Until recently, most of us used AI by opening a website, asking a question, receiving an answer, and then closing the conversation. The AI could write, explain, brainstorm, and answer questions—but it usually waited for us to direct every step.
AI agents are different.
An AI agent can be given a goal, decide which tools are needed, work through several steps, interact with files, organize information, remember useful details, and return with a completed result.
In my previous article, Choosing My Offline AI Model: Qwen3.6-27B, I explained why I chose Qwen3.6-27B as my preferred local AI model.
But the model alone is only the brain.
To give that brain tools, memory, skills, project access, and the ability to perform real work, we need an agent system around it.
That is where Hermes Agent comes in.
In this beginner-friendly guide, we will explore:
What Hermes Agent is.
Why it is different from a normal chatbot.
Why local and offline AI matters.
How to install Hermes through its desktop application.
How to connect Qwen3.6-27B using LM Studio or Ollama.
How Hermes remembers information.
Whether connecting Hermes to Obsidian is worthwhile.
What beginners can do with Hermes.
How to begin safely without becoming overwhelmed.
No advanced programming knowledge is required.
What Is Hermes Agent?
Hermes Agent is a free, open-source AI agent developed by Nous Research.
The easiest way to understand it is:
A chatbot gives you an answer. An agent can help you complete a job.
You can still talk with Hermes like a regular AI assistant, but Hermes can also use tools and take multiple steps toward a goal.
For example, Hermes can potentially:
Read and summarize documents.
Create, edit, rename, and organize files.
Search through folders and projects.
Research information on the web.
Use browser automation.
Run terminal commands.
Help write or repair software.
Create scheduled tasks.
Work through supported messaging applications.
Connect to external services and tools.
Remember your preferences and projects.
Search previous conversations.
Learn reusable workflows.
Turn repeated procedures into skills.
Delegate parts of a complicated task.
Work with an Obsidian knowledge vault.
Help manage long-running creative or technical projects.
Hermes is not an AI model. Instead, it is the system that gives an AI model access to memory, tools, skills, files, and working environments.
Think of the relationship like this:
Qwen3.6-27B is the brain.
Hermes Agent provides the tools, memory, skills, and working system.
Hermes Desktop is the friendly application through which we communicate with it.
LM Studio or Ollama loads and serves the local model.
Nous Research describes Hermes as a self-improving agent with a built-in learning loop. It can create skills from experience, improve workflows during use, preserve useful knowledge, search past sessions, and gradually develop a better understanding of the person using it.
You can read the complete description in the official Hermes Agent documentation.
What Is New in the Latest Hermes Agent Release?
At the time of writing, the latest stable release is Hermes Agent v0.20.5, released on August 19, 2026.
This release arrived during a very active period of Hermes development. Recent improvements include:
Better desktop performance.
A more convenient model picker.
Drag-and-drop support for PDFs and files.
Improved conversation summaries.
Better protection against stalled agent tasks.
More reliable installation and update behavior.
Improvements to Bot Mode and group conversations.
Better command-line controls.
Persistent memory for scheduled tasks.
Separate reasoning settings for scheduled jobs.
Easier web-search availability for new installations.
Hundreds of fixes across the desktop app, tools, providers, and agent runtime.
Beginners do not need to understand every technical change. The important point is that Hermes is developing from a terminal-focused agent into a much more accessible desktop workspace without losing the powerful system underneath.
The detailed technical changes can be found on the official Hermes Agent releases page.
How Is Hermes Different from ChatGPT or a Normal Local Chat App?
A normal local AI application usually lets you download a model and chat with it.
That is already useful, but the model may not be able to do much outside the conversation window.
Hermes adds an agent layer.
Instead of only saying:
Here is how you could organize your folder.
Hermes may be able to inspect the folder, propose a structure, ask for approval, and then organize the files.
Instead of only saying:
You should keep a project summary.
Hermes can create the summary, save it inside the project, and update it later.
Instead of only explaining how to repeat a task, Hermes can help turn the procedure into a reusable skill.
The important difference is the movement from answering to working.
Of course, this also means Hermes must be used carefully. An agent with access to files and commands is more capable than a chatbot, but it also requires clear permissions, sensible boundaries, and human supervision.
Why Is Hermes Agent Important?
Modern AI is becoming part of everyday creative and professional life.
