DigUp Plus 0.1.9

Windows portable build, 64-bit Windows 10/11. Unpack the zip anywhere and run DigUpPlus.exe — nothing is installed, and settings, logs, the index and the model live in the data folder beside it. The build isn't signed: choose More info → Run anyway if Windows asks. The zip is ~470 MB, mostly the CUDA runtime the helper carries so the model runs on NVIDIA cards.

SHA-256 of DigUpPlus-0.1.9-windows-x64.zip: 14b4955244a6c8d855da5c68dffe122338b984aeb9aeb3b93349f8edae4a61c1

Updating from 0.1.6: the helper changed in every release since, so take the whole zip, not just DigUpPlus.exe. Close DigUp Plus first, unpack it over the old folder, and keep your data folder — the portable build keeps everything there. The index upgrades itself on the first start. Recordings that 0.1.6 indexed by their tags are read once more (one ffprobe each) so the window and the panel can show their title, artist and album; until then the card shows the tag line the search matched.

What is in 0.1.9

Recordings are read per folder. Each folder's Rules (Settings → Folders, and the Add Folders window) now say how its recordings are read: Audio by tags and/or Audio by content. Tags is a moment per file (title, artist, album, genre, year, track from the container, no sound decoded); content decodes and embeds the sound. A folder set to tags only reads the types listed under “Recordings read by their tags”, and its other recordings are left out with the reason shown. The tag card in the window shows the fields and plays the file, search finds it by tag words, and the file's line says “by tags”. Choosing a rule re-reads exactly the recordings whose way of being read changed, and the time estimate on the Folders tab counts the rules as they are on screen, before saving. Adding folders no longer starts indexing first: the native picker opens, the rules are set before the first pass.

Rescan and Re-read one folder. Settings → Folders has two links on every searched folder. Rescan looks through that folder again — new, gone and changed files — and reads nothing again, so nothing leaves the results. Re-read… asks first, with the folder's own estimate, then queues every file in it, drops its old text and vectors, and reads it again; a file that's gone drops out instead of staying queued. Other folders are not touched.

Indexing holds up on real disks. A storm of dropped change events is one full sync a minute, not one per event. A file still being written (a video being copied, a PDF being printed) is looked at again until its size and time settle, for ten seconds at most, instead of being read half-done and failing. On disks where the system refuses a watch (some network and exFAT drives) their folders are synced every five minutes and the status line says “changes aren't watched”. Programs' leftovers never enter the index: Word's ~$report.docx lock beside an open document, *.tmp, *.swp, *.crdownload, desktop.ini, Thumbs.db.

Freshness is visible. Settings shows “last sync 2 min ago” beside the counts, and the window's status line ends with “synced 2 min ago” (redrawn every half minute), so it's clear how current the index is.

Agents can search, read-only. DigUp Plus has an MCP server: digup mcp speaks JSON-RPC over stdin/stdout with three tools — search (10 results a page, 50 at most, with snippets and the place that matched), read_snippet (a window of one page, passage or code stretch, 32–1024 tokens) and index_state (counts, the model, the index's age). Answers stay within about 1500 tokens; an agent can only read files the index holds and, when folders are given, only under them. Nothing writes, moves or deletes. Settings → Storage → Agents copies the config for Claude Desktop, Claude Code and Cursor (the executable, index, chosen folders and model location as absolute paths), and digup mcp --print-config prints the same for a terminal.

Fixed along the way:

• A helper started just to rescan or re-read a folder lost the request and exited before doing it; it now applies the work and only then decides there is nothing to do.
• A folder's Re-read could sit behind a long indexing pass; it now yields to it at the first file boundary.
• read_snippet without a segment id answered with the file's name entry — the folder path — instead of its text, and answered the same for pictures; it now returns the matched passage, or says there is no text.

The plans, measurements and checks behind these changes are in docs/per-folder-rules-plan.md, docs/kensaku-ideas-plan.md and docs/agent-access.md.

DigUp Plus 0.1.6

Windows portable build, 64-bit Windows 10/11. Unpack the zip anywhere and run DigUpPlus.exe — nothing is installed, and settings, logs, the index and the model live in the data folder beside it. The build isn't signed: choose More info → Run anyway if Windows asks. The zip is ~470 MB, mostly the CUDA runtime the helper carries so the model runs on NVIDIA cards.

