boilerplate-gemini-golang

This is a boilerplate for starting to create applications with the Gemini LLM (by Google).

How to Use

First:

1. Rename the .env.example file to .env.
2. Open the newly renamed .env file and add your GEMINI_API_KEY.
3. You can obtain your API key here.

With this done, we can continue 😎

You have to add your own prompt in the prompts folder. Let's add "my_custom_prompt.go".

prompts
├── my_custom_prompt.go
├── personal_assistant.go
├── python_interpreter.go
└── talk_file.go

Now our prompts folder contains these files.

Important: prompts/my_custom_prompt.go must contain a prompt template and a function that receives the prompt written by the user and converts it to useful instructions for the LLM.

This is a basic example, so we will receive the answer just in Markdown.

prompts/my_custom_prompt.go

// prompts/my_custom_prompt.go
package prompts

import (
	"app/gemini"
	"fmt"
)

var YOUR_OWN_PROMPT_TEMPLATE = `Context: %s
User Prompt: %s

[YOUR_RESPONSE_HERE]`

// var YOUR_OWN_PROMPT = `HERE YOUR OWN PROMPT (JSON, MARKDOWN, YAML, ...)`

func YourOwnHandler(userPrompt string) string {
	return fmt.Sprintf(YOUR_OWN_PROMPT_TEMPLATE, gemini.BuildHistory(), userPrompt)
}

Adapt main.go

Now that we have "YourOwnHandler", we must use it in our main.go, line 16.

// main.go
package main

import (
	"app/gemini"
	"app/prompts"
	"fmt"
	"log"
)

func main() {
	if err := loadEnv("./.env"); err != nil {
		log.Fatal(err)
	}
	for {
		userPrompt := readPrompt("Prompt: ")
		prompt := prompts.YourOwnHandler(userPrompt)
		jsonMode := true
		response, err := gemini.Request(prompt)
		if err != nil {
			fmt.Println("Error gemini request:", err)
			continue
		}
		gemini.AddMessage(prompt)
		processResponse(response, jsonMode)
	}
}

Explaining utilities

The Gemini package includes 2 functions for managing the history (memory):

• AddMessage: If you want the LLM to remember something that you wrote, you must use this function after the request was validated:

gemini.AddMessage(message)

• AddResponse: If you want the LLM to remember something that it answered in the past, you must use:

gemini.AddResponse(response)

• processResponse: When the LLM responds with a JSON object, set the second parameter to true to enable jsonMode. This ensures that the function processes the response as JSON. For example:

processResponse(response, true)

The function will then look for the "result" key in the JSON returned by the LLM.

If the LLM's response is in Markdown format, set the second parameter to false.

If you're using a custom JSON structure, make sure to update the parseResponse function and adjust the struct accordingly to correctly handle the response.

Note: AddMessage and AddResponse are used by default, so the LLM has "memory". The memory is limited to 100 messages, but you can modify this limit in history.go (MAX_MEMORY).

It's your turn!

Start building an app with LLM now ❤️