Kalamu 1.3.0

Kalamu 1.3.0 speaks French, searches the text of a whole project, lists what still needs
checking across it, and keeps each source's archival description, exported in your own columns
beside what was read from it. For what 1.2.0 changed, see the
1.2.0 release.

What's new since 1.2.0

• A French interface: choose Same as the system, English or Français in Settings ›
  Appearance and language; by default Kalamu takes the first of the system's languages it
  has. Dates, numbers and costs follow the language, and items' pages are typed as 8 (gauche).
  The default prompts' names are translated; the prompts themselves, and the files Kalamu
  writes, stay in English, and the answers' language is still chosen in New job.
• Search a whole project: every passage of its documents that holds some words, in the text
  as you corrected it, capitals, accents and Arabic vowel marks set aside. Look for the exact
  phrase, all the words in the same page, or words near each other; narrow by collection, tag,
  task or review; include analyses' answers and tables of contents if you wish. Each match opens
  its review at that place, a recording at its line. The search runs on the computer: nothing
  is sent, and its index is made again from your results when needed.
• What to check: one list, for a project, a collection or a selection, of the pages and
  answers that need you, each with its reason: failed or not read, cut off at the model's limit,
  read differently by two set-ups, read from a text corrected since, or not yet reviewed. Each
  opens its review, from which the previous and next items are a click away. Only your review
  resolves an item; models agreeing never counts as one.
• Describe each source as an archive does: repository, fonds, series, box or folder,
  shelfmark, creator, date as written, web address, citation and rights. Give the parts a
  project's documents share once, in the project; a document fills in its own. Export
  descriptions writes them as CSV and JSON, each value beside where it came from, and each
  result's provenance.json names the documents its source is, with their descriptions.
• Export profiles: the same export in your own columns, order and headings, for a
  spreadsheet or an Omeka S import: start from every column or from Dublin Core terms, add fixed
  texts, choose the list separator and what an empty cell holds. A column can also read each
  document's result: how far it was reviewed, the model, its text as corrected, or an analysis's
  answer, either as the model gave it or under your dictionary's labels. Profiles are kept for
  every project, shared as templates and imported.
• Rename a tag across a project from the table's tag filter; onto a tag in use, the two
  become one.
• Languages in New job: it now says when a transcription model takes only one of the
  languages chosen, or a language only without speakers and timestamps.
• Report a problem leads with bugs and ideas, and opens GitHub's choice of forms; offensive
  or harmful output can still be flagged there. A misreading is for the review to correct, not
  to report.

Fixed

• Gemini 3.5 Transcribe names speakers Speaker 1, Speaker 2 (the live service labels them
  spk:0, spk:1), and its cost is counted from its tokens: it showed $0.
• OpenAI caches only the instructions: implicit caching wrote each whole page to the cache at
  1.25 times the input price.
• Descriptions, profiles and the dictionary's CSV files open in Excel with their accents.
• Tokens billed for an answer a provider's filter stopped count in the job's cost.
• Splitting and joining items is saved in one step, with clearer messages; cutting out or
  exporting items no longer freezes the window.
• Project search waits for a word of three letters before searching, and a database error is
  said as such rather than counted as text not read.
• After renaming a tag, the table stays filtered on it.
• A result's provenance.json is no longer overwritten with older details when a document is
  renamed while its job finishes.

Download

• Microsoft Store: the Store copy installs without a warning and updates itself:
  Kalamu in the Microsoft Store.
• Kalamu_1.3.0_x64-setup.exe: Windows 11 on Intel and AMD PCs.
• Kalamu_1.3.0_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops).
• Kalamu_1.3.0_aarch64.dmg: macOS 13 or later on a Mac with Apple Silicon (M1 and later).
  Open it and drag Kalamu to Applications.
• kalamu-samples.zip: a short recording, scanned pages, a newspaper issue and a text to try
  the app with.
• Kalamu_1.3.0.msixbundle and the two .msix packages are the Store submission and its
  per-architecture packages, for testing; install Kalamu from the Store or the installers.

Every download bundles FFmpeg and offline Whisper; nothing else needs to be installed.
Installing 1.3.0 over an earlier version keeps your jobs, settings, presets and keys; project
search builds its index the first time Search is opened.

