01What is Hermes Agent V0.18?
Hermes Agent V0.18 is the latest update to the Hermes AI agent, also called 'The Judgment Release.' It adds new systems like Mixture of Agents, improved goal mode, and a /learn command that lets the agent pick up new skills quickly.
This update focuses on making AI agents more autonomous and reliable. You can combine multiple models for better answers, verify work with a judge, and let the agent learn from guides automatically. If you use Hermes in a terminal or an agent operating system, these features change how you work with AI.
02Main Features in Hermes Agent V0.18
The update includes several headline features. The biggest is Mixture of Agents, a system that fuses answers from multiple models like Claude Opus 4.8 and GPT 5.5 to create higher-quality outputs. It works in one shot, taking about 10–15 minutes to return a final answer.
- Mixture of Agents: combines multiple models for better results.
- /goal: an autonomous loop with a builder and judge API that iterates until work meets your definition of done.
- /learn: Hermes learns a guide or repository and saves it as a skill.
- /journey: a playable timeline of everything Hermes has learned about you.
- Proof of work: Hermes runs project checks and records evidence before declaring tasks complete.
- Every model's reasoning is visible after answers come back.
You can also update Hermes in three ways: say 'update to the latest version' in chat, click 'update Hermes' on your dashboard, or type 'hermes update' in the terminal.
03How to Set Up Hermes Agent V0.18
Setting up the new features is straightforward. Start by ensuring you have the latest version. Then explore the new tools from the chat or your agent operating system.
- Update Hermes using one of three methods: chat command, dashboard button, or terminal 'hermes update'.
- Open your agent operating system (if you use one) to access the full interface with profiles, Oracle, and other tools.
- Select Mixture of Agents from the model picker to use the multi-model system.
- Use /goal to start a goal mode, defining your builder API, judge API, and number of rounds.
- Use /learn and provide a link to a guide or repository to teach Hermes a new skill.
- Use /journey to view your timeline of learned information and past interactions.
If you prefer the terminal, all features work there too, but the operating system makes it easier to see and manage everything.
04Using Hermes Agent V0.18: Examples from the Episode
In the episode, Julian showed practical examples of the new commands. For Mixture of Agents, he gave a prompt and the system combined answers from Claude Opus 4.8 and GPT 5.5. It created an OS, a Minecraft alternative, and games—all in one go.
The output appeared in the workspace after 10–15 minutes.
For /goal, he explained that you give Hermes a task, and a separate judge checks the work. If the quality is low, the loop repeats for as many rounds as you set, without needing your input. This saves time on quality control.
For /learn, he fed a link to a guide and typed '/learn'. Hermes returned a full breakdown of what it learned and saved it as a skill for future use. He also tried /journey, which showed a timeline of everything Hermes had learned about him.
He also showed the delegate tasks feature, where you can give a task like 'run the SEO funnel for how to build an AI agent' and multiple sub-agents work in parallel automatically.
05Results and Comparisons from Goldie Bench
Julian compared Hermes Mixture of Agents against other models on his Goldie Bench. The Hermes Mixture of Agents scored at the top, outperforming Fable 5 in many ways. He noted that Fable 5 was being tested that day, but from what he saw, Hermes Mixture of Agents produced better results.
The system is also cheaper and more efficient according to the update notes. Julian observed it getting faster with each update. He also highlighted that the proof-of-work feature prevents false completion claims—Hermes now runs actual project checks and records evidence before saying 'done.'
For goal mode, the completion evidence system ensures the loop keeps judging work against your stated definition of done, rather than stopping when the model thinks it's finished.
06Tips, Limits, and Mistakes to Avoid
One tip from the episode: using Hermes in a terminal makes it hard to see everything you've created. Julian recommends an agent operating system for better visibility and management. He also suggests using Obsidian, a free app, to visualize your memory and recent conversations with Hermes.
For Mixture of Agents, remember it's one-shot—you wait 10–15 minutes for an answer, so plan accordingly. It's not like Claude CLI where you go back and forth. Also, define your 'definition of done' clearly in goal mode to avoid endless loops.
Be careful not to rely on the model's own claim of success. The new proof-of-work system helps, but always check the evidence it records.
07Who Should Use Hermes Agent V0.18?
Hermes Agent V0.18 is for anyone using AI agents seriously, from SEO professionals to content creators. Julian mentions that even non-technical users, including himself, can get the most out of it. In the AI Profit Boardroom, there are tutorials and systems built around Hermes, with 3,900 members getting support 24/7.
If you have tasks that need a team of agents—like writing five articles in parallel or managing lead generation—the delegate tasks feature is a big win. The update makes Hermes more reliable and efficient, which matters for real-world projects.
Whether you're a beginner or advanced, these tools let you automate complex workflows and save time on quality control.
08Where to Get More Help and Resources
If you want the full agent operating system Julian uses, it's available in the AI Profit Boardroom. You can find link in the comments or visit ai-profit-boardroom.com. There are tutorials for Mixture of Agents, /learn, and email/lead generation agents.
The community offers video tutorials, a six-week beginner-to-expert course, weekly coaching calls, and a map to meet local AI builders. Julian personally answers questions daily with video explanations.
