Espressif
ESP32 Edge AI Guides
520 KB SRAM and 240 MHz dual-core supports basic ML inference. TFLite Micro and Edge Impulse officially supported.
Hardware Specs
| Processor | Dual-core Xtensa LX6 @ 240 MHz |
| Cores | 2 |
| Clock | 240 MHz |
| SRAM | 520 KB |
| PSRAM | 4 MB |
| Flash | 16 MB |
| FPU | none |
| Connectivity | Wi-Fi 802.11 b/g/n, Bluetooth 4.2 BR/EDR + BLE |
| Key Features | Hardware crypto acceleration, Ultra-low-power co-processor (ULP) |
| Price | $2–$5 (chip), $5–$15 (dev board) |
Dev Boards in This Family
136 edge-AI-capable ESP32 development boards with specs and compatibility details.
M5Stack Core2
M5Stack
4416 KB RAM · 240 MHz
M5Stack FIRE
M5Stack
4416 KB RAM · 240 MHz
M5Stack Station
M5Stack
4416 KB RAM · 240 MHz
Pycom LoPy4
Pycom Ltd.
1280 KB RAM · 240 MHz
TTGO T-Beam
TTGO
1280 KB RAM · 240 MHz
TTGO T7 V1.3 Mini32
TTGO
1280 KB RAM · 240 MHz
TTGO T7 V1.4 Mini32
TTGO
1280 KB RAM · 240 MHz
Pycom WiPy3
Pycom Ltd.
1280 KB RAM · 240 MHz
AZ-Delivery ESP-32 Dev Kit C V4
AZ-Delivery
520 KB RAM · 240 MHz
FireBeetle-ESP32
DFRobot
520 KB RAM · 240 MHz
M5Stack Core ESP32 16M
M5Stack
520 KB RAM · 240 MHz
M5Stack GREY ESP32
M5Stack
520 KB RAM · 240 MHz
Hardware Guides
ESP32 Anomaly Detection with Edge Impulse
Espressif's ESP32 excels at anomaly detection via Edge Impulse. The 2-core xtensa-lx6 at 240 MHz with 520 KB SRAM handles 15 KB quantized mo…
ESP32 Anomaly Detection with TFLite Micro
The ESP32 runs autoencoder-based anomaly detection with TFLite Micro by learning normal sensor patterns and flagging deviations. Models unde…
ESP32 Fall Detection with Edge Impulse
The ESP32 is an excellent match for fall detection with Edge Impulse. 520 KB SRAM delivers 8.1x the 64 KB minimum while 240 MHz processes 20…
ESP32 Fall Detection with TFLite Micro
The ESP32 is an excellent match for fall detection with TFLite Micro. 520 KB SRAM delivers 8.1x the 64 KB minimum while 240 MHz processes 20…
ESP32 Gesture Recognition with Edge Impulse
The ESP32 is an excellent match for gesture recognition with Edge Impulse. 520 KB SRAM delivers 8.1x the 64 KB minimum while 240 MHz process…
ESP32 Gesture Recognition with TFLite Micro
For gesture recognition, the ESP32 with TFLite Micro scores Excellent. Its 520 KB internal SRAM (8.1x the required 64 KB) and 240 MHz clock …
ESP32 Image Classification with Edge Impulse
The ESP32 handles image classification effectively with Edge Impulse. 520 KB SRAM at 240 MHz provides 4.1x headroom over the 128 KB requirem…
ESP32 Image Classification with TFLite Micro
Running image classification on the ESP32 with TFLite Micro is practical. 520 KB SRAM meets the 128 KB minimum with 4.1x headroom. The 240 M…
ESP32 Object Detection with Edge Impulse
Espressif's ESP32 is a solid choice for object detection using Edge Impulse. The xtensa-lx6 core at 240 MHz with 520 KB SRAM accommodates 25…
ESP32 Object Detection with TFLite Micro
Espressif's ESP32 is a solid choice for object detection using TFLite Micro. The xtensa-lx6 core at 240 MHz with 520 KB SRAM accommodates 25…
ESP32 People Counting with Edge Impulse
Running people counting on the ESP32 with Edge Impulse is practical. 520 KB SRAM meets the 192 KB minimum with 2.7x headroom. The 240 MHz xt…
ESP32 People Counting with TFLite Micro
The ESP32 handles people counting effectively with TFLite Micro. 520 KB SRAM at 240 MHz provides 2.7x headroom over the 192 KB requirement f…
ESP32 Predictive Maintenance with Edge Impulse
The ESP32 handles vibration-based predictive maintenance with Edge Impulse by classifying accelerometer patterns into normal, warning, and f…
ESP32 Predictive Maintenance with TFLite Micro
The ESP32 is an excellent match for predictive maintenance with TFLite Micro. 520 KB SRAM delivers 8.1x the 64 KB minimum while 240 MHz proc…
ESP32 Sound Classification with Edge Impulse
The ESP32 is an excellent match for sound classification with Edge Impulse. 520 KB SRAM delivers 8.1x the 64 KB minimum while 240 MHz proces…
