Course 2 — Install LM Studio and run a local model
Before this course, finish Course 1 (Hermes installed) and Course 0 (concepts). You will need about 3 GB of free disk space and 30 minutes.
What this course does
By the end of this course, you will have:
- LM Studio installed on your laptop
- A local AI model (LFM2-700M) downloaded and running
- Hermes Desktop talking to the local model
- A 1-on-1 chat where the responses come from your laptop, not the cloud
After this course, every chat in Hermes Desktop runs locally. No internet needed.
Lesson 1 — Why local?
Before installing anything, two minutes on why this matters.
When you finish this course, every chat you have with Hermes happens on your laptop. The model file sits on your hard drive. Your prompt never leaves the machine. The reply is generated on the machine. Nothing travels.
This is the same as writing in a notebook. The words stay where you wrote them.
Cloud models (the default in Hermes when you finish Course 1) can do the same thing your local model does, but they send your messages to a server somewhere. Most cloud providers log your conversation. Some train future versions on what you typed.
Local-first isn't paranoia — it's hygiene.
Lesson 2 — Download LM Studio
What you need
- A laptop with at least 8 GB RAM
- About 1.5 GB of free disk space for the app + model
- Internet connection (only for the download)
Steps
- Open your browser
- Go to https://lmstudio.ai
- Click Download for [Mac / Windows]
- The download is about 350 MB
- Save the installer to your Downloads folder
Check
- You have
LM-Studio-Installer.dmg (Mac) or LM-Studio-Setup.exe (Windows) in Downloads
Lesson 3 — Install on Mac
- Open your Downloads folder
- Double-click
LM-Studio-Installer.dmg
- Drag the LM Studio icon onto the Applications shortcut
- Wait for the copy (about 30 seconds)
- Open LM Studio from your Applications folder
- If macOS asks for permission, click Open
Check
- LM Studio opens with a search panel
Lesson 3 — Install on Windows
- Open your Downloads folder
- Double-click
LM-Studio-Setup.exe
- Click Install when prompted
- Wait for the install (about 1 minute)
- Click Finish
- LM Studio opens automatically
Check
- LM Studio opens with a search panel
Lesson 4 — Find the model
The app has a catalogue of free, open-source models. For this course, we're using LFM2-700M from Liquid AI.
Why this model
- Small enough to run on any laptop with 8 GB RAM
- Fast enough that responses feel instant
- Smart enough for the eight essentials, daily rituals, and casual conversation
- Free and open-source
- Validated for the Afterschool Agent curriculum
Steps
- In LM Studio, look at the left sidebar. Click the magnifying glass icon (Search)
- In the search bar, type:
LiquidAI/LFM2-700M
- Look for the model card with the Liquid AI logo
- You'll see several versions. Pick the one labelled GGUF with quantisation Q4_K_M
- Click Download
- The download is about 1.4 GB. Wait for it to finish (5–15 minutes depending on your internet)
Check
- The model appears in the My Models section with a green checkmark
Lesson 5 — Test the model
Before connecting Hermes, confirm the model works on its own.
- Click the chat bubble icon (left sidebar)
- At the top, select LFM2-700M from the model dropdown
- In the message box, type: "Tell me a joke about teenagers"
- Press Enter
- Wait for the reply (3–8 seconds)
- Try: "What's one thing I should do today to be healthier?"
What you're seeing
The model file on your hard drive is reading your prompt, generating tokens (one at a time), and printing them as a reply. Every word is being generated on your laptop. To confirm, you can turn off your Wi-Fi right now and the chat will still work.
Check
- You got a coherent response to both prompts
- The model finished in under 10 seconds
Lesson 6 — Start the local server
LM Studio can also run as a small HTTP server. That's how Hermes talks to it.
- Click the developer icon in the left sidebar (looks like
- Click Start Server
- You'll see a URL appear. It will be something like
http://localhost:1234/v1
- Leave this window open. The server stays running in the background.
