Arion Release Notes

Arion 1.0.4 — May 12, 2026 · Index hygiene + clean-install hardening

1.0.4 is a focused bug-fix release. The big theme is keeping the search index honest about what's actually on disk — several cases where stale records lingered after files or directories were removed, or after an extension was toggled off, are now cleaned up on every scan.

Bug fixes

• Disabled file types now disappear from the index. Unchecking an extension in Settings → File Types and rescanning previously left the old records behind, so the document count and search results both kept showing files of that type. They're now pruned at the start of every scan.
• Deleted files and folders no longer "haunt" the index. The cleanup pass that removes records for files no longer on disk now runs before indexing on every scan, including scans that stop early because a Personal customer hit the 5,000-document cap. Previously the cleanup was skipped in exactly that case, so removing files on a full Personal index had no effect until you scanned again under the cap.
• WordPerfect files without a converter no longer pollute the index. Old versions tried to extract text directly from .wpd binary content as a fallback when LibreOffice wasn't installed — in practice that produced runs of formatting commands that looked enough like words to match garbage on real searches. Arion now reports these files as extraction errors with a clear hint about installing LibreOffice from libreoffice.org, or removing .wpd from indexed extensions in Settings → File Types.
• Personal-edition discovery messaging is clearer. When more than 5,000 files match the configured scan rules, the status bar now says “Personal indexes up to 5,000” alongside the discovered count, so it's no longer surprising to see a file count exceed the cap during the discovery phase.
• "Scanning…" stays visible while you search. Previously, running a search during a scan replaced the current-file text in the status bar — if the next file took a while to extract, there was no visible sign the scan was still running. A dedicated “⏳ Scanning…” indicator now persists in the status bar from scan start to finish, regardless of what other status text comes and goes.

Under the hood

• The PyInstaller bundle now includes the full Python standard library, which lets side-loaded packs (Video & Audio transcription, etc.) call any stdlib module from inside the bundle rather than just the imports Arion happened to use directly. Closes a class of "missing module" errors that only surfaced when an add-on pack reached for something Arion itself didn't use.

Arion 1.0.3 — May 2, 2026 · Side-loader + maintenance

1.0.3 closes the Family-Edition transcription promise from 1.0.1's release notes by introducing the on-demand side-loader, plus a handful of bug fixes surfaced during a structured QA pass.

New: on-demand transcription engine

The video and audio transcription engine (faster-whisper plus its native dependencies) is now an opt-in download from inside the app rather than bundled in the installer. Settings → Video & Audio shows a one-click Install Video & Audio Pack button when the engine isn't present; the download is ~65 MB and installs into a per-user shared location. The same engine serves Cadenzium and any future Chordalia products on the same machine, so you only download it once.

After install, restart Arion and tick Enable video & audio transcription — installing the engine and turning the feature on are deliberately separate steps. When a future Arion release ships an updated engine pack, the panel surfaces an Update Pack button automatically.

Bug fixes

• License install no longer trips on extra files. The Settings → Install License flow now validates exactly the file you pick, instead of scanning the whole directory and potentially picking up a stray expired license that happens to sort earlier alphabetically.
• Startup license validation is more forgiving. If multiple license files are present (e.g. an old expired one left behind alongside your current one), Arion now tries each and uses the first that validates, rather than refusing to start on the first failure. Tampered or corrupt files still hard-stop for security reasons.
• Personal-edition Video & Audio panel no longer shows a misleading green "ready" note when the engine is on disk from earlier Family use. The panel now shows a neutral note explaining the engine is present and will activate after a Family upgrade — no re-download needed at that point.

Under the hood

• The standard installer no longer ships onnxruntime (it moves into the side-loaded engine pack alongside faster-whisper and friends), keeping the base installer compact.
• Arion now declares a minimum required engine-pack version at startup — older packs left over from earlier installs are ignored even if their compatibility metadata still claims to cover this Arion version. Settings will prompt to install the current pack.

Arion 1.0.1 — May 1, 2026 · Maintenance release

A focused maintenance release driven by a structured QA pass against 1.0.0. Every change in this release is a real customer-visible bug fix or UX improvement — no new features, no breaking changes for existing customers.

Bug fixes

• Expired licenses now fall back to Personal mode instead of locking the application out entirely. Family or Archive customers whose subscription lapses can keep searching their existing index at Tier 1 while they renew. Genuinely corrupt license files (signature failure, malformed payload) still hard-stop, since those are integrity failures rather than lifecycle states.
• Removable drives no longer silently re-index after reconnect. When a configured scan root is unreachable (offline external drive, disconnected network share), background scans now skip the prune phase for records under that root and emit a clear "Scan root unreachable" status. Previously the daily scan would purge the entire index for that drive and re-extract everything on next reconnect.
• Excel 97-2003 (.xls) files now extract correctly. Previously listed as supported but silently failed because the underlying library only handles .xlsx. Now routed through LibreOffice headless, consistent with how .doc, .odt, and .ods are handled.
• Tools → Prune Excluded now honors "Excluded Folder Names" patterns (e.g. Cache, node_modules) in addition to full-path entries. Previously only the full-path list drove the prune.
• Switching the Ollama embedding model now works correctly for any model, not just the default nomic-embed-text. The native context window is auto-detected per model, and previously-bundled queries that would silently return a dimension-mismatch error now show an actionable "Run Tools → Build Semantic Index" message.
• Archive members (files inside .zip, .cbz, .tar) now index correctly. A regression in 1.0.0 caused a silent NameError during archive-member indexing; archive contents will be picked up on the next scan after upgrade.
• Disk-low and Personal-cap-reached aborts now keep their warning visible at the end of the scan instead of being clobbered by "Last scan ended: just now".
• Tier 3 ChromaDB index now writes to %LOCALAPPDATA%\Chordalia\Arion\<config>\ arion_vectors\ on every install (a path-resolution bug previously left it pointing at Program Files\Arion\ arion_vectors\, which is read-only on standard Windows installs).

