Model or dataset
tuya/TuyaOpen avatar
tuya/TuyaOpen

TuyaOpen: seven supported platforms, a 32-bit toolchain, and a Python 3.12 lock

Next-gen AI+IoT framework for T2/T3/T5AI/ESP32/and more – Fast IoT and AI Agent hardware integration

1,948 stars321 forksCNOASSERTION

At a glance

What is it?
This is Tuya's cross-platform C and C++ SDK for its Wi-Fi and Bluetooth microcontrollers, plus a Python layer of build tooling. What the documentation gives you is precise about hardware and vague about status: every target is marked Supported, the overview promises a Raspberry Pi the table never lists, and the build image installs 32-bit libraries so one vendor executable can run.
Who is it for?
Adopt TuyaOpen when your hardware is on that list, because the per-target serial port and module names in the support table are the kind of detail you would otherwise discover the hard way, and the local model support in the examples is the fastest way to see what the device runtime can do.
Can I use it commercially?
Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
Is it still maintained?
Yes. The repository last received commits 2 days ago.
What is it written in?
Mainly C, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • tuya/TuyaOpen1,948 ★ · +123
Star history
Repository2026-09-102026-11-15
tuya/TuyaOpen1.8k1.9k

123 Stars in the last 66 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

Seven rows, every one of them marked Supported

The target platform table has four columns: name, support status, introduction, and debug log serial port. The status column is the interesting one, because it carries the same word in all seven rows.

code
| Ubuntu                | Supported      | Can be run directly on Linux hosts such as ubuntu.           |                       |
| Tuya T2                    | Supported      | Supported Module List: [T2-U](https://developer.tuya.com/en/docs/iot/T2-U-module-datasheet?id=Kce1tncb80ldq) | Uart2/115200          |
| Tuya T5                  | Supported      | Supported Module List: [T5-E1](https://developer.tuya.com/en/docs/iot/T5-E1-Module-Datasheet?id=Kdar6hf0kzmfi) [T5-E1-IPEX](https://developer.tuya.com/en/docs/iot/T5-E1-IPEX-Module-Datasheet?id=Kdskxvxe835tq) etc. | Uart1/460800          |
| ESP32/ESP32C3/ESP32S3 | Supported      |                                                              | Uart0/115200          |
| LN882H                | Supported      |                                                              | Uart1/921600          |

There is no partial, no experimental and no planned. A reader looking for a signal about how finished a port is will not find one here, which matters most for the two rows that carry no introduction text at all: the ESP32 family and LN882H.

The last column is where the real content sits. The debug log port is a different UART and a different baud rate per target: Uart2 at 115200 on T2, Uart1 at 460800 on T3 and T5, Uart0 at 115200 on the ESP32 family, Uart1 at 921600 on LN882H, Uart2 at 115200 on BK7231N. That is one line of documentation that will save you a serial console session.

The overview promises a Pi that the support table does not list

The overview describes the hardware it supports as the Tuya T-series Wi-Fi and Bluetooth MCUs, Pi, and the ESP32 family, reached through a cross-platform C and C++ SDK. The table above lists Ubuntu, T2, T3, T5, ESP32, ESP32C3, ESP32S3, LN882H and BK7231N. Pi is not among them.

The same gap appears on the other side. The repository description names T2, T3, T5AI and ESP32, and the link bar at the top of the README points at a hardware resource page for a T5-AI board, while the table's T5 row lists T5-E1 and T5-E1-IPEX modules. So the AI-branded board is promoted in the description and in the navigation, and the matrix it would belong in says T5.

For someone choosing a board this is the section to read twice, because the table is authoritative about what the build system has been run against, and the prose is written for a marketing audience. Ubuntu is the one row that is not hardware at all, and it is there so the SDK can be built and run on a Linux host, which is also the row with no serial port value.

The build image installs 32-bit libraries for one vendor executable

The Dockerfile is short and one comment explains most of it:

dockerfile
FROM ubuntu:latest

# L511G fcelf is a 32-bit dynamically linked C++ executable.
RUN dpkg --add-architecture i386 && \

The image adds the i386 architecture and installs libc6-i386 and libstdc++6:i386 alongside the normal toolchain, because a binary called fcelf from a vendor referred to as L511G is a 32-bit dynamically linked C++ program. Everything else in that layer is ordinary: wget, git, lcov, cmake-curses-gui, build-essential, ninja-build, Python 3 with pip and venv, libsystemd-dev, libusb, and a generated en_US.UTF-8 locale with LANG set to match.

Two details in the rest of the file are worth naming. The base image is ubuntu:latest, which is a rolling tag, so a build of the same commit is not guaranteed to produce the same toolchain. And the image moves Debian's EXTERNALLY-MANAGED marker aside before creating a builder user, which is the step that lets pip install into the system interpreter rather than refusing to.

So the container exists partly to host a third-party 32-bit tool. That is the kind of dependency a project usually mentions in a support matrix and this one mentions in a Dockerfile comment.

The Python layer accepts one interpreter version and pins the build tools

The Python project is tooling, not the SDK itself. The manifest names it tuyaopen, describes it as TuyaOpen SDK Python tooling dependencies, and declares:

toml
requires-python = ">=3.12,<3.13"

An upper bound on the interpreter is unusual in project metadata, and this one excludes 3.11 and 3.13 as firmly as it requires 3.12.

The dependency list is a mix of exact pins and floors, and the exact pins are on the tooling rather than on the libraries a developer would reason about. cmake==4.0.2, ninja==1.11.1.4, click==8.1.8, requests==2.33.0, urllib3==2.7.0, certifi==2025.4.26, charset-normalizer==3.4.2, colorama==0.4.6, GitPython==3.1.50, gitdb==4.0.12, smmap==5.0.2, idna==3.15, kconfiglib==14.1.0 and PyYAML==6.0.2 are all fixed. The floors cover pyserial, cffi, cryptography, pycryptodome, pycparser, setuptools, six, future and click-option-group.

