Someday: Your Screenshots Are a To-Do List. Someday Finishes Them.

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

Every smartphone and desktop user has a digital graveyard of screenshots:

Screenshots are the lowest-friction capture mechanism ever invented. But while saving a screenshot takes a split second, acting on them is tedious, manual drudgery. You have to open Google Calendar, decode relative dates, copy venues, switch between apps, or manually type receipt figures into a spreadsheet. Because this work is repetitive, it gets postponed indefinitely.

Someday is built for the Slow and Steady ethos. It rewards laziness:

Capture it once. Let Someday do the tedious part. Review the results over coffee.

Someday monitors your screenshot directory in the background. When new screenshots appear, an open-weight vision-language model inspects the image, extracts the critical details, resolves relative dates into exact calendar timestamps, and logs everything in a calm, paper-inspired Morning Ledger. You don't have to chat with a bot or prompt an AI. You simply wake up, review the ledger, approve with one click, and export valid .ics calendar events, CSV expense sheets, or reading lists.

http://localhost:3000 (Frontend) & http://127.0.0.1:8000 (FastAPI Local Backend).ics) for Apple Calendar, Google Calendar, and Outlook.

┌────────────────────────────────────────────────────────┐
│               The Morning Ledger UI                    │
├────────────────────────────────────────────────────────┤
│ [📅 Event] Open Source AI Summit · Tomorrow 7 PM       │
│    Quote: "Tomorrow at 7 PM - 10 PM at Hacker Hub"     │
│    [Approve] -> Exports to Calendar (.ics)             │
├────────────────────────────────────────────────────────┤
│ [🧾 Receipt] Blue Bottle Coffee · $17.09 USD           │
│    Quote: "TOTAL AMOUNT PAID $17.09 USD"               │
│    [Approve] -> Exports to Expense CSV                 │
├────────────────────────────────────────────────────────┤
│ [📖 Reading] How To Architect Reliable Local AI        │
│    Quote: "The Pragmatic Engineer · Gergely Orosz"     │
│    [Approve] -> Exports to Reading List JSON           │
└────────────────────────────────────────────────────────┘

Your screenshots are a to-do list. Someday finishes them.

🌐 Live Vercel Interactive Showcase: https://someday-ledger.vercel.app

🎬 1-Minute Demo Video (AI Voiceover): Watch Demo Video

🖥️ Local-First Privacy Application: http://localhost:3000 (Next.js) + http://127.0.0.1:8000 (FastAPI + Ollama)

Someday is an open-source, local-first personal AI utility that turns forgotten screenshot folders into completed tasks. Instead of letting screenshots of event posters, dinner receipts, reading lists, and articles accumulate in digital oblivion, Someday monitors your screenshot folder, runs a local open-weight vision-language model, extracts structured actions with evidence grounding, and presents them in a calm morning ledger for your approval.

Every day, people take screenshots of things they intend to do later:

👉 GitHub Repository: https://github.com/Shreyansh00987/Someday

python-dateutil, icalendar. moondream 1.4B and gemma3:1b). Someday is architected around the principles of local-first privacy, deterministic guardrails, and asynchronous laziness:

Passive Folder Watcher (backend/watcher.py):

Open-Weight Vision AI Pipeline (backend/inference.py):

moondream:latest 1.4B with multimodal CLIP projector, and event, receipt, reading, or uncertain. Deterministic Date Resolution Engine (backend/date_engine.py):

The Morning Ledger Frontend (frontend/src/):

#F7F5EE), deep charcoal ink typography (#181613), British racing green accents (#1E3F2E), and polaroid contact-sheet thumbnails. Open innovation is not just an implementation detail for Someday — it is the only reason the product is viable:

Zero Data Leakage for Sensitive Personal Information:

Screenshots frequently contain banking balances, home addresses, private event invites, and medical appointment slips. Sending these to third-party proprietary cloud APIs is an unacceptable privacy hazard. Running open-weight models locally ensures not a single byte leaves the user's machine.

True Offline Independence:

Once installed, Someday works on an airplane, in a remote cabin, or during an internet outage. It requires zero active internet connection to transcribe and organize your day.

Freedom from Per-Token Tolls:

Proprietary vision APIs charge steep per-image fees. A backlog of 1,000 screenshots could cost tens of dollars. With open-weight models running on local hardware (our tests ran on a consumer laptop NVIDIA RTX 4050 GPU with 6GB VRAM), processing 10,000 screenshots costs $0.00 in API fees.

Model Substitutability:

Users are never locked into a single provider. In Someday's settings, you can swap between moondream, gemma3:1b, minicpm-v, or llava with a single click.

We built an automated evaluation harness (backend/eval_harness.py) and benchmarked the workflow against real screenshot test fixtures:

In International Sloth Day spirit: Capture once, do nothing, review over coffee.