- Add date range (date_from/date_to), storage provider, and tags filters to GET /api/files
- Add SavedSearch model and migration (005_add_saved_searches)
- Add CRUD API endpoints for saved searches at /api/saved-searches
- Update files.html template with new filter controls and saved searches UI
- Update files view to pass new filter parameters to template
- Add comprehensive tests for all new functionality (26 tests)
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
The upload retry logic now checks file_record.processed_file_path first
(the GPT-suggested filename stored during finalization), before falling
back to legacy hash-based and original-filename-based path patterns.
This fixes the case where the processed file has a different name than
the original (e.g., '2023-10-01_Unknown.pdf' vs 'cable_graphic.pdf')
and the retry couldn't find the file on disk.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
The status calculation required ALL steps to be explicitly marked as
success/skipped before a file could be "completed". This failed for
dynamic pipelines where:
1. check_for_duplicates was logged before the file record existed (no
file_id), so its FileProcessingStep was never updated from "pending"
2. extract_text was not marked as "skipped" for non-PDF files that go
through PDF conversion first
Fix:
- Move check_for_duplicates success log to after initialize_file_steps()
with the correct file_id so the step actually gets updated
- Mark extract_text as "skipped" for non-PDF files
- Add terminal-step fallback: if send_to_all_destinations is "success",
the file is "completed" even if intermediate steps remain "pending"
(handles any other dynamic pipeline edge cases)
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Add a terminal-step guard (send_to_all_destinations) to all status
calculation paths so that files are only marked Completed once the
entire processing pipeline has been recorded.
- get_file_overall_status: require TERMINAL_STEP to be present
- get_files_processing_status: same guard for bulk status
- get_step_summary: count missing terminal step as queued so
total_main_steps > main_completed when pipeline is incomplete
- apply_status_filter: SQL sub-query requires terminal step for
completed filter
- process_document: call initialize_file_steps after creating a new
file record so all mandatory steps are pre-created as pending
Define TERMINAL_STEP constant in step_manager.py and reference it in
file_status.py and file_queries.py to avoid magic strings.
Tests updated: add send_to_all_destinations to completed-file
fixtures; add test verifying initialize_file_steps is called for
new files.
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Both get_file_preview and download_file were ignoring the DB-stored
original_file_path and processed_file_path fields, instead relying on
local_filename (a temp path that may be gone) and guessing patterns for
the processed file. This caused "Processed file not found" and
potentially "Original file not found" in the /files/{id} view even when
the files existed at their stored paths.
- version=original: check original_file_path first, fall back to local_filename
- version=processed: check processed_file_path first, fall back to
hash/filename guessing patterns
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
- Revert search.html and files.html links back to /files/{id}
- Add GET /files/{file_id} route (file_view_page) with workdir path-
containment guards (os.path.commonpath) to prevent traversal
- Create file_view.html: document-centric page showing AI metadata,
inline PDF preview, extracted OCR text, download actions, file info,
status pill, and link to /files/{id}/detail for process pipeline view
- Existing /files/{id}/detail route is unchanged
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
The debounceSearch() function in files.html called clearFullTextSearch()
when the query was shorter than 2 characters. Since clearFullTextSearch()
sets input.value = '', every single keystroke was immediately erased —
users could paste text but not type.
Fix: debounceSearch now only hides the results panel for short queries
without touching the input value.
Also adds a dedicated /search page with Google-style results showing
content previews (document title, filename, type badges, tag badges,
sender, and OCR text snippets with highlighted matches).
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
- Rewrite tests/test_celery_worker.py to mock check_credentials.apply_async
at import time, enabling all 35 statements to be covered without Redis
- Add tests for conditional beat schedule entries (IMAP, Uptime Kuma)
- Delete app/utils/config_validator.py — dead code shadowed by the
config_validator/ package directory (Python gives packages precedence)
- Remove both files from coverage omit in pyproject.toml
- celery_worker.py now at 100% coverage (was 0%)
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
- Raise quality acceptance threshold from 65→85 (configurable via TEXT_QUALITY_THRESHOLD)
- Reject text with significant issues (excessive_typos, garbage_characters,
incoherent_text, fragmented_sentences) even when score is above threshold
(configurable via TEXT_QUALITY_SIGNIFICANT_ISSUES)
- Add compare_text_quality() for AI-powered head-to-head comparison of
original embedded text vs fresh OCR output
- Update process_document to pass original text to OCR task for comparison
- Update process_with_ocr to run comparison and keep the higher-quality text
- Add new settings to settings_service.py metadata
- Update docs/ConfigurationGuide.md with new settings
- Add comprehensive tests for new threshold and comparison logic
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
- Update _compute_processing_flow to recognize process_with_ocr as the OCR
stage and remap legacy process_with_azure_document_intelligence log entries
for backward compatibility
- Normalize legacy OCR step name in _compute_step_summary log fallback
- Add process_with_ocr to REAL_MAIN_STEPS/REAL_STEPS in step_manager,
file_status, and file_queries (keeping legacy name for old DB entries)
- Update retry logic in api/files.py to retry failed OCR via process_with_ocr
(handles both step names as aliases)
- Fix process_document.py to log process_with_ocr as skipped (not azure step)
for the local text extraction path
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
Root cause: ensure_ocr_languages_from_settings() only downloaded tessdata
when the 'tesseract' provider was active, but embed_text_layer() uses
ocrmypdf (which needs tessdata) as a fallback for ALL OCR providers.
- embed_text_layer(): call ensure_tesseract_languages(language) after
confirming ocrmypdf is on PATH, so language data is present before
ocrmypdf is invoked (prevents exit code 3 for fra/deu/etc.)
- ensure_ocr_languages_from_settings(): extend the condition from
'tesseract' in active_providers to also trigger when ocrmypdf is
on PATH, enabling proactive pre-download at startup for any config
- Tests: mock shutil.which and ensure_tesseract_languages in affected
test cases; rename azure-only test and add new test for ocrmypdf case
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
- Add app/utils/ocr_language_manager.py: detects tessdata dir, downloads
missing .traineddata files via wget/curl from tessdata_fast GitHub repo,
pre-downloads EasyOCR models, exposes async background-thread helper
- TesseractOCRProvider.process() calls ensure_tesseract_languages() before
running pytesseract; raises clear error if languages remain unavailable
- EasyOCRProvider.process() logs informational message when models download
- app/main.py: calls ensure_ocr_languages_async() at startup
- app/utils/settings_sync.py: triggers language re-check after every
settings reload so UI changes take effect without container restart
- app/api/settings.py: adds POST /api/settings/install-ocr-languages
endpoint for on-demand language installation from the admin UI
- Dockerfile: adds wget for runtime tessdata downloads
- docs/ConfigurationGuide.md: documents automatic language download
- tests/test_ocr_language_manager.py: 29 unit tests
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
- PDFs are now uploaded to Mistral Files API (POST /v1/files) and
processed via a signed document_url, resolving the 422 error caused
by passing data:application/pdf;base64,... to an image endpoint
- Images (JPEG/PNG/GIF/WEBP/BMP/TIFF) use base64 image_url directly
- Unsupported MIME types raise a clear ValueError
- Magic-byte fallback detects PDFs with no file extension
- Switches from openai chat completions to requests HTTP calls
- Adds helper method _upload_pdf_and_get_document()
- Adds TestMistralOCRProvider with 9 unit tests covering all paths
Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>
copilot-swe-agent[bot]