People now use AI to help with:
Writing and publishing.
Research.
Programming.
Design and creative development.
Business administration.
Personal organization.
Education.
Communication.
Data analysis.
Long-term project management.
Most popular AI services operate entirely in the cloud. This is convenient, but it creates several limitations:
They may require monthly subscriptions.
Usage may be restricted.
Private information may leave your computer.
They may stop working during internet outages.
The company can change its pricing or policies.
Models can be removed or replaced.
You do not fully control the system.
Hermes provides another path.
You can connect Hermes to powerful cloud models when you want maximum intelligence, but you can also connect it to a local open-weight model running on your own computer.
This makes Hermes a bridge between modern AI agents and personal, locally controlled computing.
Why Use an Offline AI Model Today?
Running Qwen3.6-27B locally gives us several important advantages.
Greater Privacy
When the model and tools are running locally, your prompts and local files can remain on your own computer.
This can be valuable when working with:
Personal journals.
Unpublished books.
Business documents.
Private research.
Source code.
Creative plans.
Client projects.
Personal knowledge archives.
You must still pay attention to any online tools you enable. A web search, cloud integration, or remote service naturally sends some information over the internet.
However, the main model conversation can remain local.
No Per-Message AI Cost
Once the model has been downloaded, you are not paying for every prompt or token.
Your main costs become the computer hardware and electricity required to run it.
Independence
A local model can continue working when:
Your internet is unavailable.
A cloud AI service is down.
A subscription expires.
A provider changes its limits.
Online AI access becomes expensive or restricted.
This is especially meaningful in places where internet access is slow, unstable, limited, or costly.
Personal Control
The model files are stored on your computer. You choose when to update, replace, or remove them.
Your Hermes memory, project files, workflows, and Obsidian notes can also remain under your control.
A Long-Term AI Workspace
A local model does not have to remain a temporary chatbot.
Combined with Hermes, it can become part of a personal workspace containing:
Your preferences.
Active projects.
Working rules.
Reusable skills.
Research notes.
Creative guidelines.
Project histories.
Scheduled routines.
The model itself can be replaced later while much of the surrounding system remains useful.
What “Offline” Does Not Mean
Offline AI does not mean every Hermes feature works without the internet.
The following can work locally:
The Qwen3.6-27B model.
Hermes’s internal memory.
Local file tools.
Terminal tools.
Local project work.
Obsidian vault access.
Many installed skills.
These normally require internet access:
Web search.
Online research.
Cloud applications.
Messaging platforms.
Remote APIs.
Online model providers.
Services such as GitHub, Google Drive, or social platforms.
You remain in control of when Hermes works entirely locally and when it is permitted to use online services.
Understanding the Complete Local Setup
Before installing anything, it helps to understand how the pieces connect:
You
↓
Hermes Desktop
↓
Hermes Agent
↓
LM Studio or Ollama
↓
Qwen3.6-27BYou communicate through Hermes Desktop.
Hermes Agent understands the goal and decides whether it needs memory, files, commands, skills, or another tool.
Hermes sends the thinking request to LM Studio or Ollama.
LM Studio or Ollama runs Qwen3.6-27B on your computer and returns its response to Hermes.
This separation is useful because you can change the model later without rebuilding your entire Hermes environment.
Installing Hermes Agent with the Desktop App
Hermes can be used through a terminal, but beginners should start with Hermes Desktop.
The desktop application uses the same Hermes core, configuration, sessions, skills, and memory as the terminal version. It is not a lightweight imitation. It is the full agent presented through a friendlier visual interface.
Hermes Desktop supports Windows, macOS, and Linux. The graphical installer is especially convenient for Windows and macOS users.
Step 1: Download Hermes Desktop
Visit the official Hermes Agent website.
Select:
Download Hermes Desktop AppChoose the correct version for your operating system:
Windows 10 or 11.
macOS 12 or newer.
Linux using the provided installation instructions.
Only download Hermes from the official website or official Nous Research GitHub repository. Avoid unknown mirrors or third-party installers.
Step 2: Run the Installer
Open the downloaded installer and follow the normal installation process.
During the first launch, Hermes Desktop may prepare the Hermes Agent runtime and supporting components.
This is normal. The application is installing the agent system that runs behind the graphical interface.