ESP32 Sound Classification with TFLite Micro
Espressif's ESP32 excels at sound classification via TFLite Micro. The 2-core xtensa-lx6 at 240 MHz with 520 KB SRAM handles 40 KB quantized…
ESP32 Voice Recognition with Edge Impulse
The ESP32 is an excellent match for voice recognition with Edge Impulse. 520 KB SRAM delivers 4.1x the 128 KB minimum while 240 MHz processe…
ESP32 Voice Recognition with TFLite Micro
For voice recognition, the ESP32 with TFLite Micro scores Excellent. Its 520 KB internal SRAM (4.1x the required 128 KB) and 240 MHz clock e…
ESP32 Wildlife Monitoring with Edge Impulse
The ESP32 handles wildlife monitoring effectively with Edge Impulse. 520 KB SRAM at 240 MHz provides 4.1x headroom over the 128 KB requireme…
ESP32 Wildlife Monitoring with TFLite Micro
Espressif's ESP32 is a solid choice for wildlife monitoring using TFLite Micro. The xtensa-lx6 core at 240 MHz with 520 KB SRAM accommodates…
Related Guides
AI Agents for Embedded Systems
What AI agents mean for embedded systems: the on-device sense-inference-act loop, plus graph orchestration on a Linux edge gateway with ForestHub edge-agents.
How to Deploy AI Models to Microcontrollers
Step-by-step guide to deploying machine learning models on ESP32, STM32, and Arduino MCUs using TensorFlow Lite Micro and Edge Impulse.
Edge Agent Platform vs. Point Solution
Why an edge agent platform of runtime, builder, deploy and fleet carries more use cases over time than a one-off point solution per problem.
Edge Agents vs Cloud Agents
Edge agents vs cloud agents: latency, privacy, cost, reliability, and determinism compared, and when to run the agent loop on-device vs in the cloud.
Edge AI Agents on Microcontrollers
How edge AI agents run on microcontrollers: the sense-reason-act loop on-device, architecture, memory constraints on ESP32 and STM32, and when to use them.
Edge AI for Manufacturing
How manufacturers deploy edge AI for quality control, predictive maintenance, and energy monitoring. MCU-based solutions without cloud dependency.
Privacy & GDPR by Design at the Edge
How an Edge Agent makes data sovereignty an architecture property through local processing, GDPR data minimization, and CRA readiness by design.
Edge AI vs Cloud AI: When to Use Which
Compare edge AI and cloud AI across latency, cost, privacy, and power. Learn when to process on-device and when cloud inference is the better choice.
ESP32 vs STM32 for AI Applications
Compare ESP32 and STM32 for edge AI. Architecture, ML performance, tooling, connectivity, and cost analysis for embedded ML projects.
EU AI Act & Cyber Resilience Act for Edge AI
What the EU AI Act and Cyber Resilience Act mean for industrial edge AI agents, and how an auditable, deterministic architecture supports compliance.
Fall Detection in the Smart Home with a Local Edge Agent
How an Edge Agent could detect a fall locally on the device and alert a stored contact, without any video or audio leaving the home.
Graph-first vs Loop-first for Edge AI Agents
Why deterministic, auditable AI wins on the machine. Graph-first makes the workflow the program and the AI agent just one bounded node among many.
How to Build Agentic Edge AI
A practical guide to building agentic edge AI: sensors, on-device inference, decision and state logic, actuation, and orchestration on ESP32 and STM32.
How to Run TensorFlow Lite on ESP32
Step-by-step guide to running TFLite Micro (LiteRT) on ESP32 with ESP-IDF. Tensor arena, operator registration, and C inference code included.
Industrial Edge Gateways for AI Agents
How to choose and architect industrial edge gateway hardware that runs AI agents — CPU/GPU/NPU, fieldbus I/O, DIN-rail, secure boot, and Docker on Linux.
HVAC Edge Agent, On-Device AI for Building Control
An HVAC Edge Agent fuses CO2, temperature and occupancy on-device, writing only setpoints to the controller, never in the loop. GDPR-friendly edge AI.
