The estimated shipping date is August 2026. We are currently in the middle of production and feature optimization. We have set our delivery timeline for August to ensure we have ample time to complete essential regulatory certifications (such as the FCC) and deliver a thoroughly tested, high-quality product to you.

We have already started processing the $100 cashback. Because our own payment system is still being set up, and Shopify's anti-fraud/AML restrictions prevent us from refunding more than the original transaction amount, we temporarily have to use bank transfering as a workaround. You can find the link to submit your payout information in the relevant Kickstarter Update.

Please note that sharing your bank information is completely optional. If you have any privacy concerns, you can wait for alternative methods (like PayPal), which we expect to be available around late August.

In short, you have two options:
1. Receive it now: Use the temporary bank transfer option.
2. Keep details private: Wait for our alternative payout methods to launch later this summer.

Since our Kickstarter campaign has officially concluded, we plan to launch Tiiny on our official website this September with an MSRP of $1,999.
In the meantime, you can head over to https://tiiny.ai/ to join our waitlist, which will lock in an exclusive launch discount for you.

Tiiny is designed specifically for running LLMs and agents, and it works alongside your computer. It comes with its own interface (TiinyOS) and SDK, so you use it through a client rather than as a completely standalone device.

Yes, it is designed to work fully offline. You can run models, chat, generate images, and process your local files without any internet connection.

Yes. Although we don't currently have a dedicated client for Linux like we do for macOS or Windows, you can still run Tiiny AI Pocket on Linux via TiinySDK.

Tiiny AI Pocket supports running open-source models up to 120B (int4) parameters.

Tiiny is well suited for MoE-based large LLMs, with inference speed depending on model size and context length.
Under typical real world deployment with int4 quantization:
• 20B models: ~25 to 40 tokens/s
• 70B models: ~14 to 23 tokens/s
• 120B models: ~10 to 20 tokens/s

Note: Lower speeds correspond to 32k long context; higher speeds reflect 4k short context.

Right now you can expect around ~64K to 128K context in practice, depending on the model and setup.

Yes. You can use our conversion tool to adapt your own models into a Tiiny-compatible format. The conversion tool will be fully open-source and available on GitHub. While our team is manually adapting existing frameworks (like Qwen, Llama, etc.) to the ONNX pipeline, the open-source release enables the community to contribute and support additional architectures independently.

Not right now — Tiiny doesn't support multi-device clustering / distributed inference yet.

Tiiny is not optimized for heavy video generation or diffusion workloads, and performance will vary depending on the model and configuration. Results may not meet expectations for such tasks. For use cases primarily focused on video generation, a dedicated GPU system would be more suitable than Tiiny.

All your private data is stored locally on your Tiiny's internal SSD — it never leaves the device unless you choose to move it. All data is also end-to-end encrypted, secured with a unique encryption key that only you possess, created during device setup.

No. Most features in TiinyOS will never send your data outside Tiiny and your computer, except for the following two optional online features:
1. Web Search: Search engines utilize your prompts for data retrieval.
2. Email Binding: We collect your device ID and email address solely to enable password resets via verification codes.

Yes. TiinyOS includes a built-in toolkit that lets you easily back up data to external drives or your PC, export it in standard formats, or permanently delete everything in a few steps.