Skip to content

Reference: workflows and tools

🤖 AI Drafted (Not reviewed)

This part of the manual is generated from the app itself, so it says what Fichero actually does rather than what someone remembered it doing. It covers 52 shipped workflows and 126 tools.

Workflows are grouped into folders that follow the route work takes: prepare the image, find the regions, read them, clean the reading, translate it, describe and extract, then catalogue, organise, export.

Image Editing

  • Enhance Images — Create contrast, sharpness, and light denoise image derivatives without modifying the originals. Useful for low-contrast scanned documents before OCR or transcription.
  • Fuzzy Clean Images — Despeckle and clean uneven scan background into derived image files without modifying the originals. Useful for noisy scanned documents before OCR or transcription.
  • Recombine Segments — Stitch segment image files into one derived image. Pairs naturally with Segment Images: run segmentation, then recombine selected segment outputs vertically or horizontally.
  • Remove Background Images — Create alpha-background image derivatives without modifying the originals. Uses a threshold method by default and can be switched to OpenCV/rembg where available.
  • Rotate / Auto-Orient Images — Create rotated or EXIF-oriented image derivatives without modifying the originals. Use this as the camera/DSLR watched-folder capture preset for auto-orienting imports and writing JPGs for downstream image editing or OCR workflows.
  • Segment Images — Detect foreground document regions and write cropped segment derivatives without modifying originals. Useful for separating multi-region scans before editing, OCR, or recombination.
  • Split Images — Split images into grid tiles or PDFs into per-page image derivatives without modifying originals. Useful before OCR, segmentation, or recombination workflows.

Detect Regions

  • Backfill Text Geometry — Use this when: pages already have a transcription but no word boxes. Apple Vision detects regions on the page (free, on-device), then the existing text is aligned onto those boxes, so every word in the transcript gets a position — measured where Vision read the same word, interpolated between anchors where it did not. Boxes are saved as provider aligned:apple_vision, never as a measured OCR pass, and a page whose alignment cannot be trusted is refused rather than given a wrong overlay.
  • Detect Regions (Apple Vision) — Use this when: you want text bounding boxes on pages before (or without) transcription. On-device Apple Vision detects line and word regions and saves them per page — fast, local, free. For hard hands Apple can miss regions; try Detect Regions (VLM) with a vision model of your choice.
  • Detect Regions (VLM) — Use this when: Apple Vision misses regions on hard material (archaic hands, damaged pages) and you want a vision LLM to find the text regions instead. Uses whatever model you pick at run time — compare providers by running this next to the Apple Vision preset.

Transcribe

  • Capture OCR + Transcribe — Use this when: you want a capture-oriented OCR preset for newly photographed pages or uploaded PDFs. It prepares pages into OCR-ready images, enhances those derived images, then transcribes back onto the original selected documents so page text, artifacts, and provenance stay attached to the source record instead of the temporary derivatives.
  • English Secretary Hand (16th–17th C.) — Use this when: you have English secretary hand — depositions, indentures, wills, parish and court records of the 16th–17th centuries. One careful whole-page pass; run Paleographer Review afterwards to refine.
  • Latin Paleography — Use this when: you have Latin manuscript text — medieval through early-modern (charters, ecclesiastical records, notarial instruments). One careful whole-page pass with abbreviation-system advice; run Paleographer Review afterwards to refine.
  • Paleografía Española (s. XVIII–XIX) — Use this when: you have 18th–19th century Spanish administrative, notarial, or secretarial hands (humanística, bastarda, letra de oficio). Replaces the Spanish Script v2 presets. One careful whole-page pass; run Paleographer Review afterwards to refine.
  • Paleografía Española (s. XVI–XVII) — Use this when: you have early-modern Spanish script — procesal, cortesana, itálica — from Spain or colonial America (notarial records, audiencia files, cartas de poder, confesiones, probanzas). One careful whole-page pass with deep period advice; run Paleographer Review afterwards to refine.
  • Paleographer Review — Use this when: a document ALREADY has a transcription (from any Transcribe preset, any model) and you want a paleographer’s review of it against the image. Three staged passes with extended thinking — abbreviations & formulary, orthography & consistency, final layer — all on the ONE model you pick at run time (transcribe with one model, review with another if you like). Writes the reviewed text as the document’s content. Replaces the Ensemble preset.
  • Prepare Images for OCR — Normalize image and PDF pages into OCR-ready derived image files without modifying the originals. Applies EXIF rotation, renders PDFs, boosts contrast, and writes prepared JPGs for downstream transcription workflows.
  • Transcribe — Use this when: you want to transcribe any file using the vision model in your provider settings, without a script-specific prompt. For better accuracy on specific document types use Transcribe HTR (historical handwriting), Transcribe Manuscript (modern handwriting), Transcribe Typescript (printed/typed), or Transcribe Paleography (archaic scripts).
  • Transcribe (Auto-Detect) — Use this when: you don’t know the document’s script type and want automatic routing. Classifies the script type (Typescript / Manuscript / HTR / Paleography) then runs the matching transcription profile — including two-pass review for historical and archaic scripts.
  • Transcribe + Review (Pipeline) — Use this when: you want the full paleography treatment in one run — the single-pass Transcribe Paleography workflow, then the Paleographer Review workflow over its result. This preset RUNS the two other workflows as sub-workflows (the chain pattern: any preset can compose others the same way). Model: whatever you pick at run time, used by both stages.
  • Transcribe HTR — Use this when: you have legible historical handwriting (16th–19th C.) that doesn’t need the full paleography treatment. One whole-page pass with period-orthography advice; run Paleographer Review afterwards for hard documents.
  • Transcribe Manuscript — Use this when: you have modern handwriting (20th–21st century) — letters, notes, diaries, forms, signatures. For historical handwriting (pre-20th C.) use Transcribe HTR; for archaic scripts use Transcribe Paleography.
  • Transcribe Paleography — Use this when: you have archaic or specialist pre-18th C. script in any language and want ONE careful, whole-page pass with extended thinking. Language-specific presets (Español s. XVI–XVII / s. XVIII–XIX, Latin, English secretary) carry deeper period advice; run Paleographer Review afterwards to refine the result. Model: whatever you pick at run time.
  • Transcribe Paleography (Economy) — Use this when you want a near-free paleography draft: Apple Vision line geometry (free, on-device) feeds a cheap local HTR backend, then ONE small-model cleanup pass corrects it against the image. Total paid calls per page: at most 1, vs the ensemble’s 15+.
  • Transcribe Typescript — Use this when: you have typewritten or printed documents — typescripts, books, pamphlets, official records with clear print. Uses Apple Vision (on-device, fast) when configured; falls back to the LLM vision model for scanned pages with poor contrast.