Two entries are Windows-only, marked for sys_platform equal to win32: click-pwsh, for PowerShell completion, and windows-curses. Combined with the three export scripts and the bat file, the Windows path is a first-class one here, and it is also the path with the extra native dependencies.

There is a uv.lock and a uv-manifest.env at the root, so the resolution is meant to be reproducible even though the manifest is not.

Submodules are disclaimed, and one badge points at a branch that is not the default

The README carries a disclaimer that is unusual in how much work it does. It says the project may contain submodules developed by third parties, that these submodules may be updated independently, and that because the update frequency is uncontrollable the project cannot guarantee they are always the latest version. Users hitting submodule problems are advised to update them as needed or file an issue.

There is a .gitmodules file at the root, so the statement describes a real mechanism rather than a hypothetical. The same paragraph then extends to commercial use: users should recognise the functional and security risks, bear responsibility for any issues, and run their own functional and safety tests. The company, it says, accepts no liability for direct, indirect, special, incidental or punitive damages.

The second loose end is small. The commit-activity badge in the link bar points at the graphs endpoint with a query parameter naming the branch dev, while the repository's default branch is master. Any contributor who clicks it is looking at a branch that is not what a clone gives them.

The device runtime is here, the agent workflows are a cloud service

The framework's name covers two halves and only one of them lives in this repository. The overview says TuyaOpen pairs with Tuya Cloud's low-latency multimodal AI, described as drag-and-drop workflows, and that it integrates top models, naming ChatGPT, Gemini, Qwen and Doubao in one sentence and Deepseek, ChatGPT, Claude and Gemini in the feature list. The device side is what you can read here: a C and C++ SDK, boards and platform directories, an apps directory, a tests directory, and eleven example folders covering Bluetooth, e-Paper, a getting-started walkthrough, graphics, low power, multimedia, peripherals, protocols, system work, tflite and Wi-Fi.

The tflite folder is the most useful signal about where the boundary is. On-device inference is in scope; the agent that decides what to infer is not.

Other named integrations point outward as well. The feature list covers voice features on the device, automatic speech recognition, keyword spotting, text to speech and speech to text, remote control, monitoring and over-the-air updates through Tuya Cloud, compatibility with Google Home and Amazon Alexa, and custom hardware under a Powered by Tuya programme. The link bar also carries a pricing page and a link to the Dify project, which is a workflow platform rather than part of this SDK.

The README names Apache 2.0 and the metadata names nothing

The licence is described in one line: distributed under the Apache License Version 2.0, with a pointer to the LICENSE file. That file is in the tree.

The repository's licence metadata, however, reads NOASSERTION rather than Apache-2.0. A tool that reads the API rather than the README therefore learns that this project's licence could not be determined, and that is a fact you may have to explain to a scanner.

This review does not resolve the discrepancy on your behalf. What can be said is that the README and the LICENSE file agree with each other and the metadata field does not, and that anyone using this SDK inside a product should read the LICENSE file rather than trusting either field.

The releases tell the rest of the story about the project's shape. v1.7.0 on 2026-05-28, v1.8.0 on 2026-06-11 and v1.9.0 on 2026-07-22 are three version bumps in under two months, and the last push to master was 2026-09-30.

Editorial conclusion

Adopt TuyaOpen when your hardware is on that list, because the per-target serial port and module names in the support table are the kind of detail you would otherwise discover the hard way, and the local model support in the examples is the fastest way to see what the device runtime can do. Do not adopt it expecting a managed platform: the agent workflows, the model integrations and the OTA side are described as Tuya Cloud services, and the README's own disclaimer pushes functional and security risk onto you for commercial use. Before you commit, check the licence yourself, because the README says Apache 2.0 while the repository metadata says nothing at all, and note that the Python tooling layer accepts only 3.12.

Frequently asked questions

Which hardware does TuyaOpen support?

The support table lists Ubuntu, Tuya T2, T3 and T5, the ESP32 family including ESP32C3 and ESP32S3, LN882H and BK7231N, and marks every row as Supported. It also gives the debug log serial port per target, from Uart0/115200 on ESP32 to Uart1/921600 on LN882H.

Does TuyaOpen build on Linux?

Yes. Ubuntu is a row in the supported target table, described as something that can be run directly on Linux hosts. The Dockerfile in the repository also uses ubuntu:latest and adds i386 libraries plus libstdc++6:i386 for a 32-bit vendor executable named fcelf.

Which Python version does the TuyaOpen tooling support?

Only 3.12. The Python project declares requires-python as >=3.12,<3.13, and it pins its build tooling exactly, including cmake==4.0.2, ninja==1.11.1.4, kconfiglib==14.1.0 and requests==2.33.0.

Is TuyaOpen's agent workflow part of this repository?

The overview describes pairing with Tuya Cloud's multimodal AI and drag-and-drop workflows, and names model providers such as ChatGPT, Gemini, Qwen, Doubao, Deepseek and Claude. What this repository provides is the device-side C and C++ SDK, boards, platform ports and examples including a tflite folder.

What licence is TuyaOpen under?

The README says it is distributed under the Apache License Version 2.0 and points at a LICENSE file in the root. The repository's licence metadata field reads NOASSERTION, so automated tooling reading the API does not get an SPDX identifier.

Official sources

  1. Issues
  2. Project website
  3. README
  4. Releases
  5. tuya/TuyaOpen on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/tuya-tuyaopen.svg)](https://hysenlabs.com/projects/tuya-tuyaopen)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/tuya-tuyaopen.svg)](https://hysenlabs.com/projects/tuya-tuyaopen)

Was this useful?