Course 3 - Connect Hermes to OpenCode Go
Before this course, finish Courses 1 and 2. You will need a payment method and your parent's $5 top-up.
What this course does
By the end of this course, you will have:
- An OpenCode Go account with $5 in it
- An API key pasted into Hermes
- The agent routing hard questions to a much bigger cloud model automatically
- A working hybrid setup: local for everyday, cloud for the hard stuff
Lesson 1 - Why cloud backup
Right now, every chat with Hermes goes through the local model in LM Studio. That model is fast, private, and free. But it has limits.
For example, the local model might:
- Get confused on a multi-step maths problem
- Not know current events
- Produce a short answer when you need a long one
- Struggle with abstract reasoning
A cloud model is bigger. It can handle those things. But it costs money per query and your messages leave the laptop.
The hybrid pattern: local by default, cloud when needed. The agent decides. You don't have to.
Lesson 2 - What OpenCode Go is
OpenCode Go is a paid gateway that lets Hermes talk to many big cloud models through one account.
The models you can use include:
- DeepSeek v4 Pro - very strong at reasoning, code, and following complex instructions
- Kimi k3 - good at long documents, multi-step tasks
- Qwen 3.7 Max - good at creative tasks and multilingual
- GLM 5.2 - balanced, reliable
- Grok 4.5 - good at conversational tasks
For this course, we recommend DeepSeek v4 Pro as the default cloud model. It's the smartest of the lot.
The cost is tiny. A typical question costs less than a cent. The $5 top-up from your parent covers many months of typical use.
Lesson 3 - Sign up
Steps
- Open your browser
- Go to https://opencode.ai/zen/go
- Click Sign Up
- Enter your email and create a password
- Verify your email (check inbox for the link)
- Sign in
Check
- You're signed in to the OpenCode dashboard
Lesson 4 - Add the $5
A parent needs to do this step. It requires a payment method.
Steps
- In the OpenCode dashboard, click Wallet or Billing
- Click Add Credit
- Enter $5.00 (or more, if you want a buffer)
- Enter your parent's payment method
- Confirm
What's stored
The $5 is a deposit, not a subscription. There's no monthly fee. You pay only when you actually use a cloud model. The $5 will last most families for several months.
Check
- Wallet shows $5.00 (or your deposited amount)
Lesson 5 - Get your API key
Steps
- In the OpenCode dashboard, click API Keys (or Settings)
- Click Create New Key
- Name it:
Hermes Desktop
- Copy the key to your clipboard
Where to paste it
Don't paste it into chat. Don't send it to anyone. The key is what links your OpenCode account to your usage.
Save it somewhere private. A note in your phone's password manager is fine.
Check
- You have a long alphanumeric string starting with
sk-...
Lesson 6 - Connect Hermes to OpenCode
Steps
- Open Hermes Desktop
- Click your profile icon (top-right)
- Click Settings
- Find Cloud Providers or Providers
- Click Add Provider or Configure OpenCode
- Paste the API key from Lesson 5
- Set the model dropdown to DeepSeek v4 Pro (or your preference)
- Click Save
Hermes will test the connection. If it works, you'll see a green dot or "connected" status.
Check
- Hermes shows "connected" for OpenCode
- The model dropdown lists DeepSeek v4 Pro
Lesson 7 - Set up the hybrid pattern
Now we tell Hermes when to use each model.
Steps
- In Hermes Settings, find Routing or Hybrid Mode
- Set the defaults:
- Default model: Local (LM Studio)
- Fallback for hard questions: OpenCode (DeepSeek v4 Pro)
- Save settings
What this means
Day-to-day chats use your local model. If you ask a question that the local model can't handle, the agent automatically tries the cloud model. You don't have to choose.
Check
- Ask a simple question: "What's the weather like today?" - answered locally
- Ask a hard question: "Explain quantum entanglement like I'm 14" - answered by the cloud model
To see which model answered, look at the bottom of the reply. It should say something like "via DeepSeek v4 Pro" or "via LM Studio".
Lesson 8 - Test the hard-question routing
The cloud model is best for:
- Multi-step reasoning
- Long documents
- Abstract concepts
- Current events
Try these
- "Plan a 7-day revision schedule for my maths exam, factoring in 2 hours of sport each day and a family dinner every night."
- "Explain the difference between a comet, asteroid, and meteor in a way a 12-year-old would understand."
- "Help me write a polite email to my teacher explaining I missed the assignment deadline because my laptop broke."
The local model will probably struggle with these. The cloud model will give you detailed, useful answers.
Lesson 9 - Watch the wallet
After a week of use, check how much credit you've spent.
Steps
- Go back to opencode.ai/zen/go
- Sign in
- Click Wallet
- See the current balance and recent usage
What to expect
If you used the cloud model for 5-10 questions per day, you should have spent less than $1. The $5 should last 2-3 months minimum.
If you run out
You can top up again. Same page, same button. Your parent does it.
Check
- Wallet shows a balance above $0
- You can see which questions used the cloud model and which used the local one
Lesson 10 - Adjust the routing
After using the system for a week, you may want to change which questions go to the cloud.
Make the cloud more aggressive
If you want the agent to use the cloud more often, edit the routing rules. Tell Hermes: "Use the cloud model for any question that takes more than 5 seconds to answer locally."
Make the cloud more conservative
If you want to save money, do the opposite: "Only use the cloud model if the local model says 'I don't know' or gives a reply under 3 sentences."
Default
For most teens, the default routing is fine. The agent decides based on question complexity.
Check
- You know where the routing settings are
- You can change them as needed
You're done with Course 3
When you finish this course, you have:
- An OpenCode Go account with $5 in it
- An API key wired into Hermes
- The hybrid pattern: local + cloud, automatic routing
- Access to DeepSeek v4 Pro for the hard questions
Your agent is now substantially smarter than the local-only setup. Same privacy promise for everyday chat. Better answers when you need them.
Next course: Course 4 - Personalise the agent. This is where the agent learns your teen's name, interests, and the essential they want to focus on.
Troubleshooting
OpenCode Go says the API key is invalid
- Copy the key again from the dashboard. Don't add any spaces.
Hermes fails to connect to OpenCode
- Make sure the API key is pasted correctly. Make sure the wallet has credit.
The cloud model is always slower than the local one
- Normal. Cloud models take 2-5 seconds over the network. Local models are instant. The cloud is for hard questions; the local is for fast ones.
The cost is higher than I expected
- Check Hermes 's usage history. Some questions are routed to the cloud by default. Adjust the routing rules.
The cloud model won't answer certain questions
- Some questions trigger content filters. Try rephrasing.
I want to use a different model
- In Hermes Settings, swap the model dropdown. Hermes supports 8+ cloud models through OpenCode Go.
I need to talk to a person
- Email go@getonetoeight.org or message the Afterschool Agent on Telegram.