01What Is Hermes Agent Goals and How It Works

Hermes Agent Goals is a new free update that lets you give your AI agent one persistent objective, and it will keep working step by step until the job is done. You type a goal like "build an SEO website" and Hermes runs a loop: it takes a turn, then a judge model checks if the goal is finished. If not, it keeps going automatically.

This is inspired by Ralph Loops, a popular automation method for Claude Code. The update was announced by one of the co-founders of Hermes. It turns Hermes from a tool you prompt constantly into an agent that works on its own.

You just set the goal, and Hermes does the thinking and the actions until completion.

In this episode, Julian Goldie shows how to set up and use this feature. He also shows how Hermes can deploy a full website while you watch. The goal loop keeps running until the judge model says the task is done, or until the turn budget is used up.

02Key Features of the Goals Update

The new Goals update brings several important features to Hermes Agent. Here are the main ones shown in this episode:

  • Autonomous goal loop: Hermes runs a loop with a supervisor model that checks if the task is completed after each turn.
  • Turn budget: You can set a maximum number of turns (default is 20) so the agent doesn't run forever.
  • Judge model: A separate AI model checks if the goal is achieved. You can change the provider and model in config.
  • Status commands: Use /goal status to see the current setup and progress.
  • Pause and resume: You can pause a goal, close your terminal, and resume it later. The goal continues where you left off.
  • Clear command: Use /goal clear to end a goal completely.

These features work on the command line (CLI) and also through the gateway via the central registry. The update is free—just run hermes update in your terminal to get it.

03How to Get Started with Hermes Agent Goals

Getting started with Hermes Agent Goals takes about two minutes. Here are the steps Julian shows in the episode:

  1. Open your terminal.
  2. Run hermes update to pull the latest updates from Hermes Agent.
  3. Type /goal followed by your goal text. For example: build a new blog about Hermes AI agent, optimize it for SEO, and continue posting daily until it's completed and ranking.
  4. Press enter. Hermes will initialize the agent with your current setup.
  5. Wait for Hermes to start working. You can see the plan and the steps being completed.
  6. Use /goal status to check progress, /goal pause to pause, /goal resume to resume, and /goal clear to stop a goal.

You can change the turn budget and the judge model in the config file. The default max turns is 20, but you can adjust it to whatever you need.

04How to Use Hermes Agent Goals: Examples

In the episode, Julian demonstrates a full use case. He sets a goal to build an SEO-optimized blog about Hermes itself. The agent creates a dark-themed, fast static website with meta tags, schema markup, sitemap, and semantic HTML.

Then it deploys the site to Netlify automatically.

You can also use goals for other tasks. Julian lists examples like: fix every error in a particular topic, port feature X from repo Y, investigate why something isn't working, and build a small CLI to rename files. The idea is to give a clear, specific, and measurable goal.

While the goal is active, you can send normal messages, and those take priority over the loop. After each turn, the judge model checks if the goal is done. If not, the loop continues.

When the goal is complete, you'll see "goal achieved" or "goal paused" if you stop it.

05Results and Tests Shown in the Episode

Julian runs a live test of the feature. He sets the goal for the SEO blog, and the agent begins working immediately. It creates a plan and lists the tasks it needs to complete.

The agent writes the HTML files and deploys the website to Netlify.

You can watch the agent loop over its status checks, showing that it is building the site step by step. The episode shows the agent creating multiple blog posts as part of the autonomous process. Julian notes that the judge model checks if the goal is done after each turn, and if not, the loop continues.

He also shows the turn budget in action. The agent has 20 turns to finish the job. If the budget runs out, the goal stops unless you increase it.

The feature is relatively new, so Julian advises testing it out on a big project and letting it run while you do other things.

06Tips, Limits, and Mistakes to Avoid

To get the most out of Hermes Agent Goals, keep these tips in mind. First, write specific, clear, defined, and measurable goals. The better the goal, the better the agent's work.

Vague goals might lead to incomplete or off-target results.

Second, understand the turn budget. If your task is big, you may need to increase the max turns in config. Otherwise, the loop will stop too early.

Third, you can change the judge model in config. Using a stronger model might improve judgment, but it could cost more.

There are a few limitations to remember. The feature is new, so there could be bugs. If you pause a goal, it respects the pause, but you must resume it manually.

If the budget is exhausted, the goal stops, not because it's done but because it ran out of turns. To avoid mistakes, check your config and set a realistic budget from the start.

07Who Should Use Hermes Agent Goals and Why It Matters

Hermes Agent Goals is for anyone who uses AI agents for multi-step projects. If you create content, do research, write code, or build websites, this feature saves you time. Instead of prompting the agent every few minutes, you set one goal and let it work.

This is especially useful for developers and marketers who want to automate repetitive tasks. The ability to close your laptop and resume later makes it very flexible. You can start a goal, go offline, and come back to see the progress.

It matters because it makes AI agents more autonomous and practical for real work. You can get a full website built or a research report done with a single command. Julian also mentions that a 30-day roadmap and example prompts are available inside the AI Profit Boardroom, his community for AI automation users.