SHA-256 of DigUpPlus-0.1.6-windows-x64.zip: fb7f3e5fc70fd5b86fedf01b488d891ee1ddfefc2d6f7ed09dade99b05d00f99

Updating from 0.1.5: the helper changed in this release, so take the whole zip, not just DigUpPlus.exe. Close DigUp Plus first, unpack it over the old folder, and keep your data folder — the portable build keeps everything there.

What is in 0.1.6

A folder can say what it holds. Settings → Folders (and the window Add Folders… opens) has Rules on every searched folder: the kinds to index — pictures, PDFs, documents, recordings, videos — and one depth switch, Scan subfolders, with that folder's subfolders listed right there. The rule is saved before indexing starts, so the first pass already leaves out what you don't want instead of reading everything and re-syncing afterwards. The folder's line says what it takes (“pictures only · top level only”), and the “why” beside its state lists what was left out.

Recordings can be read by their tags. In the same tab, “Recordings read by their tags” takes file types (mp3, m4a…): those files are indexed by their title, artist and album in a moment, without decoding or embedding the sound, and are found by those words. Changing the rule re-reads only those files.

Two search settings. Settings → Search sets how far a meaning match must stand out (0.5 to 5, 2.0 by default) and how much of the image ranking the Images filter shows at once (Standard, More, Fewer).

Fixed along the way:

• The first estimate of a folder now honours the folders and file types you skip (it used to size everything).
• The window's “N not read” reasons survive the first pass — the helper's report was dropped whenever that sync also queued new files.
• A fresh install searched with the meaning gate at 0 rather than 2.0, before Settings had ever been saved.
• Removing a folder cleans the folders skipped inside it on Windows whatever their case.
• The Add Folders… window keeps the rules the folders already had; saving it used to clear them.

The rules plan, with the measurements and the checks behind these decisions, is in docs/per-folder-rules-plan.md.

DigUp Plus 0.1.5

Windows portable build, 64-bit Windows 10/11. Unpack the zip anywhere and run DigUpPlus.exe — nothing is installed, and settings, logs, the index and the model live in the data folder beside it. The build isn't signed: choose More info → Run anyway if Windows asks. The zip is ~470 MB, mostly the CUDA runtime the helper carries so the model runs on NVIDIA cards.

SHA-256 of DigUpPlus-0.1.5-windows-x64.zip: 8e74d320f4a341c5ca1a9c694cbd4d49687cdf4248d9c85bb3690a0cf3d9acc

What is in 0.1.5

The window behaves the way it reads. The X quits DigUp Plus outright — no forgotten half-running instances; indexing pauses and resumes on the next start. Esc (after everything else there is left to cancel) and Ctrl+W put the window away into the tray instead, with the app and its indexing going on.

The views tell the truth. The window opens on Images — the whole picture ranking in cosine layers, paged as deep as it goes. All files (the mixed-score view) is the last chip: its numbers come from a different ranking, and as the default it used to hide hundreds of cosine matches behind the gate. The Media chip went — its kinds have their own chips next to Images. When the open filter has nothing but other views do, the empty state points at them with counts.

Find similar (search by example). Right-click a picture, screenshot, video or audio result → Find similar: its own embedding becomes the query, the ranking is by cosine, and the reference itself stays out of its results. No re-embedding, no cloud. A failed lookup is a visible error, never a keyword search in disguise.

Small corpora search right. A meaning match on a tiny index could never pass the relevance gate (three files measured z 1.41 against the 2.0 threshold — a mathematical ceiling). Under 8 scale-setting vectors a clearly matching file is admitted by an absolute cosine floor, while unrelated queries stay rejected.

File operations at scale (a second review pass, verified by the full test suites): a 5,000-file move now travels in chunks of 500 index updates instead of 5,000 round trips; the destination folder is read once per batch instead of per file; a copy that fails before creating its target no longer reports a "partial copy" that doesn't exist; selecting results no longer repaints every cell on every key.

Datasets as a setting. Folders that look like datasets (thousands of similar files) are still skipped by default — Settings → Folders can index them after all, for collections whose subfolders trip the rule.

The first run opens Settings, not a wizard; nothing downloads without asking. No telemetry, no cloud, no update checks — the only network call is the one-time model download.

A Windows-focused fork of ARahim3/DigUp; see the README for what it changes.

DigUp Plus 0.1.4

Windows portable build, 64-bit Windows 10/11. Unpack the zip anywhere and run DigUpPlus.exe — nothing is installed, and settings, logs, the index and the model live in the data folder beside it. The build is not signed: choose More info → Run anyway if Windows asks. The zip is ~470 MB, mostly the CUDA runtime the helper carries so the model runs on NVIDIA cards.