Clean Up

  • Clean Up Text — Transcribe each file (reusing any existing text layer), then clean the extracted text with an LLM: fix OCR noise, normalize whitespace and hyphenation, and strip page headers/footers and scanning artefacts — preserving the original meaning. The cleanup behaviour (aspect toggles, prompt, and model) is fully editable on the Clean Up Text node. The LLM node uses the $small model alias — set your default small model in Settings → AI Defaults.

Translate

  • Translate — Transcribe each file (reusing any existing text layer), then translate the extracted text into your preferred language using an LLM. Target language, source language, model, and prompt are all editable on the Translate node. The LLM node uses the $small model alias — set your default small model in Settings → AI Defaults.
  • Translate (DeepL) — Transcribe each file, then translate the extracted text through DeepL (default Dutch to English). Edit source/target language on the Translate node as needed.
  • Translate + Double-Check — Two-pass translation with an AI double-check. Transcribe each file, translate the text into your preferred language, then a second AI pass reviews the draft translation against the original source and corrects mistranslations, omissions, and altered names. Target language, model, and prompts are editable on each node. The LLM nodes use the $small model alias — set your default small model in Settings → AI Defaults.
  • Translate the Reviewed Transcription — Use this when: a page has already been transcribed and reviewed, and you want the TRANSLATION of that reviewed text — not a fresh reading of the image. Every other Translate preset re-transcribes first, which pays for the hard work twice and translates a reading you did not check. This one reads the newest transcription_review artifact and translates that.

Describe

  • Describe (visual) — Visually describe each selected image or PDF page: what is depicted, layout, figures, handwriting versus print, physical condition, and notable visual details. Saves one description artifact per page child.

Extract

  • Accounts → Spreadsheet (CSV) — Use this when: a page (or each page of a document) holds historical accounts — cuentas, tithe books, tribute rolls, expense lists — and you want spreadsheet rows out. Reads amounts as a paleographer (roman numerals, the U thousands mark, maravedís/pesos/reales), keeps the original text, and normalizes amounts to numbers where certain. One CSV artifact per page; open or export from the artifact.
  • Diary Entries — Transcribe each diary page (reusing any existing text layer), then split every page into per-day entry nodes: one child per dated entry, carrying a date attribute (diary_entry document type, date role) and the day’s bounding box on the page. The entries render immediately in the library’s Data view — timeline and calendar. Pages with several dates become several entries; a page that IS one day’s entry becomes one. Unreadable dates are kept as undated entries named by the date text on the page, never guessed. The splitter uses the $small model alias — set your default small model in Settings → AI Defaults.
  • Extract Geo — Transcribe each page (reusing any existing text layer), then extract every place name and geocode it to lat/lon. Feeds the 2D world-map and 3D globe representations. Offline geocoding by default; enable online geocoding on the Extract Geo node to resolve places the built-in gazetteer misses.
  • Extract Table — Detect and extract tabular data from each image/page using vision, emitting CSV you can paste into a spreadsheet. Feeds the spreadsheet (Table) representation in the document viewer. The output style and header handling are editable on the Table node.
  • Modernización (Spanish) — Use this when: you want the document’s text in modern Spanish orthography — same words, modern spelling, expanded abbreviations — beside the diplomatic transcription. Writes an analysis artifact; the original transcription is untouched.
  • Regesto (Archival Abstract) — Use this when: you want the archivist’s calendar entry for a historical document — date, place, document type, parties, action, and index terms — from the image plus any existing transcription. Writes an analysis artifact; never touches the document’s text.
  • Translate to English (Historical) — Use this when: you want an English translation of a historical document (Spanish, Latin, or other), faithful to its legal and notarial register, with names preserved and uncertainties carried through. Writes an analysis artifact.

