v0.10.1 — Stability, UX Polish & Model Updates

Summary

This is a maintenance-focused release that improves the reliability of Document Insights execution, smooths frontend behavior for long-running jobs, and adds support for newer Gemini models.

Document Insights & Tasks

• More reliable generation lifecycle (clean answer resets, correct resume after refresh) and reduced parallelism for improved stability.
• Fixed loaders, pagination, and duplicate execution issues across Insights grids and dialogs.
• Added UTF-8 BOM to CSV exports for correct opening in Excel and similar tools.

Authentication (Keycloak)

• Improved login and registration UX to better support SSO with external identity providers.

Models & AI

• Added Gemini 3 Flash (preview) and fixed reasoning/token accounting for Gemini 3 models.

Performance & Reliability

• Fixed job polling edge cases, reduced unnecessary reloads, and applied general stability optimizations across Insights and Digitization.
• Added a lightweight queue to trigger document splitting and project training automatically after document uploads.

v0.10.0 — Document Insights Expansion, Better Exports, Bigger File Support

Highlights

This release significantly upgrades Document Insights (supporting documents, stronger UX, and safer regeneration), adds CSV export end-to-end, improves search and ingestion for more file types, and hardens batch processing/performance for large workloads.

Document Insights

• Supporting Documents for Insights: attach supporting docs at both question and document level and include them in the execution context for better, grounded answers.
• Safer answer lifecycle: when Insight questions or document values change, related answers are invalidated/cleared to prevent stale results and misleading summaries.
• Global task prompt & task-level controls: tasks can store prompt/model and key completion parameters (e.g., temperature/top-p/max tokens), enabling consistent runs and easier tuning.
• Usability improvements: refreshed dialogs (document/question), pagination + loading fixes, unified “reloadAll()” refresh behavior, and clearer warnings when actions will delete answers.
• Duplicate Insight Task: quickly clone an Insights task from the UI, including validation and backend support.

Export & Reporting

• CSV export for Document Insights and Document Digitization: full flow across backend + frontend, supporting export for single document or all documents.

Search & File Support

• Wildcard document search by name (including accent-insensitive matching via DB support), enabling “contains” queries for quicker discovery.

• Broader ingestion support:

  • Added .doc and .docx support.
  • Added Excel support by transforming spreadsheets into large CSV representations for storage and processing.

Performance & Reliability

• Digitization stability for large files: reduced OOM risk via more controlled batching/parallelism, safer temp-file handling (especially for remote stores like S3), and improved cleanup behavior.
• Job status polling fixes: eliminated runaway/infinite frontend polling and improved interval behavior.
• Database performance improvements: optimized document deletion (indexes + native SQL paths) and reduced unnecessary entity fetching for delete operations.

AI / Model & Token Accounting

• More accurate token accounting across providers (including Gemini reasoning token handling aligned to OpenAI-style billing semantics).
• Caching + audit visibility: improved LLM caching behavior and logging of cached/reasoning tokens in audit logs.
• Backend tool execution upgrades: introduced a registry/command-style execution pattern and expanded tool support (e.g., read-document by id) to support richer agent workflows.

Platform / Ops

• Upload limit increased to 150 MB.
• Dependency/platform upgrades (notably Spring Boot upgrade and runtime/tooling alignment for migrations).
• Keycloak/keystore and compose hardening: improved configuration so empty/invalid keystore files don’t break deployments.

If you want, I can format this into the exact GitHub Release layout you use (with a “What’s Changed” PR list section), but the above is already trimmed to the core shipped capabilities.

v0.9.0 - Document Insights & Document Digitization

Document Insights Task (1 answer per document)

Create a board of documents, define predefined questions, and run a batch job that consolidates each document’s responses into a single, trustworthy answer. Includes Generate New/All/Selected actions, checkbox selection, CSV export, and live job status/polling.

Document Digitization Task (1 answer per page)

Set a prompt and export structure, choose all pages or a precise 1-based inclusive page range, and run LLM OCR per page. Parallel execution with a global generation status UI, per-document CSV export, strong file-type validation, and optimized image processing.

Features

• Tasks system (Tasks/TaskNodes/TaskEdges) powered by Spring Batch for reliable execution.
• Document Insights: consolidated per-document answers; skip a doc+question group if any completion fails.
• Insights UI: Generate New/All/Selected, checkbox selection, CSV export, improved answer dialog.
• Digitization: page range selection (from/to, 1-based inclusive) and “All Pages” flag.
• Digitization execution: parallel generation with persistent, draggable Global Generation Status.
• Digitization export: per-document CSV (Title | Prompt | Structure | Page 1 | Page 2 | …).
• OCR pipeline tuned for vision LLMs (page printing, contrast/sharpness enhancement) with large memory/speed gains.
• File-type validation with clear UX; supports PDF, DOC, DOCX, PPT, PPTX, XLS, XLSX, ODT, ODP, ODS, RTF.
• Increased max upload size to 100 MB.
• Job APIs for status and batch tracking; multi-ID polling; renamed controllers/services (JobController/JobService).
• Added .docx ingestion in related pipelines.

Major bug fixes

• Fixed “Generate All” failing after page navigation; ensured UI refreshes when jobs complete.
• Made page-number filtering inclusive and reduced unnecessary page renders during OCR.
• Resolved IndexOutOfBoundsException in Digitization processor with page-bounds validation.
• Fixed document deletion bugs and ensured related task nodes/edges are cleaned up.
• Added missing @PreAuthorize checks and tightened Super Admin-only operations.

What's Changed

• Feature 433 trigger splitter training with project by @gyfantidis in #434
• Feature 441 document digitization task by @gyfantidis in #446
• Releare v0.9.0 - Document Insights & Document Digitization by @sekasx in #449

Full Changelog: v0.8.5...v0.9.0