SHA-256 of DigUpPlus-0.1.4-windows-x64.zip: 115f749417b101e86329288614d4efffcb8b3ebaa28e6fed4de40beeac45796c

What is in 0.1.4

File operations at scale (a second review pass, verified by the full test suites before adoption).

• A 5,000-file move no longer costs 5,000 helper round trips: index changes travel in chunks of 500, each chunk commits as one transaction, and a file that fails undoes only itself.
• The destination folder is read once per batch instead of once per file and once per (2), (3) candidate — renaming a big batch is linear now, not quadratic.
• A copy that fails before creating its target no longer reports a "partial copy" that does not exist.
• Selecting results no longer repaints every cell on every key: with ten thousand results, the checkboxes follow the keys.
• The path-resolution for moved files is cached per destination folder.

Index move and switch, visible and honest.

• Move Index Folder now shows the percents while it copies: the destination fills only at the end (the databases land under temporary names and are renamed when whole), and a big index used to look stuck.
• Use Existing Index… works during the speed-check window after a start — exactly when indexing begins over a folder that lost its index and a person needs to point at the real one. Move Index Folder explains in its tooltip that it waits for the check.

The first run opens Settings, not a wizard; nothing downloads without asking. No telemetry, no cloud, no update checks.

A Windows-focused fork of ARahim3/DigUp; see the README for what it changes.

DigUp Plus 0.1.3

Windows portable build, 64-bit Windows 10/11. Unpack the zip anywhere and run DigUpPlus.exe — nothing is installed, and settings, logs, the index and the model live in the data folder beside it. The build isn't signed: choose More info → Run anyway if Windows asks. The zip is ~470 MB, mostly the CUDA runtime the helper carries so the model runs on NVIDIA cards.

SHA-256 of DigUpPlus-0.1.3-windows-x64.zip: 9c2d9c4d7b5d591da313e3632fb2170ef0ea7c70c35681b8f598722bc8ae6e0f

What's in 0.1.3

Index and model, hardened. Use Existing Index… is one transaction: the index is checked to open before anything stops, a failed save keeps the old folder, and a move and a switch can no longer interleave. A file operation's index update waits for an index move instead of landing on the old one. A malformed settings.json is kept aside as settings.json.bad instead of being silently reset.

The engine. A file whose identity is unknown — a OneDrive placeholder, a file that couldn't be opened — no longer inherits another file's vectors through a move. Installing a model swaps the old file out through a .old copy, so an antivirus holding the file can't leave you without a model.

The results window, keyboard-first. Tab switches between the search field and the results; Del deletes the file under the cursor (the ticked ones, when some are ticked); the delete confirmation opens with the keyboard on the delete itself — Enter confirms, Escape keeps the files. The preview pane is resizable and remembers its width; after a move or a delete the view stays on the file it was on instead of restarting the query; the batch bar's actions look like buttons.

Honest delete copy. A file too large for the Recycle Bin may be deleted for good — the confirmation and the report say so, and Windows asks first.

The first run opens Settings, not a wizard; the model downloads only after you choose. No telemetry, no cloud, no update checks — the only network call is the one-time model download.

This fork is a Windows-focused take on ARahim3/DigUp; see the README for what it changes.

DigUp Plus 0.1.0

The first own release of this fork: a Windows portable build, no installer.

Download DigUpPlus-0.1.0-windows-x64.zip, unpack it anywhere and run DigUp Plus.exe. Nothing is installed and no
admin rights are needed. The exe isn't signed yet, so Windows may ask first: More info → Run anyway. It needs 64-bit
Windows 10 or 11.

Settings live in %APPDATA%\com.sucotasch.digupplus; the index and the model in %LOCALAPPDATA%\DigUpPlus. Keeping
them out of the unpacked folder means a new zip can replace the old one without re-indexing. The 865 MB model downloads
once, on the first launch.

What's in it

• DigUp Plus is its own product now (0.1.0, com.sucotasch.digupplus), so its index, model and settings never mix with
  an installed upstream DigUp.
• The upstream updater is gone: Settings shows the version and opens the releases page. Updates are manual.
• This fork's work on top of DigUp: paged search results, the Images filter in cosine layers, Move To / Copy To / Delete
  in the search window, and settings for where the index and the model live.
• The macOS app isn't built by this fork; the Windows and Linux shell is what ships here.