Check
- The server is running and the URL is displayed
- The status indicator shows a green dot or "running"
Don't close LM Studio
The server only runs while LM Studio is open. If you close LM Studio, the local server stops and Hermes will fall back to a cloud model. Keep LM Studio running in the background while you use Hermes.
Lesson 7 — Connect Hermes to the local model
Steps
- Open the Hermes Desktop app (if not already open)
- In the top-right corner, click your profile icon (or the menu button)
- Click Settings
- Find the section Local Model or Provider
- Choose LM Studio as the local provider
- Paste the URL:
http://localhost:1234/v1
- Set the model name:
lfm2-700m
- Click Save
- Back in the main chat, type: "Hello. Are you running locally?"
- The agent should reply. Look at the bottom of the reply — it should say something like "via LM Studio" or "local model"
Check
- The agent replies
- The reply metadata shows the local model is in use, not the cloud
Lesson 8 — Test it offline
The real test of local-first is that it works without internet.
- Turn off your Wi-Fi (or unplug your ethernet)
- Wait 5 seconds
- Send a message to Hermes: "Are you still there?"
- The agent should reply normally
- When done, turn Wi-Fi back on
What just happened
The agent ran on your laptop. No internet. The model, the tools, the memory — all local. The only thing that requires internet is the cloud fallback, which you can also disable if you want everything 100% local.
Check
- The agent replied while Wi-Fi was off
Lesson 9 — Adjust the model (optional)
If the model is too slow or too dumb, you can swap it. Try a bigger one.
Speed up
If responses feel slow (10+ seconds), your laptop is struggling. Try a smaller model or close other apps.
Make it smarter
If responses feel too short or basic, try a bigger model:
- Qwen 1.5B (2.5 GB, 4 GB RAM needed) — noticeably smarter
- Llama 3.1 8B (4.7 GB, 8 GB RAM needed) — much smarter, slower
In LM Studio, search for the new model, download it, and select it in the chat dropdown. Then in Hermes, update the model name in settings.
When you outgrow the local model
That's what Course 3 is for — it adds a cloud backup that kicks in only for hard questions.
Check
- You know where to find a list of free models in LM Studio
- You know how to swap models if you want to
Lesson 10 — Make it a habit
Every time you start your laptop for the next week:
- Open LM Studio (it stays in the menu bar / tray)
- Confirm the local server is running (Developer tab)
- Open Hermes
- Chat normally — the agent is using your local model
If you forget to start LM Studio, Hermes will use the cloud model instead. That's fine for now — the goal is to use the local model most of the time, not all of the time.
You're done with Course 2
When you finish this course, you have:
- LM Studio installed
- LFM2-700M running locally
- The LM Studio server running on
http://localhost:1234/v1
- Hermes Desktop connecting to that local server
- A agent that works offline
Next course: Course 3 — Connect Hermes to OpenCode Go. That adds a cloud backup for the hard questions — fast, smart, but only when you need it.
Troubleshooting
LM Studio installer is blocked by Windows SmartScreen
- Click "More info" → "Run anyway". New apps receive this warning.
The model download is failing
- Check your internet connection. The download is 1.4 GB. If it keeps failing, try a different network.
The agent is slow after connecting to LM Studio
- Your laptop may be running other things. Close browsers, video calls, and other heavy apps. Try a smaller model.
The agent says "I can't connect to the local server"
- Make sure LM Studio is still open. The server stops when you close LM Studio.
- Check the URL in Hermes settings matches the one LM Studio shows.
The model replies don't make sense
- The model is small. Some replies will be basic. That's fine for everyday chat, planning, and the rituals. Save the deep questions for Course 3 (cloud backup).
I want to use Ollama instead of LM Studio
- Ollama is a command-line tool. Hermes Desktop supports it the same way. Documentation at ollama.com. Skip this for now — LM Studio is friendlier.
Hermes is using the cloud model even though LM Studio is running
- Restart Hermes after starting LM Studio. The agent checks the local server on startup. If the server isn't running then, it falls back to cloud.
I need to talk to a person
- Email go@getonetoeight.org or message the Afterschool Agent on Telegram.