We built Hypermode on top of Modus, an open source, serverless framework for crafting intelligent functions and APIs, powered by WebAssembly. With Hypermode, you can deploy, secure, and observe your Modus apps.

To get started, create your first Modus app. You can import this app into Hypermode in the next step.

You can import your Modus app via the Hypermode Console or through the terminal with Hyp CLI.

npm install -g @hypermode/hyp-cli

hyp link

When Hypermode creates your project, a runtime is initiated for your app as well as connections to any Hypermode-hosted models.

After deploying your app, Hypermode lands you in your project home. You can see the status of your project and the API endpoint generated for your app.

From the Query page, you can run a sample query to verify it’s working as expected. In the following query, we’re going to use the generateText function to generate text from the shared Meta Llama 3.1 model based on the prompt “How are black holes created?”

query myPrompt {
  generateText(text:"How are black holes created?")
}

Let’s dig deeper into the behavior of our AI service when we ran the query by looking at the Inferences page. You can see the step-by-step inference process and the inputs and outputs of the model at each step of your function. After invoking the Meta Llama model from the function code, we can see the function execution. In this case, it took Llama 4.4 seconds to reply to the prompt. We can also see the parameters on both the inputs and outputs.

{
  "model": "meta-llama/Llama-3.2-3B-Instruct",
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant. Limit your answers to 150 words."
    },
    {
      "role": "user",
      "content": "How are black holes created?"
    }
  ],
  "max_tokens": 200,
  "temperature": 0.7
}

Hypermode makes it simple to iterate quickly. Let’s make a few changes to your app to explore how easy customizing your API is.

Our API is responding using language that is more formal than we want. Let’s update our generateText function to respond using exclusively surfing analogies.

func GenerateText(text string) (string, error) {
  model, err := models.GetModel[openai.ChatModel]("text-generator")
  if err != nil {
      return "", err
  }

  input, err := model.CreateInput(
      openai.NewSystemMessage("You are a helpful assistant. Only respond using surfing analogies and metaphors."),
      openai.NewUserMessage(text),
  )
  if err != nil {
      return "", err
  }

  output, err := model.Invoke(input)
  if err != nil {
      return "", err
  }

  return strings.TrimSpace(output.Choices[0].Message.Content), nil
}

func GenerateText(text string) (string, error) {
  model, err := models.GetModel[openai.ChatModel]("text-generator")
  if err != nil {
      return "", err
  }

  input, err := model.CreateInput(
      openai.NewSystemMessage("You are a helpful assistant. Only respond using surfing analogies and metaphors."),
      openai.NewUserMessage(text),
  )
  if err != nil {
      return "", err
  }

  output, err := model.Invoke(input)
  if err != nil {
      return "", err
  }

  return strings.TrimSpace(output.Choices[0].Message.Content), nil
}

export function generateText(text: string): string {
  const model = models.getModel<OpenAIChatModel>("text-generator")

  const input = model.createInput([
    new SystemMessage(
      "You are a helpful assistant. Only respond using surfing analogies and metaphors.",
    ),
    new UserMessage(text),
  ])

  const output = model.invoke(input)

  return output.choices[0].message.content.trim()
}

Save the file and push an update to your Git repo. Hypermode automatically redeploys whenever you push an update to the target branch in your Git repo. Go back to the Hypermode Console and run the same query as before. You should see the response now uses surfing analogies!