Known limitations

• Still in English: the macOS application menu, model descriptions, the error a failed page
  keeps, and the files Kalamu writes. German will follow.
• Search finds words of three letters or more, and one word or another is not offered yet.
• Items are whole pages of a PDF. Two articles on one page each take the whole page; a
  multi-page image's items can be made and corrected, but not read or exported on their own.
• Unsigned installers: Windows SmartScreen will warn. Choose More info, then Run anyway,
  or install from the Store.
• The Mac app is not signed with an Apple Developer ID, so the first time macOS refuses to
  open it: open System Settings → Privacy & Security, choose Open Anyway beside Kalamu,
  and confirm. macOS may also ask once, and again after an update, to let Kalamu use the key
  saved in your Keychain.
• The Mac version is new: it is built and tested on Macs in GitHub Actions, but has not yet
  been used on a real Mac. Please report what does not work with
  Report a problem. Intel Macs are not
  supported.
• Offline Whisper, tables of contents and corrections have had fewer runs against the live
  services; OpenAI's batches had not answered within 35 minutes in the last check. Try a small
  job first.
• A workflow run again starts from the models' text, not the corrections made in the last run.

Included software

Kalamu bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD). The
exact source of the bundled FFmpeg, its build configuration and licences are in the
ffmpeg-9.0.2-1 release. Offline
transcription uses whisper.cpp (MIT). The fonts are under the SIL Open Font License. Every other
component is listed with its licence in licenses/THIRD-PARTY-NOTICES.txt.

See the roadmap for what comes
next.

Kalamu 1.2.0

Kalamu 1.2.0 reads the articles of a newspaper issue, or the letters of an archival file, on
their own: a table of contents' entries become items you can correct, then extract, analyse or
export one by one. It also brings simpler default prompts and a data folder named for Kalamu.
For what 1.1.2 changed, see the
1.1.2 release.

What's new since 1.1.2

• Items of a table of contents: in a table of contents' review, Make items from the
  entries turns each checked entry into an item: an article, a letter, a chapter, with its
  title and pages. Correct an item's pages (1, 3-5, 8 (left)), join a continuation, split,
  add or remove items; a page shared by several items is named as such, since a whole-page cut
  cannot isolate one article's text. Send the items you select to a new text extraction or
  analysis: each is read as a PDF of its own pages, its results are written under the item's
  name beside the source, and provenance.json names the item. Giving an item other pages
  marks the jobs that read it out of date. Export writes each selected item's pages as a PDF,
  with the list as CSV and JSON.
• One default prompt per kind of material: text extraction keeps three, for a printed or
  typed document, a newspaper or magazine, and a handwritten document or manuscript, each for
  any language and either reading direction, Arabic and Ajami included. The lines (as on the
  page, or as running text) are chosen in New job for every prompt. Thirteen defaults replace nineteen,
  with one shared set of marks ([illegible], [word?], [margin: ...]). Saved pipelines,
  workflows and shared files that named a retired prompt use its replacement; an edited copy
  of one stays your own prompt.
• The data folder is called Kalamu Data: on Windows, %LOCALAPPDATA%\Kalamu Data instead
  of io.github.fmadore.kalamu; on macOS, ~/Library/Application Support/Kalamu. The first
  start of 1.2.0 moves the earlier folder once, with your jobs, results and settings; if it
  cannot, the old folder stays in use. Settings shows where it is, also for the Store version.
• Searchable PDFs, shown before you start: New job's diagram for text extraction says
  whether the chosen model also gives a searchable PDF (today Mistral OCR) or text only.
• Clearer interface text: a copy-editing pass over the whole interface; the glossary names
  the controls it explains as New job does.

Fixed

• Mistral Large 4 is estimated at its current price, half what the catalogue had.

Download

• Microsoft Store: the Store copy installs without a warning and updates itself:
  Kalamu in the Microsoft Store.
• Kalamu_1.2.0_x64-setup.exe: Windows 11 on Intel and AMD PCs.
• Kalamu_1.2.0_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops).
• Kalamu_1.2.0_aarch64.dmg: macOS 13 or later on a Mac with Apple Silicon (M1 and later).
  Open it and drag Kalamu to Applications.
• kalamu-samples.zip: a short recording, scanned pages, a newspaper issue and a text to try
  the app with.
• Kalamu_1.2.0.msixbundle and the two .msix packages are the Store submission and its
  per-architecture packages, for testing; install Kalamu from the Store or the installers.

