Most "AI assistant" tools wait for you to ask. The bet here is the opposite: the model that already watched your whole day knows what to surface before you type anything. By capturing screenshots as the primary signal and running them through a vision-language model, it reconstructs context you never bothered to log — and turns that into summaries and todos that show up on their own.
What Sets It Apart
- Screenshots as the primary context source, not files or chat history — it sees what you actually did across every app, which is far richer than activity logs that only timestamp window switches.
- A real context lifecycle, not a retrieval bolt-on: capture, document chunking, entity extraction and normalization, then merging and deduplication before anything reaches the LLM. This is why insights stay coherent over days instead of returning raw search hits.
- Local-first by construction — SQLite plus ChromaDB on disk, no cloud dependency, and it can drive locally-run models through LMStudio. The cost and privacy story is structurally different from ChatGPT Pulse.
- Proactive delivery: daily and weekly recaps plus actionable todos are pushed to the homepage, rather than waiting behind a prompt.
Who It's For
Great fit if you are a knowledge worker, researcher, or student drowning in scattered context and willing to let a tool watch your screen in exchange for automatic recall. The Electron desktop app runs on macOS and Windows. Look elsewhere if continuous screen capture is a non-starter for you, if you need a hosted multi-device service rather than a single-machine tool, or if you want polished consumer reliability — this is early (v0.1.x) and aimed at people comfortable wiring up their own model endpoints.
Where It Fits
Against Dayflow it offers richer proactive insight than plain activity logging plus context-aware Q&A; against ChatGPT Pulse the trade is local processing and open-source customization for lower API cost, at the price of running infrastructure yourself.
