[{"element_type":"markdown","content":"\u003cp\u003eI’m Jacob, the builder behind DeCampLabs. PHNTM One is my attempt to give everyday AI tasks a dedicated place on the desk, using a Raspberry Pi rather than depending on a cloud AI account.\u003c/p\u003e\n\n\u003cp\u003e\u003cimg src=\"https://www.phntmcore.com/images/phntm-one-standby-20261008-1448.webp\" alt=\"PHNTM One showing its standby screen beside a compact keyboard\"\u003e\u003c/p\u003e\n\n\u003cp\u003e\u003cem\u003eOwner-supplied image exported from ChatGPT, October 8, 2026. Screen details may differ from the current interface.\u003c/em\u003e\u003c/p\u003e\n\n\u003ch2\u003eWhat I built\u003c/h2\u003e\n\n\u003cp\u003ePHNTM One combines a Raspberry Pi 5 with 8 GB of memory, a 10.1-inch touchscreen, 256 GB microSD storage, a microphone, and a speaker. A compact wireless keyboard with a touchpad gives another way to interact with it.\u003c/p\u003e\n\n\u003cp\u003eThe idea is simple: capture a thought, ask about a document, or pick up a project without opening another cloud chat on the computer you are already working on.\u003c/p\u003e\n\n\u003ch2\u003eWhat it does\u003c/h2\u003e\n\n\u003cul\u003e\n\u003cli\u003e\n\u003cstrong\u003eLocal conversation:\u003c/strong\u003e ask questions through text or voice. In Private mode, the model runs on the device.\u003c/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eNotes and ideas:\u003c/strong\u003e type or dictate a thought, then ask for help organizing or developing it. The original stays separate from the AI draft, so you can review the suggestions.\u003c/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eDocument questions:\u003c/strong\u003e bring in PDF, text, and Markdown files and ask about them. Answers can include file and page references when supporting passages are found. References still need checking; AI can make mistakes.\u003c/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eMemory:\u003c/strong\u003e explicitly save a useful detail, review it later, correct it, or ask the device to forget it.\u003c/li\u003e\n\u003c/ul\u003e\n\n\u003cp\u003eLocal chat, memory, Notes, document questions, and voice can work without internet. Optional connected services have separate data paths; enabling cloud-assisted answers can send the question and selected context to that provider.\u003c/p\u003e\n\n\u003cp\u003e\u003cimg src=\"https://www.phntmcore.com/images/phntm-one-color-dashboard-20261008-1448.webp\" alt=\"PHNTM One dashboard and compact keyboard in colored light\"\u003e\u003c/p\u003e\n\n\u003cp\u003e\u003cem\u003eOwner-supplied image exported from ChatGPT. This picture is not a performance measurement.\u003c/em\u003e\u003c/p\u003e\n\n\u003ch2\u003eThe software and the trade-offs\u003c/h2\u003e\n\n\u003cp\u003eThe device uses a Python application, Ollama for local model inference, Gemma 3 4B for the primary chat model, and local speech tools. The touchscreen is the main interface.\u003c/p\u003e\n\n\u003cp\u003eRunning a small model on a Raspberry Pi means accepting limits. Answers can take tens of seconds, longer Notes can take several minutes, and complex reasoning is stronger on larger cloud models. PHNTM One can also give a wrong answer. My focus is making the device useful while keeping those limits visible.\u003c/p\u003e\n\n\u003ch2\u003eSee it in use\u003c/h2\u003e\n\n\u003cul\u003e\n\u003cli\u003e\u003ca href=\"https://www.phntmcore.com/films\"\u003eWatch the recorded demonstrations\u003c/a\u003e\u003c/li\u003e\n\u003cli\u003e\u003ca href=\"https://www.phntmcore.com/#gallery\"\u003eExplore the device photographs\u003c/a\u003e\u003c/li\u003e\n\u003cli\u003e\u003ca href=\"https://www.phntmcore.com/manual\"\u003eRead the owner manual and limitations\u003c/a\u003e\u003c/li\u003e\n\u003cli\u003e\u003ca href=\"https://www.phntmcore.com/\"\u003eVisit the project website\u003c/a\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\n\u003cp\u003eThe demonstrations use real device footage, with edits and presentation details explained on the website. They are not response-speed benchmarks.\u003c/p\u003e\n\n\u003cp\u003eI’d love feedback from other makers: what everyday task would make a local AI desk companion worth keeping beside your keyboard?\u003c/p\u003e\n","metadata":{"markdown":"I’m Jacob, the builder behind DeCampLabs. PHNTM One is my attempt to give everyday AI tasks a dedicated place on the desk, using a Raspberry Pi rather than depending on a cloud AI account.\n\n![PHNTM One showing its standby screen beside a compact keyboard](https://www.phntmcore.com/images/phntm-one-standby-20261008-1448.webp)\n\n*Owner-supplied image exported from ChatGPT, October 8, 2026. Screen details may differ from the current interface.*\n\n## What I built\n\nPHNTM One combines a Raspberry Pi 5 with 8 GB of memory, a 10.1-inch touchscreen, 256 GB microSD storage, a microphone, and a speaker. A compact wireless keyboard with a touchpad gives another way to interact with it.\n\nThe idea is simple: capture a thought, ask about a document, or pick up a project without opening another cloud chat on the computer you are already working on.\n\n## What it does\n\n- **Local conversation:** ask questions through text or voice. In Private mode, the model runs on the device.\n- **Notes and ideas:** type or dictate a thought, then ask for help organizing or developing it. The original stays separate from the AI draft, so you can review the suggestions.\n- **Document questions:** bring in PDF, text, and Markdown files and ask about them. Answers can include file and page references when supporting passages are found. References still need checking; AI can make mistakes.\n- **Memory:** explicitly save a useful detail, review it later, correct it, or ask the device to forget it.\n\nLocal chat, memory, Notes, document questions, and voice can work without internet. Optional connected services have separate data paths; enabling cloud-assisted answers can send the question and selected context to that provider.\n\n![PHNTM One dashboard and compact keyboard in colored light](https://www.phntmcore.com/images/phntm-one-color-dashboard-20261008-1448.webp)\n\n*Owner-supplied image exported from ChatGPT. This picture is not a performance measurement.*\n\n## The software and the trade-offs\n\nThe device uses a Python application, Ollama for local model inference, Gemma 3 4B for the primary chat model, and local speech tools. The touchscreen is the main interface.\n\nRunning a small model on a Raspberry Pi means accepting limits. Answers can take tens of seconds, longer Notes can take several minutes, and complex reasoning is stronger on larger cloud models. PHNTM One can also give a wrong answer. My focus is making the device useful while keeping those limits visible.\n\n## See it in use\n\n- [Watch the recorded demonstrations](https://www.phntmcore.com/films)\n- [Explore the device photographs](https://www.phntmcore.com/#gallery)\n- [Read the owner manual and limitations](https://www.phntmcore.com/manual)\n- [Visit the project website](https://www.phntmcore.com/)\n\nThe demonstrations use real device footage, with edits and presentation details explained on the website. They are not response-speed benchmarks.\n\nI’d love feedback from other makers: what everyday task would make a local AI desk companion worth keeping beside your keyboard?\n"}}]