🤖 Run AI Agents on Microcontrollers

TinyAgent is a lightweight AI agent framework designed to run on ESP32 and Arduino-compatible microcontrollers. It brings the power of LLM-driven autonomous agents to the embedded world — with minimal memory footprint and maximum flexibility.

Modern AI agents (LangChain, AutoGPT, etc.) require powerful servers. But what if your IoT device could think for itself? TinyAgent bridges the gap:

• 🧠 ReAct Loop — Observe → Think → Act cycle running on your ESP32
• 🌐 LLM Integration — Call OpenAI, Ollama, or any compatible API over WiFi
• 🔧 Tool System — GPIO, sensors, HTTP requests, MQTT — all as agent tools
• 📦 Tiny Footprint — Designed for constrained environments (~20KB RAM usage)
• ⚡ Easy Setup — Arduino IDE & PlatformIO compatible

• Multiple LLM backends: OpenAI API, Ollama, any OpenAI-compatible endpoint
• Built-in tools: GPIO read/write, analog sensors, HTTP requests, MQTT publish
• Custom tools: Register your own tools with a simple callback API
• Streaming support: Process LLM responses as they arrive
• Configurable: System prompts, max iterations, tool selection — all runtime-configurable
• Memory efficient: Reuses buffers, streams JSON parsing, minimal allocations

┌─────────────────────────────────────────────────┐
│                   TinyAgent                      │
│                                                  │
│  ┌───────────┐    ┌──────────┐    ┌───────────┐ │
│  │  Observe   │───▶│  Think   │───▶│    Act    │ │
│  │ (sensors)  │    │  (LLM)   │    │  (tools)  │ │
│  └───────────┘    └──────────┘    └───────────┘ │
│       ▲                                  │       │
│       └──────────────────────────────────┘       │
│                  ReAct Loop                      │
├─────────────────────────────────────────────────┤
│                 Tool Registry                    │
│  ┌────────┐ ┌────────┐ ┌───────┐ ┌──────────┐  │
│  │  GPIO  │ │ Sensor │ │ HTTP  │ │   MQTT   │  │
│  └────────┘ └────────┘ └───────┘ └──────────┘  │
├─────────────────────────────────────────────────┤
│               LLM Client (WiFi)                  │
│         OpenAI  │  Ollama  │  Custom             │
└─────────────────────────────────────────────────┘

PlatformIO (recommended):

; platformio.ini
[env:esp32]
platform = espressif32
board = esp32dev
framework = arduino
lib_deps =
    redbasecap-buiss/TinyAgent
    bblanchon/ArduinoJson@^7.0.0

Arduino IDE:

1. Download this repository as ZIP
2. Sketch → Include Library → Add .ZIP Library
3. Install dependency: ArduinoJson by Benoît Blanchon

#include <WiFi.h>
#include <TinyAgent.h>

TinyAgent agent;

void setup() {
    Serial.begin(115200);
    WiFi.begin("your-ssid", "your-password");
    while (WiFi.status() != WL_CONNECTED) delay(500);

    // Configure LLM
    agent.getLLMClient().configure("https://api.openai.com/v1/chat/completions",
                                   "your-api-key", "gpt-4o-mini");

    // Set agent behavior
    agent.setSystemPrompt("You are a helpful IoT assistant. Use tools to interact with hardware.");
    agent.setMaxIterations(5);

    // Register built-in tools
    agent.registerBuiltinTools();

    // Run the agent
    agent.run("What is the current temperature reading from the analog sensor on pin 34?");
}

void loop() {
    // Agent runs in setup(), or call agent.run() here for continuous operation
    delay(10000);
}

agent.getToolRegistry().registerTool(
    "read_dht",
    "Read temperature and humidity from DHT22 sensor",
    "{\"type\":\"object\",\"properties\":{}}",
    [](const String& params) -> String {
        float temp = dht.readTemperature();
        float hum = dht.readHumidity();
        return "{\"temperature\":" + String(temp) + ",\"humidity\":" + String(hum) + "}";
    }
);

• basic_agent — Minimal agent setup with GPIO tools
• sensor_monitor — Autonomous sensor monitoring and reporting
• home_automation — MQTT-based home automation agent

• Core ReAct loop
• OpenAI API integration
• Built-in GPIO and sensor tools
• MQTT tool
• Ollama native support
• Conversation memory (SPIFFS/LittleFS persistence)
• Multi-agent communication
• BLE tool
• OTA update tool
• Web dashboard for monitoring
• Function calling (structured tool use)
• ESP-IDF native support (no Arduino)

• ESP32 board (ESP32, ESP32-S2, ESP32-S3, ESP32-C3)
• Arduino framework
• ArduinoJson >= 7.0.0
• WiFi connectivity
• LLM API endpoint (OpenAI, Ollama, or compatible)

Contributions are welcome! Please read CONTRIBUTING.md for guidelines.

MIT License — see LICENSE for details.

Made with ⚡ for the embedded AI community