The official Hermes installation guide contains the current platform instructions.
Step 3: Open Hermes Desktop
Hermes opens into a chat-centered workspace.
Depending on the current release, the application allows you to:
Start and manage conversations.
Choose models.
Attach files.
Browse projects and folders.
Inspect created artifacts.
View tool activity.
Manage settings and profiles.
Configure skills and connections.
Monitor the model’s context usage.
Work across multiple sessions.
Hermes Desktop automatically manages its own local Hermes backend, so beginners do not normally need to open a separate web dashboard.
Step 4: Choose a Model Provider
Hermes needs an AI model before it can think.
You can connect it to:
Nous Portal.
Supported commercial AI providers.
OpenRouter.
Another cloud API.
A local model through LM Studio.
A local model through Ollama.
Another OpenAI-compatible local server.
For this guide, we will connect our local Qwen3.6-27B model.
Two Ways to Run Qwen3.6-27B
Hermes does not load the large model file directly.
Instead, Qwen3.6-27B is loaded by a local model runner. Hermes communicates with that runner through a local API.
For our setup, you can choose between:
LM Studio
Choose LM Studio if you:
Prefer buttons and menus.
Want a graphical model browser.
Like adjusting settings visually.
Want to see whether the model server is active.
Regularly test different GGUF models.
Are completely new to local AI.
Ollama
Choose Ollama if you:
Are comfortable entering a few simple commands.
Want the model to run as a background service.
Prefer a lightweight setup.
Want easy integration with scripts and agents.
Do not need a detailed graphical model-management interface.
You only need one.
However, LM Studio and Ollama can both remain installed on the same computer. Just avoid loading two copies of Qwen3.6-27B simultaneously, because that can consume a large amount of RAM and VRAM.
Option One: Connecting Hermes to LM Studio
LM Studio is my recommended starting point for non-technical users because its settings are easier to see.
Step 1: Install LM Studio
Download LM Studio from its official website and complete the normal installation.
Open it after installation.
Step 2: Add Qwen3.6-27B
If you followed my Qwen3.6-27B model guide, use the same model and quantization selected there.
Otherwise, search for the appropriate Qwen3.6-27B GGUF model inside LM Studio.
A quantized model is a compressed version that uses less memory. Smaller quantizations are easier to run, while larger quantizations can preserve more quality but require more RAM or VRAM.
Step 3: Load the Model
Select the downloaded Qwen3.6-27B model in LM Studio.
Before loading it, locate the context-length setting and set it to at least:
65536This gives Hermes approximately 64K tokens of context.
Context is the model’s temporary working space. Hermes needs a large context because its instructions, tools, memory, project information, and conversation must all fit within it.
A 64K context also consumes additional memory. If Qwen3.6-27B stops fitting on your hardware, you may need:
A smaller model quantization.
More CPU offloading.
More system RAM.
Fewer GPU-loaded layers.
A more memory-efficient model runner configuration.
Step 4: Start the LM Studio Server
In LM Studio:
Open the Developer section.
Select Qwen3.6-27B.
Confirm the context length.
Select Start Server.
LM Studio normally uses this local address:
http://localhost:1234/v1Keep LM Studio open and leave the server running while Hermes is using the model.
Step 5: Configure Hermes
Open a terminal and run:
hermes modelChoose:
Custom endpoint (self-hosted / VLLM / etc.)Enter:
API base URL: http://localhost:1234/v1
API key: lm-studio
Model name: Use the exact model identifier displayed by LM Studio
Context length: 65536The API key is only a placeholder for a normal local LM Studio server. It is not a paid cloud key.
Save the configuration and open a new Hermes session.
Step 6: Test the Connection
Ask Hermes:
Reply with a short greeting, tell me which model you are using, and do not use any tools.
Watch the LM Studio server panel. You should see the request arrive.
Then try:
Create a text file named “Hermes Test” containing five beginner-friendly ideas for using a private local AI assistant.
Hermes should use its file tool while Qwen3.6-27B provides the reasoning.
Option Two: Connecting Hermes to Ollama
Ollama is a convenient choice when you want your local model server to operate quietly in the background.
Step 1: Install Ollama
Download Ollama from its official website and complete the installation.
On Windows, Ollama normally starts as a background application.
