Where to go next
You finished the learning path. Here's a map of the rest of the territory.
If you worked through every step, you now know enough Hermes to be useful. You've installed it, configured a model, held real conversations, set up a personality, taught it to remember things, linked it to Telegram, installed skills, plugged in MCP, scheduled recurring jobs, seen Skills evolve on their own, and maybe even deployed to a server.
That's most of what a beginner needs. But Hermes has more depth than any single learning path can cover. Here are the directions worth exploring once you're comfortable.
Voice mode
Full voice interaction — press a key to record, get a spoken reply, join Discord voice channels, send voice notes in Telegram and get voice notes back. The speech-to-text and text-to-speech can run locally if you don't want audio going to the cloud.
Start with /voice on inside a chat.
Sandboxes: Docker, SSH, and friends
By default Hermes runs shell commands on your local machine. You can instead have it run commands inside a Docker container, over SSH on a remote box, or in more exotic environments like Modal, Daytona, or Singularity.
Useful when you don't want Hermes touching your primary machine, or when you want a clean, reproducible environment for every session.
Other messaging platforms
You did Telegram. The same hermes gateway also speaks Discord, Slack, WhatsApp, Signal, and email. The setup pattern is similar — create an account or bot on the platform, put the credentials in ~/.hermes/.env, and start the gateway.
Security and approval modes
By default Hermes asks before running a command. That gets annoying fast. Read up on approval modes (manual, smart, off) and pick what fits your risk tolerance. For a machine where you trust Hermes, smart is a good middle ground.
There's also a YOLO mode. Use at your own risk.
Delegation — sub-agents
Hermes can spawn sub-agents to work on pieces of a problem in parallel. Useful for anything that splits cleanly into independent chunks.
RL training
The Hermes team also ships tools for reinforcement learning — fine-tuning your own models using the agent itself as a training loop. This is a whole field, not a side feature. If you're model-curious, this is where Hermes gets really interesting.
Python library usage
Hermes isn't only a CLI. You can import it as a Python library and embed the agent in your own code. Useful when you want to build a custom application on top of Hermes instead of using its front ends.
The ACP editor integration
Hermes can plug into editors via the Agent Client Protocol. If you use an ACP-compatible editor, you can run Hermes directly inside it as a coding assistant.
The official docs
We deliberately kept this learning path narrow. For depth, the official docs at hermes-agent.nousresearch.com go much further. They're organized by topic (Getting Started, Features, Messaging Platforms, Integrations, Developer Guide, Reference). Now that you have a mental model, they'll be much easier to navigate.
Contributing
Hermes is open source under the MIT license. The repo is at github.com/NousResearch/hermes-agent. If you spot bugs, have ideas, or want to build skills or integrations, the project is active and takes pull requests.
One last thing
The best way to keep learning Hermes is to use it for something you actually need. Pick a small recurring annoyance in your week — a report you run, a file you reorganize, a summary you write — and try to get Hermes to do it. You'll learn more in one real use case than in a dozen tutorials.
Good luck.