Personal Instruments
Active development

SnipDig

A local screenshot retrieval tool that recognizes text in selected images, explains deterministic matches, and opens the source at the relevant text region.

Category: Personal Instruments
Platform: iPhone + iPad + Mac
Status: Active development

What if you could find a screenshot by the words visible inside it?

Selected import, app-managed copies, Apple Vision OCR, transparent matching, region highlighting, and live saved queries make retrieval useful without broad photo access, cloud services, or semantic guessing.

User value

Find the screenshot by the words inside it.

Recover reservation codes, receipts, addresses, error messages, and reference notes from selected screenshots without browsing by date or thumbnail alone.

Import only what you choose

Select images through PhotosPicker on iPhone and iPad or through file selection and drag-and-drop on Mac without granting a full photo-library scan.

Search the visible text

Match exact words, phrases, case variants, normalized punctuation and spacing, and optional conservative typos through deterministic local rules.

Return to the source region

Open the screenshot from a ranked result and emphasize the recognized-text block that produced the match.

SnipDig macOS Search screen showing a query for receipt and one fictional result with recognized text, match count, OCR confidence, and match type. — **Current deterministic search**Current macOS Debug capture using fictional demo records and the mock runtime.

Features

A local path from selected image to exact text region.

The current code connects import, managed image storage, OCR, explainable search, screenshot detail, and live saved queries across Apple devices.

01
Selected multiplatform import
Use PhotosPicker on iPhone and iPad or PNG, JPEG, HEIC, and TIFF file selection and drag-and-drop on Mac.
02
Managed local image copies
Copy accepted bytes into SnipDig storage, generate local thumbnails, filter duplicates through source identifiers and hashes, and leave the original item untouched.
03
Apple Vision OCR pipeline
Queue single or batch recognition, preserve original text, confidence, and normalized bounding boxes, and represent pending, completed, failed, and retry states.
04
Deterministic OCR search
Support exact, literal phrase, case-insensitive, normalized, optional diacritic-insensitive, and option-gated typo matching without semantic expansion.
05
Ranked, explained results
Return original-text snippets and match labels, rank stronger evidence above weaker evidence, and carry the matched OCR block into screenshot detail.
06
Views as live queries
Save, rename, edit, pin, unpin, or delete a query that reruns against current OCR data, with optional deterministic suggestions from local usage signals.

How it works

Select, recognize, search, inspect.

SnipDig makes screenshots retrievable without asking the user to classify them into a new folder system.

01
Choose the images
Select screenshots or other supported images through the platform-specific picker or Mac drag-and-drop surface.
02
Create managed copies
Hash and filter duplicates, copy accepted bytes into local application storage, and generate session thumbnails.
03
Recognize the text
Run Apple Vision through the live service boundary to produce OCR blocks with text, confidence, and source-image coordinates.
04
Search remembered words
Enter a literal query and apply visible normalization or optional typo settings without expanding it into hidden related concepts.
05
Inspect the matched region
Open the ranked result, see the recognized block in the original image, and copy selected, matched, or complete OCR text.
06
Save repeated retrieval
Create a View that reruns the query against the current index without owning or duplicating screenshots.

Current macOS demo

Saved queries and selected import stay explicit.

These Debug captures use fictional records and deterministic mock services. They demonstrate implemented interfaces, not the unresolved production persistence and live-runtime wiring.

SnipDig macOS Views screen with five fictional saved OCR queries, one pinned View, and a Suggested Views section. — Current fictional demo Views in the mock runtime.

Views are searches, not folders

Manual and suggested Views store query metadata and rerun against the current OCR index without owning image or result records.

SnipDig macOS Library showing ten fictional screenshot records and a selected local-import explanation. — Current fictional Library state in the mock runtime.

Library import boundary

The Library explains selected import, app-managed copies, local OCR, duplicate protection, and the absence of a full-library scan.

Product concept

The screenshot is the source. OCR is the index.

Useful screenshot details can become searchable without replacing the image, forcing folder organization, or inferring concepts that are absent from recognized text.