Open PowerShell or a terminal and check that it is installed:
ollama --versionStep 2: Find Your Model’s Exact Name
If you already added Qwen3.6-27B to Ollama, run:
ollama listThis displays the models available on your computer.
Copy the exact name Ollama shows for Qwen3.6-27B. The identifier may differ depending on how the model was downloaded or imported.
Hermes must use the same model name that Ollama recognizes.
Step 3: Configure a 64K Context
Hermes currently requires at least 64,000 tokens of context for agent workflows.
Ollama may automatically select a smaller context according to the available hardware. We must therefore make sure Ollama itself—not only Hermes—is configured for 64K.
One way to launch Ollama with the required context is:
OLLAMA_CONTEXT_LENGTH=64000 ollama serveThe permanent setup differs between Windows, macOS, and Linux. Follow the current official Hermes and Ollama guide for your operating system.
After the model is loaded, check the active context:
ollama psLook at the CONTEXT column. It should report approximately 64K or more.
This is important. Entering 65,536 in Hermes does not automatically force Ollama to provide that amount. The server must actually be configured for it.
Step 4: Make Sure Ollama Is Running
Ollama normally provides its local API at:
http://localhost:11434/v1On many Windows installations, the background Ollama application starts the service automatically.
If necessary, start it with:
ollama serveKeep the Ollama service running while Hermes is using the model.
Step 5: Configure Hermes
Run:
hermes modelChoose:
Custom endpoint (self-hosted / VLLM / etc.)Enter:
API base URL: http://localhost:11434/v1
API key: ollama
Model name: The exact Qwen3.6-27B name shown by “ollama list”
Context length: 65536Ollama does not normally need a real API key for local access. The word ollama can be used as a harmless placeholder if Hermes requires a value.
Save the configuration and begin a new Hermes session.
Step 6: Test Ollama
Ask Hermes:
Reply with a short greeting and tell me which local model you are using.
Then try:
Examine the current folder without changing anything. Explain what kinds of files it contains.
Hermes should use Qwen3.6-27B for reasoning and its file tools to inspect the folder.
LM Studio or Ollama: Which Should You Choose?
| LM Studio | Ollama |
|---|---|
| Friendly graphical interface | Lightweight background service |
| Easier for complete beginners | Requires a few commands |
| Excellent for testing GGUF models | Convenient for regular agent use |
| Visual context and GPU settings | Simple API integration |
| Visible server activity | Works well with scripts |
| Best for experimentation | Best for quiet everyday operation |
My beginner recommendation is:
Start with LM Studio.
Learn how the model, context, and local server work.
Try Ollama later if you want a cleaner background service.
Neither program automatically makes Qwen3.6-27B more intelligent. They are different systems for loading and serving the model.
Performance can vary according to:
Model format.
Quantization.
Context length.
GPU offloading.
Available VRAM.
Available system RAM.
Server configuration.
You can keep both applications installed and switch Hermes between these endpoints:
LM Studio: http://localhost:1234/v1
Ollama: http://localhost:11434/v1Avoid running Qwen3.6-27B in both applications simultaneously.
The Model Must Support Tool Calling
Hermes is more than a chat interface. It expects the model to produce structured tool calls.
Tool calling allows the model to ask Hermes to:
Read a file.
Search a folder.
Edit a document.
Run a command.
Open a webpage.
Search the internet.
Create a scheduled task.
Save something to memory.
A local model may chat successfully but struggle with tools if:
Its chat template is incorrect.
Tool calling is disabled.
The local server has limited tool support.
The context window is too small.
The model is too heavily compressed.
The selected model was not trained for agentic work.
Test the connection gradually:
Ask a normal question.
Ask Hermes to list files without changing them.
Ask it to read one harmless file.
Ask it to create a test file.
Ask it to complete a small multi-step task.
If normal chatting works but tool use fails, the problem may be the model template or local server configuration rather than Hermes itself.
Hermes officially supports custom and self-hosted models through OpenAI-compatible endpoints, as explained in its AI provider documentation.
How Hermes Memory Works
Hermes includes persistent memory without needing Obsidian or another memory service.
Its built-in memory uses two small files.
MEMORY.md
This contains the agent’s working knowledge, such as:
Environment details.
Project conventions.
Important lessons.
Repeated procedures.
Things Hermes has learned while working.
USER.md
This contains useful information about the user, such as:
Preferences.
