It works with existing memory from Hermes, Claude Code, or other agents. No need to erase files.
- One clean system: You can run multiple clients and projects from a single Agent OS instance instead of building separate ones.
- Agent teams: Use tools like mixture of agents (a panel of different agents) where an aggregator fuses answers to get better results.
- Communication: Add all your agents to a Telegram group chat so they can talk to each other automatically.
- Orchestration: Use Hermes as an orchestrator with Paperclip to spawn free Claude Code agents when needed.
03How to Get Started with Agent OS
Follow these steps to set up your Agent OS:
- Keep your current setup: Do not delete your existing memory files or agents. Agent OS can integrate with them.
- Simplify your architecture: Use one single Agent OS for all your projects, even if you have many clients. Avoid building two or more separate systems.
- Add agents to a group chat: In Telegram, create a group chat and add all your agents. They will start communicating automatically.
- Use Paperclip for spawning: If you need free Claude Code agents from Hermes, set up Paperclip with an organization child structure.
- Focus on one primary agent: If you only have one clear use case, use just that agent (like Hermes) instead of many working together.
04How to Use Agent OS: Examples from the Episode
Julian shows how to solve common problems with Agent OS. For example, if you want your agents to work as a team, you simply add them to a Telegram group chat. No extra coding needed.
For spawning free Claude Code agents from Hermes, Julian recommends Paperclip. But he warns: only do this if you have a clear reason. Otherwise, just use one agent like Hermes alone.
With the mixture of agents feature, you set up a panel of agents that each give answers. Then an aggregator reviews all answers, looks at mistakes and benefits, and fuses them into one improved answer. This gives you better intelligence than a single model can.
05Results and Comparisons Shown in the Episode
Julian does not run formal tests, but he shares clear outcomes from real user questions:
- Simpler setups work better: Users who tried to build two separate Agent OS instances found they were overcomplicating things. Julian says one clean system makes agents faster and easier to run.
- Terminal timeouts: A user named Trevor hit agent turn caps when using mixture of agents for big builds. The agents timed out. Julian does not give a fix for the timeout, but the takeaway is that complex setups can cause problems.
- No benefit from too many agents: In one example, the user had Claude, Hermes, sub-agents, and Kanban boards. Julian saw no reason for all of them to work together. He says one agent would do the job.
06Tips, Limits, and Mistakes to Avoid
Here are the most important lessons from the episode:
- Do not overbuild: The biggest mistake is building too many agents and sub-agents. It makes your system weaker, not stronger. Keep it simple.
- One Agent OS is enough: Even with multiple clients or projects, use one single Agent OS. Separate instances add complexity without benefit.
- Forget old memory files: You do not need to delete them. Only consider it if you are trying to save on token costs.
- Watch out for turn caps: When using mixture of agents, the agent turn cap can cause timeouts. Plan for shorter or simpler agent runs.
- Use group chats for communication: Telegram group chats let agents talk without extra setups like shared boxes.
07Who Should Use Agent OS
Agent OS is for anyone who wants to run AI agents around the clock without constant manual setup. It works for non-technical users who need a simple system, and for experienced builders who want to streamline multiple agents.
Julian mentions that the community includes a beginner-to-expert course, weekly coaching calls, and a map to meet local builders. If you want one clean system that ties together Hermes, Claude Code, and other agents, Agent OS is a good fit. It matters because it saves you weeks of wasted work by avoiding overcomplicated setups.
