Your First Week with Hermes

What You Have Built So Far

Across eight lessons you have installed Hermes, connected a model provider, learned the CLI, understood configuration and profiles, given your agent an identity, made it aware of a real project, learned to inject files and diffs directly into a conversation, and taught it to remember. None of that was abstract -- every lesson had something to actually run. This lesson is the one place those pieces get used together, in order, as a single exercise.

The Walkthrough

Do this with a real project on your machine, not a throwaway example -- the point is to end this lesson with something genuinely useful, not just a demo.

1. Pick a project. Choose a directory you actually work in regularly.

2. Confirm your SOUL.md is the way you want it. From Lesson 5: open ~/.hermes/SOUL.md and re-read it. Does it still sound like how you want this agent to talk to you? Edit if not.

3. Write or check that project's AGENTS.md. From Lesson 6: at the root of your chosen project, make sure there is an AGENTS.md covering architecture, conventions, restrictions, and key paths. If you wrote one in Lesson 6 for a different project, write a fresh one specific to this one.

4. Start a fresh session inside that project directory.

cd /path/to/your/project
hermes --tui

Ask it directly: "What do you know about this project?" You should hear back something that reflects what you wrote in AGENTS.md, without having explained anything yourself in this conversation.

5. Point it at something real with @ syntax. From Lesson 7: reference an actual file or your current uncommitted changes:

What does @file:src/main.py do, and are there any obvious issues with it?

What changed? @diff

6. Teach it something about you, deliberately. From Lesson 8: tell it one real preference -- something you would actually want it to remember going forward, not a throwaway test fact.

Remember that I prefer terse commit messages, one line, no body text unless I ask for one.

7. Close the session and verify memory survived. This is the step people skip, and it is the one that actually proves the system works. Start a completely new session and ask:

What do you know about how I like commit messages written?

If the new session answers correctly without you re-explaining, every layer is working together: identity, project context, and persistent memory.

If Something Did Not Work

A short digest of the most common early issues, drawn from Hermes' own troubleshooting documentation:

• Rate limited (HTTP 429 errors). Wait and retry, or check whether your provider plan has a usage cap you are hitting.
• "Context length exceeded." Your conversation has grown too large for the model's window. Run /compress to summarize older history, or switch to a model with a larger context window via /model.
• Model not responding or erroring immediately. Double-check the model identifier matches your provider's exact naming -- these have to match precisely, a near-miss spelling will fail rather than fall back to something close.
• A terminal command got blocked and asked for approval. This is expected, working safety behavior, not a bug. Hermes intentionally requires approval for destructive-looking commands (recursive deletes, table drops, and similar). You will go deeper on the full approval system in Hermes Agent Advanced.
• Memory did not show up in a new session. Re-read the frozen-snapshot section of Lesson 8 -- a write made mid-conversation will not appear until the conversation after the one where it was written. If you tested by asking in the same session, that is expected; the real test is a fresh session afterward.

Where You Are Now

You have a real, working, personalized Hermes agent. It has a consistent voice, it understands at least one of your actual projects without re-explanation, you know how to pull specific files and diffs directly into a conversation, and it remembers things about you across sessions rather than starting from zero every time. That is a genuinely useful tool already -- many people stop here and get real value from exactly this setup for a long time.

Hermes Agent Advanced is where this stops being something you have to actively open a terminal to use. It covers scheduling your agent's work so it runs without you (cron), connecting it to Telegram, Discord, Slack, and twenty more platforms so you can reach it without a terminal at all, extending what it can do with tools, skills, and external MCP servers, giving it voice and the ability to browse the web and generate images, coordinating more than one agent at once, and the security and production practices that matter once an agent is acting on its own rather than only when you are watching.