Changelog — FileMind

v0.4.2-beta September 9, 2026

v0.4.2-beta — September 9, 2026

Fixed

Send Feedback in the sidebar now opens the feedback form inside the app instead of a browser, and reports carry the app version like the header button already did
Ask now works with Claude models that think before answering (Sonnet 5, Opus 5). It had been reading past the answer and retrying three times before giving up with a message that pointed at unrelated settings

v0.4.1-beta September 7, 2026

Added

In-app feedback now carries the app version and platform, so a report can be matched to the build it came from

Changed

The setup wizard's AI step no longer calls itself optional — it has been required since the skip button went, because Ask needs a model

Fixed

The built-in model (and LM Studio or any custom local server) was recorded as if it were OpenAI: Settings showed a 'no known price' warning for it, and a reached monthly spend limit would have blocked it. Local models are free and are never blocked

v0.4.0-beta September 6, 2026

Added

Built-in AI model: enable local AI in one click during setup — no separate install, no account, no API key
LM Studio support, and automatic detection of any local LLM server already running on your machine
Curated Ollama model catalog with hardware-aware recommendations, and model downloads from inside the app
Guided API-key setup for OpenAI, Claude and Gemini, with a Test Connection button that checks the key before saving it
AI spend tracking for cloud providers: exact token log, estimated cost, and a monthly limit you set
Bundled search and OCR models — the first scan now works with no internet connection at all
Reranking pass on search results, so the most relevant passages surface first
Automatic updates: new beta builds install themselves in the background
API keys and other secrets are encrypted at rest using Windows DPAPI
Error screen now has a button that opens the logs and a path to report the problem

Changed

Much faster startup — roughly 19 seconds to ready on a clean machine, down from 56
The trial PDF limit is lifted for beta builds, so you can index your whole library
Ask now sizes how much of your library it reads to fit the model's context window automatically
Thinking-mode control for Ollama models that support it

Fixed

Setup could fail to create a library on Windows, which blocked first run entirely
Application data was split across two folders; everything now lives in one place under %APPDATA%\FileMind
The [S1] citation markers inside an answer look clickable and now actually open the cited PDF at that page
The PDF viewer failed to open a file when its folder had been entered by typing the path rather than picking it
Ask and search could leave a background process running after quitting the app
OCR did not run on packaged builds, so scanned PDFs stayed unsearchable
Secure connections failed on some machines whose certificate store was out of date
Ollama model names could match the wrong model when several shared a family name
The Gemini model list was out of date and offered models that no longer exist

v0.3.0-beta August 7, 2026

Added

PDF scanning: recursive folder scan with automatic DOI and arXiv ID extraction
Intelligent rename proposals grounded in Crossref and arXiv metadata
Confidence scoring (High / Medium / Low) for every rename proposal
Batch rename with one-click approval for high-confidence proposals
Full-text search across entire PDF library
Semantic search using local embeddings
Hybrid search combining keyword and semantic results
Ask My Library: natural-language Q&A with page-level citations
BibTeX, RIS, CSV export, and direct Zotero integration
Undo journal: reverse any rename or batch operation in one click
Trial tier: full feature set, up to 100 PDFs, no time limit
Pro license: unlimited PDFs, 2-machine activation
Lab license: unlimited PDFs, up to 5 seats
Windows (x64) support — macOS build ships later; license already covers both platforms
Dark mode following OS preference