Communication style.
Expectations.
Stable personal information.
Preferred ways of working.
These memory files persist across sessions.
At the beginning of a new session, Hermes loads a snapshot of the memory into the model’s instructions.
You could say:
Remember that I prefer beginner-friendly explanations without unnecessary technical language.
Or:
Remember that Morning Glory is my creative identity and that my blog covers art, AI, local technology, and creative independence.
Hermes’s built-in memory is intentionally small and curated.
According to the current documentation:
MEMORY.mdhas a limit of approximately 2,200 characters.USER.mdhas a limit of approximately 1,375 characters.
This may sound small, but it keeps the most important information short enough to be loaded into every session.
If internal memory became an enormous diary, it would consume context, distract the model, and make every conversation slower.
You can read the technical details in the official Hermes persistent-memory guide.
Does Hermes Remember Every Conversation?
Hermes’s curated memory does not contain every sentence you have ever written.
That would be inefficient and undesirable.
Instead, Hermes can preserve important facts in memory and search past sessions when it needs more detail.
This creates several layers:
Current conversation: What you are discussing now.
Session history: Records of earlier conversations.
Internal memory: Important facts that should remain easily available.
Project context: Instructions and knowledge stored with a project.
Skills: Reusable methods for completing particular jobs.
External knowledge: Detailed notes stored in Obsidian or another memory provider.
It helps to tell Hermes explicitly when something matters:
This is an approved project decision. Please remember it.
You can also tell it what not to preserve:
This is a temporary experiment. Do not add it to long-term memory.
Periodically ask:
Show me what you currently remember about me. Identify anything outdated, duplicated, unnecessary, or missing. Ask before changing it.
Should We Connect Hermes to Obsidian?
Yes—but Obsidian should serve a different purpose from Hermes’s internal memory.
Obsidian should not replace the built-in memory. It should become the agent’s detailed personal knowledge library.
A useful comparison is:
Hermes memory is the small notebook it carries everywhere.
Obsidian is the organized library it visits when more information is needed.
Conversation history records what was previously discussed.
Project files contain the materials for the active job.
Skills contain reusable working methods.
Hermes includes a bundled Obsidian skill that can:
Read notes.
Search notes.
Create notes.
Edit notes.
Append information.
Add links between related notes.
Obsidian files use Markdown, a simple text format. The information remains readable and editable even without Hermes.
This makes Obsidian useful as a transparent, local-first knowledge system that both you and your agent can understand.
A Beginner-Friendly Obsidian Structure
Create an Obsidian vault named:
Morning Glory AI MemoryInside it, create these folders:
00 - Inbox
01 - About Me
02 - Active Projects
03 - Creative Knowledge
04 - AI Models and Tools
05 - Tutorials and Workflows
06 - Decisions
07 - Session Summaries
08 - Reference
Archive00 – Inbox
Use this for new information that has not yet been organized.
Hermes can periodically review the Inbox and suggest where each note belongs.
01 – About Me
Store stable information such as:
Your preferred name.
Communication preferences.
Creative identity.
Long-term goals.
Languages.
General working style.
Do not use it to store passwords, banking details, private keys, or highly sensitive personal information.
02 – Active Projects
Create one folder or master note for each active project.
Examples:
A book.
A blog.
An art collection.
A music project.
A virtual-world project.
A business idea.
An AI experiment.
Each project note can contain:
Purpose.
Current status.
Important decisions.
Next steps.
File locations.
Approved references.
03 – Creative Knowledge
Store:
Writing principles.
Art direction.
Visual styles.
Character rules.
Worldbuilding information.
Branding guidelines.
Creative references.
04 – AI Models and Tools
Store:
Qwen3.6-27B settings.
LM Studio settings.
Ollama settings.
Hermes configuration.
Hardware information.
Successful model parameters.
Troubleshooting solutions.
05 – Tutorials and Workflows
Store repeatable processes such as:
How to prepare a blog post.
How to organize a book project.
How to review a manuscript.
How to prepare an image for printing.
How to back up a completed project.
How to test a new local model.
06 – Decisions
Save important decisions and why they were made.
This prevents you and the agent from repeatedly debating choices that have already been settled.
07 – Session Summaries
Save concise summaries of important Hermes sessions.
