A calm macOS browser proof of concept: it observes how it is used (metadata only, locally, with consent) and sparingly offers to fold what it notices back into the interface, chiefly as workspaces.
- Behaviour spec:
Sill-PoC-PRD.md(source of truth for what happens, when) - Visual spec:
Project Browse — Design Package.pdf(source of truth for everything visual) - Stack: Swift + SwiftUI, WKWebView on system WebKit, SQLite. No network calls of its own, no telemetry, no LLM. PoC, light mode first.
make run # release build → build/Sill.app → open
make debug # debug configuration
make cleanRequires Xcode (tested with 26.6 / Swift 6.3 on macOS 26).
| Milestone | Status | |
|---|---|---|
| M0 | Login viability spike | Passed — Google/GitHub/Microsoft sign-ins persist across restart (docs/M0-login-spike.md) |
| M1 | Shell (sidebar-first, D2a) | Built — awaiting the owner's full-workday gate (docs/M1-shell.md) |
| M2 | Workspaces, hibernation, benchmark | Built — H4 numbers need a clean measurement run (docs/M2-workspaces.md) |
| M3 | Consent + history import | Built — owner to grant FDA + run Safari import (docs/M3-consent-import.md) |
| M4 | Learning engine | Built — harness 4/4, 1 FP (docs/M4-learning-engine.md); real-history review pending import |
| M5 | Home, cards, Learning page | Built — demo-seed verified; owner to drive confirm flow (docs/M5-surfaces.md) |
| M6 | Palette + instrumentation | Built (docs/M6-palette-instrumentation.md) — fixed a real-URL-guessing bug found via owner's import |
| M7 | Sense-check fortnight (no code) | Ready to start — all milestones M0–M6 built |
H1 detection ≥50% confirmed · H2 at least one genuine surprise · H3 trust ("my data, shown to me") · H4 measurable lightness vs Arc/Chrome · H5 daily-drivable for two weeks. A clean negative on any of these is a successful PoC outcome.