What's taking shape.

I started working on Josephine in July, and the system tray is now ready. That is a small milestone, but an important one: the tray is meant to be Josephine's quiet home on Windows, keeping the project present without leaving a big application window open all day. I am building the interface with PyQt6, with a tray-first approach that can grow into a lightweight interaction when summoned.

The version in progress right now is Josephine v0 Lite?a deliberately small, strictly offline prototype, not the full long-term vision. Its design is centered on low resource use and short, useful interactions:

- A quantized local Gemma 3B model. - On-demand voice input and concise offline speech output. - A simple route between commands and local reasoning. - No cloud APIs and no heavyweight always-open interface.

The Lite plan also treats resources as a feature: the model should be able to sleep and unload after a few minutes of inactivity, while a minimal tray state and subtle processing feedback keep the experience clear. So far, the tray is the part that is ready; the rest of these capabilities are what I am working toward. v0 is a way to test the core idea carefully before expanding Josephine into something bigger.

For an assistant that lives close to your everyday work, privacy and availability should be part of the design?not settings you have to remember to turn on. Josephine is being shaped around a local core, so ordinary interactions can work on the device without depending on a network connection. That can mean less waiting, fewer privacy trade-offs, and help that is still there when the internet is not.

The experience I am aiming for starts with a quiet tray presence and an intentional shortcut. From there, a person should be able to type or ask by voice, let a local model handle everyday reasoning, and route clear requests to focused Windows actions. The assistant should make its state understandable, ask for context only when it is useful, and avoid taking over the screen or running constantly in the background.

The wider Josephine vision may eventually include optional, clearly signposted handoffs for work that benefits from more compute. But the first Lite version is a stricter experiment:

- Keep the core offline. - Keep the scope small. - Learn whether a local assistant can be genuinely useful before adding more moving parts.

Windows once had Cortana, but after Microsoft retired it, I never felt that a true system-level assistant took its place. There are plenty of chatbots inside apps, but I wanted a companion that could help with the computer itself?without making me open another tab or send every request to the cloud.

As local language models have become more capable, the idea of a private, on-device assistant feels increasingly practical. Josephine is my attempt at that: a quiet Windows companion, available from the system tray, with the potential to connect natural-language requests to useful workflows. The goal is not another chatbot that talks for the sake of talking; it is something that can help me think, find information, and get things done while staying out of the way.

The name comes from Jo March in Little Women. I admire her independence, curiosity, creativity, and directness. I want Josephine to carry some of that spirit?to feel warm and companionable, but also capable and honest, rather than like a generic assistant putting on a personality. This is the beginning of the project, so I am starting small and letting the character and capabilities grow with it.