Do not copy entire conversations unless there is a real reason. Preserve the results, decisions, and next steps.
08 – Reference
Store useful information that is not currently part of an active project.
Archive
Move completed, replaced, or outdated notes here instead of deleting them immediately.
Connecting the Obsidian Vault to Hermes
Hermes’s bundled Obsidian skill uses a setting called:
OBSIDIAN_VAULT_PATHThis tells Hermes where the vault is stored.
On Windows, an example might look like:
OBSIDIAN_VAULT_PATH=C:\Users\YourName\Documents\Morning Glory AI MemoryReplace YourName with your actual Windows username and use the real location of your vault.
The recommended place to define this is Hermes’s .env configuration file.
After configuring the path, tell Hermes:
Use my Obsidian vault as a long-term knowledge library. Save only information that is genuinely useful for future work. Store stable personal information in About Me, project knowledge in Active Projects, reusable procedures in Tutorials and Workflows, and important choices in Decisions. Do not store passwords, API keys, banking information, or temporary conversation details.
Hermes’s bundled skill and vault-path convention are described in the official Obsidian skill documentation.
Is Obsidian Necessary?
No.
Start with Hermes’s built-in memory.
Add Obsidian when your projects, research, creative rules, and technical notes become too detailed for the small internal memory.
Obsidian is especially valuable for people managing several long-term projects, such as:
Books.
Illustrations.
AI experiments.
Blog articles.
Music.
Art collections.
Business plans.
Virtual environments.
Technical setups.
The important thing is not to save everything.
A useful knowledge library contains selected information that will help you later. An unfiltered archive of every conversation quickly becomes clutter.
Ask Hermes to save:
Important decisions.
Stable preferences.
Final procedures.
Approved creative rules.
Current project status.
Lessons learned from mistakes.
Information likely to be useful again.
What Can Beginners Do with Hermes?
The best way to understand Hermes is to give it small, useful jobs.
Introduce Yourself
Try:
Interview me with ten simple questions so you can understand my goals, projects, preferences, and the kind of help I need. Afterward, suggest what is worth adding to your internal memory. Ask before saving anything.
Organize a Project
Try:
Examine this project folder without changing anything. Explain what it contains in simple language and recommend a clearer structure.
Summarize Documents
Try:
Read these documents and create a clear summary. Identify contradictions, missing information, and important decisions. Do not edit the original files.
Create a Blog Workflow
Try:
Help me create a repeatable workflow for researching, writing, illustrating, fact-checking, and publishing a blog post. Save the final approved workflow to my Obsidian vault.
Organize an Obsidian Inbox
Try:
Review the notes in my Obsidian Inbox. Suggest where each note belongs, but ask for approval before moving or rewriting anything.
Research a Subject
When connected to the internet:
Research this subject using reliable primary sources. Separate confirmed facts from opinions, explain it in beginner-friendly language, and provide direct links.
Learn a Repeated Task
Try:
I am going to teach you how I prepare an art product for sale. Observe the steps, ask questions when something is unclear, and afterward create a reusable checklist.
Create a Project Brief
Try:
Ask me the questions needed to turn this idea into a clear project brief. Keep the language simple and identify any decisions I still need to make.
Compare Files
Try:
Compare these two versions of the document. Explain what changed, identify anything that may have been lost, and do not edit either file.
Create a Scheduled Routine
After you become comfortable with Hermes:
Every Friday, review my active-project notes and prepare a short summary of progress, unfinished work, and suggested next steps.
Scheduled tasks require Hermes’s supporting services to be running at the scheduled time.
A Sensible First-Week Plan
Day 1: Install and Test
Install Hermes Desktop.
Connect Qwen3.6-27B through either LM Studio or Ollama.
Confirm that:
The model responds.
The context is at least 64K.
A simple read-only tool call works.
A harmless test file can be created.
Day 2: Introduce Yourself
Tell Hermes about:
Your preferred name.
Your work.
Your communication preferences.
Your current projects.
The type of help you want.
Ask Hermes to suggest what belongs in internal memory.
Day 3: Create One Safe Workspace
Choose one test or project folder.
Allow Hermes to inspect it without making changes.
Ask for an explanation of what the folder contains and how it could be improved.
Day 4: Add Obsidian
Create the vault structure described in this guide.
Connect the vault path to Hermes.
