Snapsort

Everyone screenshots. No one files.

My desktop was a graveyard of Screenshot 2026-01-14 at 3.42.17 PM.png. So was every designer's and developer's I asked. The screenshot is the fastest capture tool we have and the worst retrieval tool.

Invisible until useful

Snapsort lives in the menu bar, not the dock. No window to manage, no app to remember. It watches, sorts, and gets out of the way. That position drove every choice after it.

Latency, cost, and privacy are design decisions

The pipeline is hybrid by design. A 2,787-line local classifier on Apple Vision and OCR handles the easy majority: free, instant, nothing leaves the Mac. Only files below a 0.75 confidence gate go to a cloud model, worst first.

One provider abstraction speaks to four APIs plus any OpenAI-compatible endpoint, so local models are welcome too. Images resize to an 800px edge before upload. Rate limits honor Retry-After with a cooldown clock. Keys live in the Keychain, validated on entry.

The prompt is a production surface

Classification quality lives or dies in the prompt, so I treated it like shipped UI: 387 lines, versioned, loaded at runtime so I can iterate without recompiling. Its output becomes real folder names, so it ends in a strict JSON contract.

RULE 1: GROUND TRUTH IS THE IMAGE
Classify only what you can SEE. If a brand mark is not visible, the brand is unknown.

RULE 2: BRAND BEATS GENERIC
"Stripe" beats "Dashboard". "Figma" beats "App UI".

ANTI-HALLUCINATION GUARD
When you cannot identify the brand AND cannot confidently identify the
content type, set folder to "Unsorted" with confidence ≤ 0.5.
"Unsorted" is honest; a wrong guess is not.

OUTPUT FORMAT (STRICT)
Return ONLY this JSON. No markdown fences. No commentary.
{"category":"...","folder":"...","confidence":0.0-1.0,"description":"snake_case_tokens"}

What the model sees is designed

The image alone is not enough context, and too much context invites hallucination. Every input the model receives is deliberate, weighted, and bounded.

An AI touching your files must earn it

This is where UX craft matters most in AI products. Snapsort acts on a user's file system, so every action is consented, disclosed, and reversible.

15 testers turned false positives into rules

I ran a beta with 15 developers and designers, the exact audience the classifier serves. Their real screenshot libraries became my eval set. Every misclassification was traced to its cause and became a permanent rule in the prompt or the local classifier.

• "Ask anything" is not ChatGPT. That phrase appears in Perplexity, Gemini, Claude, and others. Generic chat bars now route to Chats, brand only when a brand is visible.
• One keyword does not make a form. The form rule tightened to require 3+ matching signals after false positives on unrelated UI.
• Weak phrases are traps. "Inbox zero", "design system", "compose" got removed from the catalog entirely, with comments documenting why.
• The app must never classify itself. An anti-recursion guard stops Snapsort's own UI from creating a "Snapsort" folder loop.
• Canonical plurals. Login/Logins chaos normalized to one convention across the AI and local paths.

Built with AI, shipped like a product

The whole build ran through CLI agents, with Claude Code as the daily driver and every line reviewed by hand. Then the unglamorous last mile: signing, notarization, a DMG pipeline, and a landing page.