Retrieval before organization

People can keep the existing screenshot habit and recover a detail from remembered words without classifying every image in advance.

Tolerance without hidden meaning

Normalization and optional typo matching can handle OCR and memory imperfections while keeping exact source text, rank order, and match explanations visible.

A View is a query, not a collection

Saved retrieval stays current as the library changes and can be deleted without affecting screenshots, OCR blocks, or manual content ownership.

Architecture

A single-process local OCR and retrieval pipeline.

Shared SwiftUI navigation, platform-neutral domain models, actor-backed services, local files, Apple Vision, deterministic search, and geometry mapping support iPhone, iPad, and Mac without a SnipDig server.

01Selected image payload
02Duplicate filter and managed local copy
03Apple Vision OCR blocks
04Deterministic matching and ranking
05Matched block in screenshot detail
06Optional live saved query

Import remains selected

Platform pickers produce neutral payloads, original files remain outside SnipDig's deletion boundary, and only managed copies and records are removed.

Implemented work stays in process

OCR, matching, ranking, history signals, Views, and suggestions run locally with no network, account, analytics, CloudKit, or third-party package dependency.

Original text remains available

Normalized representations support matching, while source OCR text and mapped bounding boxes remain the basis for snippets, copying, and image-region emphasis.

Demo and live environments stay distinct

A live environment selects the Apple Vision service, but the current application entry point launches the mock environment with fictional records and in-memory repositories.

Technology

Apple-native, local, and explainable.

Application

Swift
SwiftUI
iOS and iPadOS 17+
macOS 14+

Import and images

PhotosUI
macOS file picker
Drag and drop
ImageIO
CryptoKit hashes

OCR and search

Apple Vision
Original OCR blocks
Deterministic normalization
Transparent ranking

Current storage

Managed local image files
Session thumbnail cache
In-memory metadata repositories

Current boundary

No networking
No accounts
No analytics
No cloud sync
No generated helper output

Development

The retrieval system is built. Production storage and runtime wiring are not.

Seven implementation phases cover adaptive interfaces, selected import, managed image files, Vision OCR service code, deterministic retrieval, source-region navigation, and live-query Views. The current executable still launches fictional mock services, and metadata, OCR blocks, settings, Views, history, and usage signals disappear with the process.

Stage: Active development
Platforms: iPhone · iPad · Mac
Automated suite: 573 tests · 0 failures
Default runtime: Mock demo environment

Implemented

Shared SwiftUI Search, Views, Library, Settings, import, and screenshot-detail interfaces across three Apple platforms
Selected PhotosPicker, macOS file-picker, and drag-and-drop import with app-managed files, thumbnails, and duplicate filtering
Apple Vision OCR service, serial queue and batch models, confidence controls, OCR blocks, bounding boxes, and retry states
Exact, phrase, case-insensitive, normalized, optional diacritic-insensitive, and option-gated typo matching with deterministic ranking
Original-text snippets, match labels and explanations, screenshot-detail routing, region geometry, copy actions, and basic zoom
Manual live-query Views, deterministic suggestions, local usage-signal controls, helper policy scaffolding, and broad XCTest coverage

Before a dependable personal library

Implement durable local persistence behind the existing repositories, make environment selection explicit, and exercise a complete live import-to-Vision-to-search flow outside test doubles. Reconcile stale typo copy and literal OR demo queries, complete hands-on accessibility review, and capture a real staged image with its matched OCR region before expanding optional helper, Spotlight, or App Intents work.

SnipDig lab notes

The project, over time.

Notes about the thinking, architecture, development, and decisions behind this work.

Find the screenshot by the words inside it.

Import only what you choose

Search the visible text

Return to the source region

A local path from selected image to exact text region.

Selected multiplatform import

Managed local image copies

Apple Vision OCR pipeline

Deterministic OCR search

Ranked, explained results

Views as live queries

Select, recognize, search, inspect.

Choose the images

Create managed copies

Recognize the text

Search remembered words

Inspect the matched region

Save repeated retrieval