Catalogue

  • 1 · Import → Artifacts — Step 1 of the discrete import-artifacts pipeline. Reuse already-imported page content to register only the lightweight per-page artifacts needed for review: import receipts and transcription artifacts. Safe to re-run: existing artifacts are skipped.
  • 2 · Extract Entities — Step 2 of the discrete import-artifacts pipeline. Reuse the transcription artifacts from step 1 to extract and persist entity rows only. Safe to re-run: entity upserts reuse existing rows and accumulate per-page provenance without duplicates.
  • 3 · Extract SVO → Claims — Step 3 of the discrete import-artifacts pipeline. Reuse the transcription artifacts from step 1 plus the persisted entity rows from step 2 to extract and persist SVO KnowledgeClaim rows only. Safe to re-run: the canonical claim writer deduplicates repeated page-scoped claims.
  • 4 · Merge / Dedup — Step 4 of the discrete import-artifacts pipeline. Reapply the persisted entity-resolution rules, claim-suppression rules, and conservative trivial-claim pruning to the existing KG rows for the selected document subtree. Safe to re-run: already-merged or already-suppressed rows are left unchanged.
  • 5 · KG Persist / Finalize — Step 5 of the discrete import-artifacts pipeline. Recompute cross-source corroboration, backfill canonical KG embeddings when missing, and refresh the kg.nt graph snapshot from the existing KG rows. Safe to re-run: corroboration writes only when values change, embeddings only backfill when counts are stale, and the graph refresh only runs when the snapshot is missing or claim support changed.
  • 6 · Catalogue — Step 6 of the discrete import-artifacts pipeline, and the one the full ‘Catalogue’ preset performs as its final act. Writes the folder-level archival description from the entities, claims and curation stages 1-5 already produced — it does NOT re-transcribe or re-extract, so corrections made after those stages are preserved. Run this when the work below is already done and only the description needs (re)generating. Fails loudly and names the missing stage when no claims exist for the target yet. Output attaches to the container — a selected folder stays on that folder; a single file stays on itself.
  • Catalogue — Run the whole Catalogue pipeline: the six numbered stage presets in this folder, in order, as one chain. 1 · Import → Artifacts registers per-page import receipts and transcription artifacts from the already-imported page content; 2 · Extract Entities and 3 · Extract SVO → Claims read those artifacts and persist entity and claim rows ($small model); 4 · Merge / Dedup reapplies curation rules; 5 · KG Persist / Finalize recomputes corroboration, embeddings and the graph snapshot; 6 · Catalogue writes the narrative archival description ($small model). Every stage is also runnable standalone from this folder, and each is safe to re-run. Pages must already carry text — run a Transcribe workflow first for scans without content.
  • NER per-page (local) — Local NER extraction over a folder of text files (.md / .txt). No transcription step (text already has page_content); no folder cleanup; no aggregate-combine output. Each file gets its own per-doc people / places / organizations / dates / events / keywords artifacts, and factual KG claims persist inline during extraction. Uses the $small alias — point it at Apple Intelligence in Settings → AI Defaults to keep everything on-device. Designed for field notes, journals, and any plain-text corpus where you want entity extraction without burning cloud tokens.

Organize

  • Group Same Documents — Score the selected files together, group duplicate or variant forms of the same underlying document, then organize duplicate clusters into subfolders through the audited action registry.

Books

  • Split Chapters — Split a book PDF into chapter child documents. Uses the embedded PDF outline when present, then page-heading heuristics, and falls back to one whole-book chapter.

Convert

  • AI Convert to HTML — An AI vision model reads each image/page and WRITES it out as semantic HTML — a generative re-creation of the structure and, optionally, inline styles, not a mechanical file conversion; the markup is the model’s reading of the page. Feeds the HTML/WebKit representation in the document viewer. The model alias and style options are editable on the Convert node.
  • AI Convert to Markdown — An AI vision model reads each image/page and WRITES it out as clean Markdown — a generative re-creation of the content (headings, lists, tables), not a mechanical file conversion; wording and layout are the model’s reading of the page. Feeds the Markdown representation in the document viewer. The model alias and layout options are editable on the Convert node.
  • AI Redraw as SVG — An AI vision model REDRAWS each image/page as SVG — a generative sketch of the document’s layout and text, not a vector trace of the original image; positions and shapes are approximate. Feeds the SVG (WebKit) representation in the document viewer. The model alias is editable on the Convert node.

Export

  • Export to Desktop (MD + DOCX + XLSX) — Export the selected folder or whole library to the Desktop as a Markdown folder, a Word .docx, and an Excel .xlsx. No LLM required — runs entirely on the local export service.