Focus67Dumby
A fully offline floating sticky-note style focus tool. It uses a small local Ollama model to analyze active window titles and downsampled screenshots, distinguishing productive work, related tutorials and entertainment content, and sends no data to the cloud.
Timeline (last 90 days)
Focus67Dumby is a 350x530 dark amber floating sticky-note app that runs 100% offline and uses a small local open-source model scheduled by Ollama to judge whether the user is currently working or distracted.
Focus67Dumby is a submission for the second week of the Hacktoberfest open-source AI challenge.
Focus67Dumby's tech stack is Python 3.11, using CustomTkinter, Win32 ctypes and Pillow, and calls open-weight models through Ollama's local REST API (127.0.0.1:11434).
Focus67Dumby's model dropdown discovers locally installed models in real time via Ollama's GET /api/tags, allowing users to hot-switch according to their hardware.
Focus67Dumby's perceptual-diff caching mechanism scales the in-memory screenshot to 512x288 (JPEG quality 65) and computes a 16x9 grayscale perceptual signature; if the pixel change is below 3% and the window title is the same, it hits the local session cache, achieving 0 ms latency and 0 GPU calls.
In the first 30 seconds of a sprint, Focus67Dumby enters a workspace calibration phase, automatically adding windows the user has touched to the trusted workspace and maintaining a rolling 4-window breadcrumb trail.
Focus67Dumby uses a two-strikes strategy with escalating interventions: when two consecutive checks find the user on an off-task window, it escalates the intervention according to the selected level, from native Windows notifications to notification plus sound, and then to a pop-up takeover plus sound (forcing topmost via Win32 WS_EX_APPWINDOW, flashing a red border, and shaking the sticky note left and right).
All GUI, screen-capture and Win32 imports in Focus67Dumby are wrapped in lazy-loading guards, allowing a deterministic offline CLI simulation to run in a headless Linux Docker container (python -m src.main --headless --demo) and passing all unit tests within 0.75 seconds.
At one poll every 10 seconds over an 8-hour workday, each user generates 2,880 multimodal API calls per day; at closed-source API pricing, a heavy user's token cost is $20 to $40 or more per month.
Focus67Dumby's Pro desktop edition uses a $19 one-time lifetime license, while the core repository remains 100% open source.
1 more timeline events
All Facts (11)
Feedback from Focus67Dumby's first real user exposed two problems: workflows require jumping across multiple windows, and after unchecking "always on top" on the borderless Windows sticky note it may be covered by the browser; this feedback gave rise to the 30-second workspace calibration feature and a Win32 taskbar persistence fix.
50%UnverifiedFocus67Dumby is a 350x530 dark amber floating sticky-note app that runs 100% offline and uses a small local open-source model scheduled by Ollama to judge whether the user is currently working or distracted.
50%UnverifiedFocus67Dumby's tech stack is Python 3.11, using CustomTkinter, Win32 ctypes and Pillow, and calls open-weight models through Ollama's local REST API (127.0.0.1:11434).
50%UnverifiedFocus67Dumby's model dropdown discovers locally installed models in real time via Ollama's GET /api/tags, allowing users to hot-switch according to their hardware.
50%UnverifiedFocus67Dumby's perceptual-diff caching mechanism scales the in-memory screenshot to 512x288 (JPEG quality 65) and computes a 16x9 grayscale perceptual signature; if the pixel change is below 3% and the window title is the same, it hits the local session cache, achieving 0 ms latency and 0 GPU calls.
50%UnverifiedIn the first 30 seconds of a sprint, Focus67Dumby enters a workspace calibration phase, automatically adding windows the user has touched to the trusted workspace and maintaining a rolling 4-window breadcrumb trail.
50%UnverifiedFocus67Dumby uses a two-strikes strategy with escalating interventions: when two consecutive checks find the user on an off-task window, it escalates the intervention according to the selected level, from native Windows notifications to notification plus sound, and then to a pop-up takeover plus sound (forcing topmost via Win32 WS_EX_APPWINDOW, flashing a red border, and shaking the sticky note left and right).
50%UnverifiedAll GUI, screen-capture and Win32 imports in Focus67Dumby are wrapped in lazy-loading guards, allowing a deterministic offline CLI simulation to run in a headless Linux Docker container (python -m src.main --headless --demo) and passing all unit tests within 0.75 seconds.
50%UnverifiedAt one poll every 10 seconds over an 8-hour workday, each user generates 2,880 multimodal API calls per day; at closed-source API pricing, a heavy user's token cost is $20 to $40 or more per month.
50%UnverifiedFocus67Dumby's Pro desktop edition uses a $19 one-time lifetime license, while the core repository remains 100% open source.
50%UnverifiedFocus67Dumby is a submission for the second week of the Hacktoberfest open-source AI challenge.
50%