Modbus and OPC-UA for Edge AI Agents
How to connect AI agents to PLCs over Modbus and OPC-UA at the Linux edge gateway: registers, the OPC-UA information model, subscriptions, and write safety.
MQTT for Industrial Edge AI Agents
How MQTT pub/sub, QoS, Sparkplug B, and edge brokers feed sensor data into an AI agent graph on a Linux edge gateway. A practical guide for OEMs.
On-Device Knowledge Retention with Edge Agents
How an on-device Edge Agent captures implicit service knowledge, makes it searchable, and guards against the skills shortage, fully offline at the machine.
Predictive Maintenance Across a Distributed Fleet
Local anomaly detection plus federated learning. Edge Agents explain faults on site and the fleet learns together, without centralizing raw data.
Predictive Maintenance on a Linux Edge Gateway
How to scale predictive maintenance from a single vibration node to a plant-wide system on a Linux edge gateway over MQTT, Modbus, and OPC-UA.
Predictive Maintenance with Edge AI
Deploy edge AI for predictive maintenance. Vibration analysis, anomaly detection, and ROI calculation for MCU-based monitoring on factory equipment.
Small Language Models (SLM) on Edge Hardware
Why an on-device small language model handles most industrial tasks, and how the intelligence cascade from rules to cloud keeps data on-premises.
Smart Home Setup with a Local Edge Agent Assistant
An Edge Agent turns smart home setup into a guided local dialog. On-device AI, private, no mandatory account, with MQTT device discovery.
TinyML Getting Started Guide
Learn TinyML from scratch. Set up your first machine learning project on a microcontroller with ESP32, Arduino, or STM32 using TFLite Micro or Edge Impulse.
The Virtual Service Technician as an Edge Agent
An on-device Edge Agent acts as a virtual service technician that detects machine faults and guides operators step by step, with no ticket needed.
What Is an Edge Agent? Definition and Architecture
An Edge Agent is an AI agent that runs on the device itself. It perceives, reasons and acts locally within a bounded, deterministic graph.
What Is Edge AI Orchestration?
Edge AI orchestration runs models, sensors, and actions as a graph on a Linux edge gateway, not monolithic firmware. Built with ForestHub edge-agents.
Why AI Agents Run on the Machine, Not the Cloud
Four reasons for on-device AI over the cloud. Latency, cost, connectivity and data sovereignty. How an Edge Agent works right on the device.
Other Microcontrollers
EFR32MG
Silicon Labs
256 KB RAM · 40 MHz
ESP32-C3
Espressif
400 KB RAM · 160 MHz
ESP32-C6
Espressif
512 KB RAM · 160 MHz
ESP32-S2
Espressif
320 KB RAM · 240 MHz
ESP32-S3
Espressif
512 KB RAM · 240 MHz
GAP8
GreenWaves Technologies
512 KB RAM · 250 MHz
i.MX RT1052
NXP
512 KB RAM · 600 MHz
i.MX RT1062
NXP
1024 KB RAM · 600 MHz
i.MX RT1064
NXP
1024 KB RAM · 600 MHz
LPC55xx
NXP
320 KB RAM · 150 MHz
nRF52832
Nordic Semiconductor
64 KB RAM · 64 MHz
nRF52833
Nordic Semiconductor
128 KB RAM · 64 MHz
nRF52840
Nordic Semiconductor
256 KB RAM · 64 MHz
RA6M5
Renesas
512 KB RAM · 200 MHz
SAMD51
Microchip
256 KB RAM · 120 MHz
SAME51
Microchip
256 KB RAM · 120 MHz
STM32F3
STMicroelectronics
80 KB RAM · 72 MHz
STM32F4
STMicroelectronics
192 KB RAM · 168 MHz
STM32F7
STMicroelectronics
512 KB RAM · 216 MHz
STM32G4
STMicroelectronics
128 KB RAM · 170 MHz
STM32H5
STMicroelectronics
640 KB RAM · 250 MHz
STM32H7
STMicroelectronics
1024 KB RAM · 480 MHz
STM32L4
STMicroelectronics
128 KB RAM · 80 MHz
STM32L5
STMicroelectronics
256 KB RAM · 110 MHz
STM32U5
STMicroelectronics
786 KB RAM · 160 MHz
STM32WB
STMicroelectronics
256 KB RAM · 64 MHz
Orchestrate ESP32 Edge AI with ForestHub
The ESP32 runs inference on-device. ForestHub on your Linux edge gateway ingests its results over MQTT, orchestrates the sense-reason-act loop as a deterministic, auditable graph, and acts on the line.
Get Started Free