Pixel 7 Gets a Second Life as a Handheld AI Cyberdeck Running a Local 3B Model
Older smartphones often end up forgotten in a drawer, but one Pixel 7 owner has shown that aging Android hardware can still be surprisingly useful. Instead of treating the device as obsolete, the enthusiast transformed Google’s former flagship into a compact cyberdeck capable of running an AI model locally, without relying on an internet connection.
The project uses a Pixel 7 powered by Google’s Tensor G2 chip and 8GB of RAM. While the Tensor lineup has never been known as the strongest mobile silicon for raw performance, it still has enough capability to handle smaller AI workloads when configured properly. In this case, the device was used to run Qwen2.5-abliterate:3b, a 3-billion-parameter AI model with its safety filters removed.
The result is a portable, self-contained AI machine that looks like a miniature terminal-style computer. The custom cyberdeck enclosure was reportedly created using 3D-printed CAD files generated with Claude Code. The Pixel 7 sits inside the case, while the screen displays a Linux-like terminal interface with command-line tools. The lower portion of the build appears to have space for a power bank, giving the setup enough battery life to run for longer sessions.
Performance, however, is modest. Running a 3B AI model on a Pixel 7 is possible, but not exactly fast. The device manages around 5 tokens per second, which means responses arrive slowly compared with cloud-based AI tools or more powerful desktop hardware. Still, for a smartphone running a local language model entirely on-device, the project is impressive.
The Tensor G2 and 8GB of memory place clear limits on what the Pixel 7 can realistically handle. Larger and more demanding models would likely be impractical, especially if the goal is to keep the experience usable. Even so, a 3-billion-parameter model can still perform basic tasks, answer simple prompts, assist with lightweight text generation, and offer a private AI experience without sending data to remote servers.
One of the most interesting parts of the project is the offline aspect. Local AI models are becoming more appealing as users look for privacy, portability, and independence from cloud services. A phone-based setup like this may not replace a powerful workstation, but it proves that older mobile devices can still play a role in the growing world of on-device AI.
The use of an uncensored model also adds another layer to the discussion. Models with removed safety filters can be useful for experimentation and research, but they also require responsible handling. As with any AI tool, the value depends heavily on how it is used and whether the outcome remains safe and ethical.
More than anything, this Pixel 7 cyberdeck highlights the creative potential of repurposing old smartphones. Instead of letting unused devices collect dust, hobbyists can turn them into portable terminals, offline AI assistants, home automation controllers, retro gaming machines, or compact development tools.
A 3B AI model running at 5 tokens per second may not sound groundbreaking at first, but the bigger idea is important: even aging Android phones still have enough power to be useful in new and unexpected ways. As AI models become more efficient, projects like this could become much more common, giving older smartphones a second life as personal, private, on-device AI computers.