UX improvements

• License-error dialogs no longer duplicate the support contact line.
• Settings → Search Tier — selecting Tier 2 or Tier 3 with no configuration shows a "Configure (Tier N)" shortcut button that switches to the relevant config tab, instead of a dead-end "Test Connection" button that fails when there's nothing to test.
• Settings → Ollama dropdowns now filter to model types appropriate to each slot — chat models for the Chat dropdown, embedding models for the Embedding dropdown.
• Launching a second Arion on a machine with a single existing config (Family+) opens the config picker with a "Create New…" shortcut instead of a dead-end warning.
• Large archive scans now show per-member progress in the status bar instead of appearing frozen on the archive's filename for minutes at a time.
• System Status diagnostics distinguishes missing packages, bundled-but-broken native dependencies, and successful loads — with troubleshooting hints appropriate to each state.
• Cleaner log output — directory-walk lines are no longer at INFO level (they were drowning out search activity, license events, and scheduler decisions). Embedded model name now appears in the per-Ask context log.

Installation prerequisites

Tier 3 (Local AI) requires the Microsoft Visual C++ 2015-2022 Redistributable (x64). Most modern Windows machines already have it installed; if Settings → System Status reports a bundled-but-failed-to-load library, install vc_redist.x64.exe from Microsoft and restart Arion. CPUs from before roughly 2013 may also lack the AVX2 instructions required by the Tier 3 components.

Notable change for Family Edition

Video and audio transcription is temporarily unavailable. Arion 1.0.1 ships without the speech-to-text engine. The standard build no longer bundles faster-whisper and its native dependencies. An on-demand installer for the transcription engine — shared with Cadenzium and downloaded once per machine — is planned for 1.0.2. Until then, video (.mp4, .mkv, etc.) and audio (.mp3, .wav, etc.) files are skipped during scans on Family Edition. Settings → Video & Audio shows an explanatory note when the engine isn't installed.

Arion 1.0.0 — April 29, 2026 · Initial release

Arion 1.0.0 is the first public release. It indexes documents on your machine (and optionally your cloud storage) and gives you fast full-text search with optional AI question-answering — all under a clear single-tenant model where Chordalia Systems never sees your data. Indexing and queries either stay on your machine or, if you opt into Tier 2 cloud AI, talk to your own Azure account.

Editions

• Personal — free. Up to 5,000 indexed documents, single scan root, Tier 1 keyword search. No license file or account required.
• Family — one-household paid edition. Unlimited documents, multiple scan roots, all search tiers, plus video and audio transcription, optional local web interface, cloud source connectors, and per-user document visibility. See current pricing →
• Archive — one license per Arion installation, with the license tier capping how many users that installation supports. Tiers: Small (up to 5), Office (up to 25), Site (up to 100), Enterprise (above 100, custom quote). Adds image description, MariaDB-backed shared index for multi-machine installs, role-based access control, and audit log. See current pricing →

Search

• Tier 1 (Local Search) — SQLite FTS5 full-text keyword search. No network calls, no AI, runs on any modern machine. Available in all editions.
• Tier 2 (Cloud AI Search) — Azure AI Search and Azure OpenAI in your own Azure subscription. Semantic search and AI question-answering. Family and Archive.
• Tier 3 (Private AI) — Ollama and ChromaDB on hardware you control. Same semantic search and AI Q&A as Tier 2 with no data leaving your machine. Family and Archive.
• Hybrid retrieval for Ask mode — combines semantic and keyword matching so questions about specific terms reliably surface the right documents, even buried in long ebooks or reports.

File formats

• PDF (text via pdfplumber)
• Word: .docx (python-docx) and .doc (legacy, via LibreOffice headless conversion)
• WordPerfect: .wpd, .wp, .wp5, .wp6 (LibreOffice)
• OpenDocument: .odt (LibreOffice)
• Rich Text: .rtf (striprtf, pure Python)
• EPUB and HTML
• Spreadsheets: .xlsx, .xls, .csv, .tsv, .ods
• Plain text and Markdown
• Family+: Video (MP4, MKV, MOV, WMV, WebM, and more), audio (MP3, WAV, M4A, FLAC, OGG, and more), and images (JPG, PNG, WebP, HEIC — Archive only)
• Archive contents: search inside .zip, .cbz, and .tar archives without extracting them

Family+ features

• AI question-answering with source citations
• Semantic search beyond exact keywords
• Video and audio transcription via local faster-whisper
• Cloud source connectors: Dropbox and OneDrive (read-only, OAuth)
• Multiple configurations side-by-side — one index per collection, each with independent autostart
• Custom AI system prompt
• Per-user document visibility (private folders per user)
• Optional local web interface for browser-based access on your local network

Archive features

• Image description via Ollama vision models — search your photos by content
• MariaDB-backed shared index for multi-machine installs
• Role-based access control
• Audit log of searches, document access, and admin actions
• One license per Arion installation, with the license tier capping the user count (Small, Office, Site, or Enterprise). Each tier is available as annual subscription or perpetual.

Platform and install

• Windows 10 / 11, 64-bit. Inno Setup installer (~62.5 MB).
• System tray app with optional launch-at-login
• Daily background scanning (configurable hour) plus on-demand Scan Now
• Diagnostics tab checks ExifTool, LibreOffice, Whisper, Azure, ChromaDB, and other dependencies
• Privacy: no telemetry, no crash reporting, no calls home. See the Arion Privacy Policy for details.

Planned for upcoming versions