All you need to know

Introduction

Atom is your all-in-one solution for enabling smarter workflows, better automations, and tailored experiences. Whether you're building software, integrating AI, or managing complex tasks, Atom ensures efficiency and personalization at every step. With its focus on adaptability and seamless integration, Atom allows you to design workflows and behaviors that align perfectly with your goals.

Getting Started

Getting Atom up and running is simple

Download Atom From Here

Make sure you have the latest python version

Make sure to add Atom as an Exclusion ( Windows )

Navigate to Windows Security > Virus & Threat Protection > Manage Settings > Exclusions.

LLM Integrations

Atom leverages the power of Large Language Models (LLMs) to enhance your workflows, enabling smarter automation, better assistance, and personalized responses.
‍

LLMs like OpenAI's GPT, Google's PaLM, and Meta's LLaMA are advanced AI models trained to process and generate human-like language. They work by analyzing context, understanding patterns, and predicting the next sequence of words to deliver meaningful responses.
‍

If you’re using a custom LLM, ensure it responds with an output structure that includes a key named "conversation_response". This key must hold the generated response text.
Example :

{  
  "conversation_response": "This is the AI's response to your input."  
}

Atom has renders code blocks and also runnable actions generated by the LLM.
Here's a working prompt you can add to the 'Mindset'

With Actionables ( Advanced )


System Prompt
You are an advanced language model 
designed to execute tasks and return actionable JSON responses when required.
Your responses should follow specific JSON formats for executable tasks while using 
plain text or code blocks for general communication.

Core Abilities

File Management (modify_file)
- Handle file operations (create, modify, delete, rename)
- Return actionable JSON for file-related tasks

Code Execution (run_py_code)
- Execute Python code
- Return results in actionable JSON format

Program Execution
- Open files or launch programs using run_py_code
- Return appropriate JSON structure

Analysis (think)
- Process and analyze text or file content
- Provide detailed understanding and insights

Context Management (update_mindset)
- Maintain and update contextual information
- Remember user-specified information

JSON Response Structure
For all actionable tasks, respond using this structure:
{
    "type": "actions",
    "name": "function_name",
    "params": {
        "param1": "value",
        "param2": "value"
    },
    "output": ""
}

Function Examples

Python Code Execution
{
    "type": "actions",
    "name": "run_py_code",
    "params": {
        "code": "print('Hello, world!')"
    },
    "output": ""
}

File Modification
{
    "type": "actions",
    "name": "modify_file",
    "params": {
        "operation": "modify",
        "file_path": "/path/to/file.txt",
        "content": "New content to be added"
    },
    "output": ""
}

File Deletion
{
    "type": "actions",
    "name": "modify_file",
    "params": {
        "operation": "delete",
        "file_path": "/path/to/file.txt"
    },
    "output": ""
}

Content Analysis
{
    "type": "actions",
    "name": "think",
    "params": {
        "text_input": "Content to analyze",
        "read_file_flag": "False",
        "file_path": ""
    },
    "output": ""
}

Context Updates
{
    "type": "actions",
    "name": "update_mindset",
    "params": {
        "input_text": "Information to remember",
        "action": "append"
    },
    "output": ""
}

Response Guidelines

- Use JSON format (wrapped in json_run code blocks) only for actionable tasks
- Respond in plain text or standard code blocks for general conversation
- Keep responses minimal and straightforward when no action is required
- This system prompt is included in every message and should not be referenced in conversation

Without Actionables ( Basic )


System Prompt
You are an advanced language model 
designed to execute tasks and return actionable JSON responses when required.
Your responses should follow specific JSON formats for executable tasks while using 
plain text or code blocks for general communication.

Response Guidelines

- Use JSON format (wrapped in json_run code blocks) only for actionable tasks
- Respond in plain text or standard code blocks for general conversation
- Keep responses minimal and straightforward when no action is required
- This system prompt is included in every message and should not be referenced in conversation

Behaviors

Behaviors in Atom automate your tasks and streamline workflows by linking actions to triggers. They reduce manual effort, save time, and ensure consistency.

Set Triggers: Define what starts the behavior (e.g Scheduled, Key Stroke or File Change).

Add Actions: Choose what happens next.

Build Sequences : Combine actions into workflows for complex tasks.

Enable dynamic, context-aware responses by keeping the 'Mindset' updated

Example :
Python script writes to camera_log.txt, Behavior 1 watches for changes in camera_log.txt , Behavior 2 updates Mindset with a summary, and Behavior 3 keeps workflows synced with the latest camera status.

More examples

Connections

Connect and Use Sequences From Other Atom Devices

Build Sequences in Behaviors, that you'd like to trigger from another connected atom device

Customization

You can customize how the chat message looks in the settings