KryonOS v3.0.0 - Local AI Is Coming to ESP32

[Haris16-code]

We're working on KryonOS v3.0.0, bringing a major new capability to the KryonOS ecosystem: native local AI inference on ESP32-S3 boards with PSRAM.

This update will introduce a new API for loading and running local AI models directly from binary (.bin) files, allowing compatible models to run entirely on the device without relying on cloud-based inference.

What's Coming in v3.0.0?

• Local AI Model Loading: A new API to load compatible AI model binaries directly on the ESP32-S3.
• Native C++ Inference: Model execution and inference processing will be handled natively at the C++ level, rather than through the JavaScript runtime.
• PSRAM-Aware Execution: Designed to take advantage of available PSRAM on ESP32-S3 variants to accommodate larger models within hardware limits.
• On-Device AI: Run supported models locally, reducing dependence on internet connectivity and remote AI APIs.
• JavaScript Integration: The goal is to let KryonOS applications interact with local AI through a high-level API while the underlying inference engine handles the computational workload in native code.

Our goal is to bring local AI capabilities to resource-constrained microcontrollers and make them accessible to developers building applications with KryonOS.

This is an upcoming feature, and actual model compatibility, memory requirements, and performance will depend on the model architecture and hardware configuration.

KryonOS is evolving beyond a graphical operating system and JavaScript application platform into an ecosystem for on-device AI.

Follow KryonOS development news and announcements: https://kryonos.harislab.tech/subscribe

[Haris16-code]

@korey-barrett , what do you think about the upcoming update?

[korey-barrett]

WOW now that's really exciting and i congratulate you on taking your project to the next level and i'm looking forward to the new version coming out! Is there you could have the display(touch) height/width scaleable or changeable to a different size as sticking the code at 320x240 means it's very difficult to be able to use a display other than that. i mention this as you could open up your project to the majority of esp32s3 (i understand this may not be possible with the esp32 models as they don't normally have PSRAM) users as it would mean they would't be locked into only one dimension size without re-writing the whole app. i hit this problem when i tried the use version 2 on a esp32s3 board that has a 480x480 display(touich) size and i was able to do it but it took a lot of work and i'm not happy with the result. i very much understand that this project has "experimental' for anything other than the what you have tested and know to work but i really would love to be able to experiment without having to use 240x320 (for example i have bought one of the new esp32s31 korvo 1 boards that uses a 800x480 touch screen) I can use my equipment and compute time to experiment and can advise you of any questions or results you want to have (and won't put any agent as a contributor or anywhere else) which you can then use to assist you with the main build or ignore if not required. let me know your thoughts. korey

…

On Fri, 2 Oct 2026 at 21:20, Haris ***@***.***> wrote: We're working on *KryonOS v3.0.0*, bringing a major new capability to the KryonOS ecosystem: *native local AI inference on ESP32-S3 boards with PSRAM.* This update will introduce a new API for loading and running local AI models directly from binary (.bin) files, allowing compatible models to run entirely on the device without relying on cloud-based inference. What's Coming in v3.0.0? - *Local AI Model Loading:* A new API to load compatible AI model binaries directly on the ESP32-S3. - *Native C++ Inference:* Model execution and inference processing will be handled natively at the C++ level, rather than through the JavaScript runtime. - *PSRAM-Aware Execution:* Designed to take advantage of available PSRAM on ESP32-S3 variants to accommodate larger models within hardware limits. - *On-Device AI:* Run supported models locally, reducing dependence on internet connectivity and remote AI APIs. - *JavaScript Integration:* The goal is to let KryonOS applications interact with local AI through a high-level API while the underlying inference engine handles the computational workload in native code. Our goal is to bring local AI capabilities to resource-constrained microcontrollers and make them accessible to developers building applications with KryonOS. This is an upcoming feature, and actual model compatibility, memory requirements, and performance will depend on the model architecture and hardware configuration. *KryonOS is evolving beyond a graphical operating system and JavaScript application platform into an ecosystem for on-device AI.* Follow KryonOS development news and announcements: https://kryonos.harislab.tech/subscribe — Reply to this email directly, view it on GitHub <#18?email_source=notifications&email_token=BL43U4CZXI2KNV7YCDWZ3DT5R6FQBA5CNFSNUABBM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63RPGEYDSMZVGY2DFJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BL43U4ARNIZURXCXVEWP7Z35R6FQBAVCNFSNUABJKJSXA33TNF2G64TZHMYTENZRGU3DOOBZGY5UI2LTMN2XG43JN5XDWMJQHEZTKNRUGKQXMAQ> . Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/BL43U4CGCRVEI6EXD4QQI5T5R6FQBA5CNFSNUABBM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63RPGEYDSMZVGY2DFJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KUZTPN52GK4S7NFXXG> and Android <https://github.com/notifications/mobile/android/BL43U4CY7AWKSBQS4235POT5R6FQBA5CNFSNUABBM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63RPGEYDSMZVGY2DFJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. Download it today! You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Haris16-code]

@korey-barrett Thank you very much for the detailed feedback and for offering to help test KryonOS on your hardware. I really appreciate it.

For KryonOS v3.0.0, we will be making the existing display properties more flexible so applications can work with different display resolutions.

KryonOS already provides:

System.screenWidth()
System.screenHeight()

Currently, these return the fixed dimensions of the default hardware (240x320). In v3.0.0, these APIs will instead return the actual width and height of the display being used.

For example:

System.screenWidth()   // 480
System.screenHeight()  // 480

or:

System.screenWidth()   // 800
System.screenHeight()  // 480

This means applications will be able to detect the actual display dimensions instead of assuming a fixed 240x320 screen, making it much easier to build interfaces that adapt to different display sizes.

Since KryonOS currently has native support for the ILI9341, ILI9341-based displays will continue to work. For other display controllers and hardware configurations, your testing would be extremely valuable.

Since KryonOS is open source, I highly appreciate you experimenting with the hardware you already have. Please feel free to fork the KryonOS repository and make any changes or experiments you want on your fork.

Once you have tested everything properly, you can open a Pull Request with your changes. We will review the implementation and, if everything looks good, merge it into KryonOS.

Please also keep us updated about your testing, results, issues, and any questions you have. You can contact us directly at:

kryonos@harislab.tech

If you would also like to support the development of KryonOS, your support helps us continue running this open-source project and purchase new hardware for testing. We plan to use supported contributions to acquire boards and displays so we can add officially tested and verified hardware support to KryonOS.

You can support the project here:

Support KryonOS

Thanks again for offering to help expand KryonOS hardware support. Your testing could be very useful for making KryonOS work across a much wider range of ESP32-S3 boards and displays.