When you build an AI toy, the first decision is not which model to pick but which platform to build on. Today's mainstream supply falls into three technology roadmaps: the cloud-based Tuya T5E, the on-device open-source Espressif ESP32, and the lightweight command-driven PY32. The question is not which one is better, but which one fits you. This article breaks the options down objectively, from architecture through to selection, as a reference for people working in the industry.
Roadmap 1: The cloud-based Tuya T5E
The Tuya T5E platform runs a dual stack of local SDK plus cloud API: TuyaOS runs locally and handles device logic, connectivity and wake-up, while the cloud provides AI conversation, multilingual support and global data centers. Its strengths are a mature ecosystem and out-of-the-box readiness; 60+ languages and pre-compliance for overseas markets make it the first choice for a fast move into global markets.
The trade-off is that openness is bounded by the platform: there is limited freedom for secondary development, and accounts and data live inside the Tuya ecosystem. It suits toy factories that want to get product into overseas markets as quickly as possible and do not want to build their own cloud and LLM stack.
Roadmap 2: The on-device open-source Espressif ESP32
The Espressif ESP32S3 platform is also a local SDK + cloud API dual stack: ESP-IDF runs locally and can handle on-device command parsing, camera pre-processing and customizable wake words, while the cloud side connects to a private cloud that delivers AI conversation and emotional expressions. Its strength is an extremely high degree of freedom - the app, WeChat Mini Program, admin console, account system and dedicated LLM can all be customized end to end.
The trade-off is that deep customization also means deep investment, so it suits toy factories that want to build their own brand and keep data sovereignty. Solution providers that have packaged this roadmap into a product line - the LX Series from Nablai, for example - lead with complete brand sovereignty, backed by the Nablai App and the AMS vendor management platform.
Roadmap 3: The lightweight PY32 command platform
The PY32 platform is a pure UART command module: command frames drive the expression screen, vibration and audio. There is no AI, no cloud and no SDK - it is command-driven performance rather than semantic understanding. Cost and barrier to entry are both the lowest of the three, which suits low-cost market tests or purely performative scenarios. The NT Series from Nablai falls into this category.
It does not belong in the same category as the two smart module dual stacks above, and the two should not be mixed up when you select: whether you want a companion that can talk or a prop that can perform is what determines whether PY32 is the right answer.
Dual stack vs. command: the key distinction
In this industry you often hear TY described as purely cloud and LX as purely local, the two set up as opposites. That is wrong. Both TY002 and LX002 are a local SDK + cloud API dual stack; the difference lies in ecosystem ownership and brand sovereignty, not in architecture type. NanoToy (PY32) is the true pure-command module. Get this wrong and your whole selection framework is skewed.
Selection decision matrix
Advice for decision makers
Start with the commercial goal: choose a TY-class platform to move overseas fast, an LX-class platform for brand sovereignty, an NT-class platform to validate the market. Many of the success stories follow a stepped path - PY32 to test the water, then TY to go global, then LX to build a long-term brand - rather than a single leap. Once the underlying platforms converge, what actually separates products is content operations, character customization and how data sovereignty is arranged.
FAQ
Q: Is a dual stack more expensive than a pure cloud approach? A: The local SDK in a dual stack gives you better offline experience and responsiveness; cost depends on how deep the customization goes, so it should not be judged simply as cloud vs. local.
Q: Does the Espressif route force you to build your own cloud? A: Not necessarily. You can connect to a solution provider's private cloud, or build your own. What matters is whether the accounts and the data are yours.
Q: Can PY32 be upgraded into a smart module later on? A: The architecture is different, so it is usually a change of platform rather than an upgrade. Using PY32 for early-stage validation is the more economical choice.
Pulling the discussion back to selection itself, there is one judgement that is constantly muddled: some people treat TY as purely cloud and LX as purely local. That is a false dichotomy. Both are a local SDK + cloud API dual stack; the difference is ecosystem ownership and brand sovereignty. Understand that and you will not get dragged off course in a selection meeting by the fake opposition between a cloud camp and a local camp.
For brand owners the more practical questions are: do I want my user data sitting inside somebody else's ecosystem, and who are my characters and content deeply bound to? Answer those clearly and the roadmap becomes clear on its own. Technology roadmaps also shift as the business moves through stages: many teams use TY to validate overseas markets quickly, move to LX to build their own brand once that works, and finish with NT as a low-cost entry-level product. The above is industry observation and does not constitute an endorsement of any vendor; base your specific selection on each vendor's public datasheets and measured data.
Shenzhen Nablai Intelligent Technology Co., Ltd. — AI module specialists for smart toys
📧 contact@nablai.com.cn 🌐 www.nablai.com.cn