Kelvin 0.1.0 — first public build

Kelvin is a local-AI photo editor for macOS, and this is a pre-alpha. It is useful, it is
honest about what it can't do yet, and it would rather tell you both up front.

What it does: drop in a photo — RAW, JPEG, HEIC, PNG or TIFF — and a small vision model reads
the scene on your Mac. Kelvin offers three or four finished candidate edits; you pick one, tune
it, and apply it across the shoot — every photo re-read and corrected on its own, the look
adapted per frame rather than copied. The editor has the full adjustment set, a colour mixer,
geometry, and a mask kit (subject, sky, radial, graduated, brush, colour range, luminance, skin —
including one person's skin at a time — and per-person). Masks come with presets, from stormy
skies to soft skin, and any mask you tune can be saved as your own. A Looks library adds film
looks and black-and-white conversions on top of whichever candidate you picked. There's culling
with keep/reject, a filmstrip that groups by burst, day, place or near-duplicate, batch apply
scoped to your keepers if you like, and export to JPEG, HEIC, PNG or 16-bit TIFF.

What it doesn't do yet: preference learning is logged but not wired into a loop — Kelvin
doesn't yet get better from your picks. Auto-masks can be refined and inverted but not brushed by
hand. Nothing generative, on purpose. Known rough edge: the edit panel can stutter while a
render or scene read is in flight — a fix is scoped. Expect pre-alpha manners elsewhere too, and
bug reports are genuinely welcome.

Your photographs stay on your machine. The model ships inside the app — nothing is uploaded,
there's no account and no telemetry. A released copy makes exactly one kind of network request,
and only with your permission: an update check against usekelvin.app, which fetches a few KB of
release metadata and nothing else. Decline it and the app never touches the network at all.

One thing you should know about exports: exported files carry the original photograph's
metadata by default, including its GPS location, exactly as your camera wrote it. There's a
switch in the export panel to strip it — covered by tests that read the exported file back to
check. Kelvin won't silently discard your metadata, and it won't silently publish your location
either; the choice is visible and it's yours.

How it was built: by one person working with an AI assistant (Claude), and the commit history
says so on every commit rather than hiding it. The history is kept un-squashed on purpose — it
records what was measured, what was reversed, and why, which is worth more than a tidy graph.

Requirements: an Apple Silicon Mac on macOS 14 or later. The download is large (~1.4 GB)
because the vision model travels inside the app; that's the price of nothing ever being fetched
behind your back.

Found a problem? Report a bug —
the app can prefill the report with your build and macOS version from Settings. Security issues:
see SECURITY.md.