Skip to main content

Introduction

What are LLM Peripherals?​

LLM peripherals are actively controllable, physically manipulable hardware devices.

People use peripherals like keyboards to control LLMs. Similarly, LLMs use their own peripherals to interact with the physical world.

The result is not a physical agent as a workflow, but a system where the LLM can decide when and what physical actions to take within a single reasoning process. Context prompting is fundamentally key.

What we're building is not a Swiss Army knife, but situation-specific tools that perform very particular functions.

The Problem to be resolved first​

There are situations that require intelligence for judgment, but the resulting physical actions are simple.

For example:

  • Looking at a plant and deciding whether to water it
  • Considering weather and time to decide whether to open curtains
  • Responding to strangers when no one is home

These situations require complex judgment, but the actual actions are simple operations like "water the plant", "open curtains", or "send notification".

Project SABA's Vision​

Our goal is the gradual intellectualization of life and space.

When there's something inconvenient or that you want to make smarter, anyone should be able to easily intellectualize it.

To achieve this, the solution must be:

  1. Affordable
  2. Easy to install
  3. Easy to use

Project SABA is a project to achieve these goals.

Technical Features​

Plug & Play

Configure Wi-Fi once, and your device is ready.

No Schemas Required

No complex configuration files or API schemas needed. Just describe what your device does in natural language.

Intent-Based Design

Instead of "rotate motor 50°," you define "open_living_room_curtain."

LLM-Native

Works seamlessly with Claude, GPT, and other LLMs via the Model Context Protocol (MCP).

How It Works​

1. Hardware as Tools​

Your devices become tools that LLMs can autonomously use during conversations—just like they use code interpreters or search engines.

2. Simple Setup​

Connect your device, configure Wi-Fi/MQTT once, and it's ready. The LLM can immediately understand and control it through SABA's core server.

3. Semantic Function Design​

A motor can do thousands of things. What matters is the intent:

  • water_plant (not "activate pump for 3 seconds")
  • open_curtain (not "rotate motor 90 degrees")
  • press_coffee_button (not "extend actuator 2cm")

LLMs understand the purpose of a tool from its name. The name is more important than anything else.

4. Works with Any Hardware​

From simple motors and sensors to cameras and complex actuators. ESP32-based for affordability, with plans to support more platforms.