01What Hermes Agent Curator Is and How It Works

Hermes Agent Curator is a feature inside Hermes Agent that picks the right agent, tools, and model for your task automatically. You just type what you want, and it builds a team and runs the job without manual setup.

Curator has three main parts. The task reader understands your goal. The agent picker scores every agent, tool, and model to find the best fits.

The runner executes the plan, handing output from one agent to the next, and gives you the final answer.

In this episode, Julian Goldie shows how Curator replaces the old way of picking agents yourself. Instead of choosing an agent and hoping it works, Curator does the picking for you. It is like hiring a manager who knows every worker on your team.

02Main Features of Hermes Agent Curator

Hermes Agent v113 shipped Curator as the headline feature. It is open source, runs locally, and works with any model like Claude, GPT, Gemini, or Olimar.

  • Task reader - Understands what you really want, not just the surface ask.
  • Agent picker - Scores agents, tools, and models to pick the best for each job.
  • Runner - Runs agents in order and connects outputs.
  • Learning picker - Gets better with each use, seeing which agents you keep or reject.
  • Kanban integration - Curator slots on top of the existing Kanban feature, assigning agents to cards automatically.
  • Feedback button - Thumbs down teaches Curator to avoid bad picks next time.

Curator breaks big tasks into smaller jobs and runs them in the right order. It saves time because the right agent gets picked the first time, avoiding wasted runs.

03How to Set Up Hermes Agent Curator

Julian Goldie explains you need a few good agents set up first. Curator is only as good as the agents you give it. Here is how to get started:

  1. Install Hermes Agent on your machine or server. It is open source and free.
  2. Build or add agents with clear jobs, like research, writing, coding, or planning.
  3. Connect your models - local ones like Olimar or LM Studio, or cloud like Claude, GPT, Gemini.
  4. Open Curator and describe your task in plain words.
  5. Set it to show the plan first - review the agent list and steps, then approve before it runs.
  6. Run and give feedback - use thumbs up or down to train Curator for better picks.

For a full 30-day roadmap, Julian mentions the AI Profit Boardroom, which starts with day one as your first agent, day five as Curator setup, day 15 as Kanban, and day 30 as a full library.

04How to Use Hermes Agent Curator: Examples

Julian shows two real-life examples. First, say you run a small business and want a content plan for next month. You tell Curator.

It pulls a research agent to find trends, a writer to draft posts, a planner to map dates, and a reviewer to check work. You just press go and get a full month of content.

Second, you want to grow your email list. You drop one card on the Kanban board that says “grow my email list this month.” Curator breaks it into smaller cards for the lead magnet, landing page copy, email sequence, and social posts. Then it picks an agent for each and runs them.

You wake up and the work is done.

You can also use Curator with Kanban like Trello for AI agents. Each card is a job, each column is a stage, and agents move cards. With Curator, you do not even assign agents to cards; it just does it.

05Results and Speed Gains from Curator

Julian reports that people testing Curator say jobs that used to take 30 minutes now take five. The speed gain comes because the right agent is picked the first time, so there are no wasted runs or bad outputs.

For example, a small business content plan that might have taken a weekend with a freelancer or team can now be done with one prompt. Growing an email list that used to take a team or freelancer is now one card on a board.

Curator learns from your feedback, so results improve over time. The team behind Hermes is open about mistakes and built a feedback button to train the picker. In this episode, Julian emphasizes that the open-source nature means you can see the code and change it, which gives you control over how it works.

06Tips, Limits, and Mistakes to Avoid

Julian is clear that Curator is not magic. It is only as good as the agents you give it. If you have one agent, it picks that one every time.

With 10 well-built agents, it shines. So setup still matters.

Curator can pick an agent you did not expect. To stay safe, set it to show the plan before it runs. Review the agent list and steps, then hit approve.

If you skip that, you might get surprised by bad output. Julian recommends running the safe way for the first week, then letting it run alone once you trust it.

  • Watch token usage - Running multiple agents per task can rack up costs fast with paid models. Use local models to avoid this problem.
  • Give feedback - Use the thumbs down button when Curator picks a bad fit. It learns from that.
  • Build a good agent library - The more clear agents you have, the better Curator performs.
  • Do not skip plan review - Approving first prevents surprises.

07Who Hermes Agent Curator Helps Most

Hermes Agent Curator helps small business owners who found agent setup too hard. You do not need to be a coder or know which model works best for each job. You describe the work and Curator handles the rest.

If you already use AI agents, Curator saves hours every week by stopping tweaks and tests. You just ship work. For teams, one person can run the work of five because Curator does the picking and routing that used to be manual.

Julian says agents are about to become normal for everyone, not just coders. Curator hides the hard parts, like Gmail made email easy and Squarespace made websites easy. Coming features like memory across runs, cost tracking, and agent sharing will make it even faster.

08The Future of AI Agents and Curator

Julian talks about a bigger trend. AI agent tools are splitting into two camps: closed source that locks you in, and open source like Hermes. Curator makes the open camp stronger because now open tools are not just open—they are easy too.

The team behind Hermes is already working on Vero.14. They are adding memory across runs so Curator remembers what worked. They are adding cost tracking to show which agents and models cost the most.

They are adding agent sharing so you can use agents built by others.

In 90 days, Curator and similar tools will be normal. Early movers will have workflows built, agent libraries set up, and know which prompts work. Julian encourages getting your agents ready now so you are already running when it becomes the standard.

09Risks and Safety with Curator

Julian is honest about risks. Since Curator picks agents on its own, it can pick one you did not expect. That can be good, but also bad if you do not watch the output.

Always review the plan first for the first week.

Token usage is another risk. Running multiple agents per task can use tokens fast, especially with paid models. Using local models like Olimar or LM Studio avoids this problem because everything runs on your machine.

That is why many users pair Hermes with local models.

Julian suggests checking plans and watching outputs until you trust Curator. Once it proves reliable, you can let it run on its own. The open-source code means you can see exactly what it does and make changes if needed.