Pocket-size device supports up to 120B models locally with 0 token fees, without requiring cloud or high-end GPU
DOVER, Del., March 12, 2026 /PRNewswire/ -- Tiiny AI Inc. (www.tiiny.ai), a U.S. AI Infra startup, today announced that its Tiiny AI Pocket Lab surpassed US$1 million in Kickstarter pledges within five hours of its March 11 launch.
Tiiny AI Pocket Lab is a pocket-size personal AI supercomputer that plugs into a laptop or PC and turns it into a local AI terminal. Built as a dedicated backend for always-on AI, it is designed to support up to 120B-parameter models locally without requiring internet, cloud services, separate servers, or a high-end GPU. It also offers one-click deployment, 0 token fees, and an always-on operating profile within a 65W power envelope.
The milestone signals early demand for what Tiiny AI calls "AgentBox" — a new class of dedicated local devices designed to run always-on AI models and agent workflows privately and persistently, including OpenClaw-based agents. It also reflects a broader shift from cloud-based AI toward edge-cloud synergy, where routine agent tasks run locally for privacy, reliability, and greater cost control, while the cloud is used selectively when needed.
In the first five hours, the campaign reached US$1,009,664 in pledges from 728 backers. The backer base was led by the United States (about 70% of backers, by backer count), followed by Germany and Canada, with additional backers from the United Kingdom and Spain, as of 3:00 pm EDT on March 11.
"We are in a broader shift from cloud-based AI to edge-cloud synergy, with a new hardware layer emerging: agent-native devices built to run always-on workflows locally," said Samar Bhoj, GTM Director at Tiiny AI. "Our focus is to make that experience accessible with one-click deployment, local-first operation, 0 token fees, stronger privacy controls, and a practical edge-cloud synergy model that uses the cloud only when needed."
Pricing and Availability
- Kickstarter campaign page: https://www.kickstarter.com/projects/tiinyai/tiiny-ai-pocket-lab
- Kickstarter super early-bird price: US$1,399
- Website-only deposit program: Tiiny AI offers a US$9.90 deposit on its website to lock in a US$1,299 price (US$100 lower than the super early-bird price).
- Estimated delivery: August 2026 (final schedule and regional availability detailed on Kickstarter)
AgentBox: A New Category Built for Always-On AI Workflows
As AI adoption accelerates, cloud-based AI is encountering friction for many everyday workflows: recurring usage costs, privacy and data-sovereignty concerns, reliability risks from network or cloud outages, and long-context capabilities. At the same time, agent frameworks are pushing AI beyond chat and into multi-step workflows, such as organizing files, building local knowledge bases, generating reports, and executing tasks over time.
Against this backdrop, Tiiny AI sees the emergence of "AgentBox" — a new class of compact local AI devices designed to support always-on agent workflows. Tiiny AI Pocket Lab is built to support up to 120B-parameter models locally, while offering privacy, reliability, and greater control as a persistent local AI engine that works alongside users' existing devices.
This "AgentBox" is powered by Tiiny AI's proprietary optimization technologies, including TurboSparse and PowerInfer, designed to improve throughput and efficiency for local inference on compact hardware.
The company offers a ready-to-use open-source ecosystem. Tiiny AI Pocket Lab supports one-click installation of more than 50 leading open-source models including OpenAI GPT-OSS, Llama, Qwen, GLM, Mistral, and Phi, and enables seamless deployment of over 100 popular open-source AI agents such as OpenClaw, OpenCode, Flowise, Presenton, Libra, Bella, and SillyTavern. Users receive continuous updates, including official OTA hardware upgrades.
To further strengthen ecosystem openness, Tiiny AI said the product will also support user-imported .gguf models from Hugging Face beyond the officially adapted mainstream models. The company plans to launch a corresponding model conversion tool in July 2026. SOTA open-source LLMs will receive priority support from Tiiny AI, while longer-tail models can be converted and imported by users based on their own needs. This is intended to make the product not only a local AI device, but also a more controllable local AI runtime environment with greater transparency and user oversight.
Built for Local-First Security and Control
As AI agents become more capable and more autonomous, security concerns are expanding beyond the model itself into broader areas such as permission management, behavior auditing, and execution control.
In response, Tiiny AI emphasizes three design principles: local-first operation, least privilege, and human-in-the-loop control. The company says sensitive workflows should remain locally processed whenever possible; agent access to tools, files, and APIs should be limited to only what is necessary for a given task; and human confirmation or audit mechanisms should remain in place for actions involving sensitive data, external systems, or critical operations.
According to Tiiny AI, this is part of what distinguishes AgentBox from a generic local AI setup and cloud-based AI: it is meant to provide a more transparent and controllable environment for running AI continuously in everyday workflows.
In December 2025, Tiiny AI Pocket Lab was verified by Guinness World Records under the category "The Smallest MiniPC (100B LLM Locally)", recognizing its form factor and ability to run a 100B-parameter LLM locally.
Formed in 2024, the team brings together global engineers from MIT, Stanford, HKUST, SJTU, Intel, and Meta, with deep expertise in systems engineering, AI inference optimization, and hardware–software co-design. Their research has been published in premier academic conferences including SOSP, OSDI, ASPLOS, and EuroSys. In 2025, Tiiny AI secured a multi-million dollar seed round from leading global investors.
Headline specs (configuration dependent):
- Memory: 80GB LPDDR5X
- Storage: 1TB PCIe 4.0 SSD
- Performance: ~18–40 tokens/second (config dependent)
- Power: 30W TDP (65W adapter required); designed within a 65W power envelope
- Dimensions/weight: 142 × 80 × 22 mm; 300g
- Compatibility: macOS & Windows
(Note: performance figures are configuration dependent; Tiiny AI can provide test conditions upon request.)
Press kit:
- Photo: https://drive.google.com/drive/folders/1tKjR0GlV827FTm15Fso5lmimtgFZNy2x?usp=sharing
- Demo videos: https://drive.google.com/drive/folders/1B9s7veRCiedIIdFeec7YiN8i4_78fhnM?usp=drive_link
- Corporate video: https://drive.google.com/file/d/168GlrOr2iC1zVWmbjGFuQhf0htrqOa48/view?usp=sharing
###
About Tiiny AI Inc.
Tiiny AI Inc. is a U.S. AI Infra startup. The company's mission is to make advanced AI personal, private, and accessible, bringing powerful model and agent workflows to local devices with simplified deployment and predictable costs. With its founding team formed in 2024 by engineers from MIT, Stanford, HKUST, SJTU, Intel, and Meta, the company develops breakthrough technologies such as TurboSparse and PowerInfer that enable up-to-120B LLMs to run on pocket-sized consumer devices for the first time. Tiiny AI's mission is to make advanced AI accessible, private, and personal.
** This press release is distributed by PR Newswire through automated distribution system, for which the client assumes full responsibility. **
AgentBox Emerges as Tiiny AI Pocket Lab Hits $1M in 5 Hours on Kickstarter, a Shift from Cloud AI to Edge-Cloud Synergy
![[MWC 2026] GSMA Releases Experience Specifications for AI Calling Native Applications](https://image.bastillepost.com/704x369/wp-content/uploads/global/2026/03/media2_2932316_image1_FB.jpg.webp)