Save one project overview and one approved workflow.
Day 5: Explore Skills
Ask:
Which bundled Hermes skills are most useful for my creative work, writing, research, local AI, and project organization? Explain them without installing or configuring anything yet.
Choose only the skills you genuinely need.
Day 6: Teach One Workflow
Choose a task you repeat regularly.
Teach Hermes the process, correct its misunderstandings, and ask it to create a final checklist or skill.
Day 7: Review Memory and Safety
Ask:
Show me what you currently remember about me and my projects. Identify anything outdated, duplicated, unnecessary, uncertain, or missing. Ask before making changes.
Also review which folders and services Hermes can access.
Safety Rules for New Hermes Users
An AI agent can perform actions, so it deserves more caution than a normal chatbot.
During your first weeks:
Keep approval prompts enabled.
Leave YOLO mode turned off.
Begin inside a dedicated test folder.
Ask Hermes to inspect before editing.
Request a plan before large changes.
Keep backups of important work.
Never store passwords in Obsidian notes.
Never give the agent private API keys through normal chat.
Do not allow it to publish or send anything without approval.
Do not let a newly tested local model delete files.
Review terminal commands before approving them.
Give each important project a clearly defined workspace.
A useful permanent instruction is:
Work only inside the project folder I provide. Explain your plan before making significant changes. Ask for approval before deleting, overwriting, installing, purchasing, publishing, uploading, sending, or sharing anything.
The latest Hermes Desktop includes a per-session YOLO mode.
YOLO mode bypasses dangerous-command approval prompts. It may be convenient for experienced users working in disposable environments, but beginners should leave it disabled.
The goal is to let Hermes help—not to give an unfamiliar model unrestricted control over the computer.
Common Problems
Hermes Cannot Connect to LM Studio
Check that:
LM Studio is open.
Qwen3.6-27B is loaded.
The local server has been started.
Hermes uses
http://localhost:1234/v1.The model name matches LM Studio’s identifier.
Hermes Cannot Connect to Ollama
Check that:
Ollama is running.
ollama listshows the model.Hermes uses
http://localhost:11434/v1.The Hermes model name exactly matches the Ollama name.
Chat Works but Tools Fail
Possible causes include:
Incorrect chat template.
Weak tool-calling support.
Insufficient context.
An incompatible local server mode.
Excessive model compression.
A model-generation error.
Try a basic file-listing task before testing complicated workflows.
The Model Is Extremely Slow
Possible reasons include:
The model does not fit in VRAM.
Too many layers are being processed by the CPU.
The 64K context consumes too much memory.
The selected quantization is too large.
Another model instance is already loaded.
LM Studio and Ollama are both holding large models in memory.
Close unused model runners and check RAM, VRAM, context, and GPU offloading.
Hermes Forgets Something
Hermes does not automatically place every statement in internal memory.
Tell it clearly:
This is a stable fact I want you to remember across future sessions.
For detailed information, store the approved version in the appropriate Obsidian note or project context file.
Final Thoughts
The most exciting thing about Hermes Agent is not that it gives us another chat window.
It gives us a foundation for building a personal AI environment.
Qwen3.6-27B provides the local intelligence.
LM Studio or Ollama provides the engine that runs the model.
Hermes gives that intelligence access to tools, memory, files, skills, projects, and multi-step work.
Obsidian can provide a detailed, human-readable knowledge library that remains under our control.
Together, these components create something much more meaningful than a temporary chatbot.
The model can be upgraded as better local models become available. Skills can be developed around the way we work. Projects can preserve their own instructions. Important knowledge can remain in local files we can inspect, edit, and back up.
For people living with expensive, unstable, limited, or restricted internet access, this combination is particularly powerful. It offers privacy, continuity, independence, and creative freedom that cloud-only tools cannot always provide.
The goal is not to give AI unlimited control.
The goal is to gradually build a trusted creative and working partner—one that operates inside clear boundaries, remembers what genuinely matters, learns useful procedures, and keeps its knowledge in places we can understand and own.
That is why Hermes Agent is important.
And with Hermes Desktop, Qwen3.6-27B, LM Studio or Ollama, and an optional Obsidian knowledge vault, creating your own locally controlled AI assistant is becoming possible even for people who do not consider themselves technical.
No comments:
Post a Comment