Every download bundles FFmpeg and offline Whisper; nothing else needs to be installed.
Installing 1.2.0 over an earlier version keeps your jobs, settings, presets and keys.

Known limitations

• Items are whole pages of a PDF. Two articles on one page each take the whole page; a
  column or passage cannot be cut out yet. A multi-page image's items can be made and
  corrected, but not read or exported on their own. The numbers printed on the pages are not
  read yet: an item's pages are numbered as the file's.
• Unsigned installers: Windows SmartScreen will warn. Choose More info, then Run anyway,
  or install from the Store.
• The Mac app is not signed with an Apple Developer ID, so the first time macOS refuses to
  open it: open System Settings → Privacy & Security, choose Open Anyway beside Kalamu,
  and confirm. macOS may also ask once, and again after an update, to let Kalamu use the key
  saved in your Keychain.
• The Mac version is new: it is built and tested on Macs in GitHub Actions, but has not yet
  been used on a real Mac. Please report what does not work with
  Report a problem. Intel Macs are not
  supported, and the application menu is in English only.
• Gemini (Gemini 3.5 Transcribe included), offline Whisper, economy mode, tables of contents,
  corrections and the dictionary's proposed merges have had fewer runs against the live
  services than OpenAI, Claude and Mistral text extraction: try a small job first.
• A workflow run again starts from the models' text, not the corrections made in the last run.
• English interface only; French and German will follow.

Included software

Kalamu bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD). The
exact source of the bundled FFmpeg, its build configuration and licences are in the
ffmpeg-9.0.2-1 release. Offline
transcription uses whisper.cpp (MIT). The fonts are under the SIL Open Font License. Every other
component is listed with its licence in licenses/THIRD-PARTY-NOTICES.txt.

See the roadmap for what comes
next.

Kalamu 1.1.2

Kalamu 1.1.2 is the first release for Macs with Apple Silicon, and makes Gemini 3.5 Transcribe,
Google's speech-to-text model, Gemini's model for recordings. For what 1.1.1 changed, see the
1.1.1 release.

What's new since 1.1.1

• Kalamu on a Mac: a disk image for Apple Silicon, macOS 13 and later, with FFmpeg and
  offline Whisper inside, as on Windows. Shortcuts use ⌘ where Windows uses Ctrl; keys are kept
  in the macOS Keychain; closing the window keeps Kalamu in the Dock, the menu-bar icon shows
  running jobs, and ⌘Q asks first while jobs run. Recordings the Mac's player cannot play as
  they are (Ogg, WebM) are converted to FLAC for the review.
• Gemini 3.5 Transcribe is Gemini's model for recordings, in place of Gemini 3.8 Flash,
  which is still offered. It labels speakers and times every line, and a chosen language is
  passed on to it; with none, it detects the language. YouTube links still go to Gemini 3.8
  Flash, which watches the video from its link.
• The languages a model supports: under Gemini 3.5 Transcribe and Whisper, New job lists the
  languages they document, in alphabetical order, and offers only those.
• The interface names the keychain and the notification area as each system calls them.

Fixed

• A folder added from a drive a Mac has used no longer brings in the hidden ._name files
  macOS leaves there.

Download

• Microsoft Store: the Store copy installs without a warning and updates itself:
  Kalamu in the Microsoft Store.
• Kalamu_1.1.2_x64-setup.exe: Windows 11 on Intel and AMD PCs.
• Kalamu_1.1.2_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops).
• Kalamu_1.1.2_aarch64.dmg: macOS 13 or later on a Mac with Apple Silicon (M1 and later).
  Open it and drag Kalamu to Applications.
• kalamu-samples.zip: a short recording, scanned pages, a newspaper issue and a text to try
  the app with.
• Kalamu_1.1.2.msixbundle and the two .msix packages are the Store submission and its
  per-architecture packages, for testing; install Kalamu from the Store or the installers.

Every download bundles FFmpeg and offline Whisper; nothing else needs to be installed.
Installing 1.1.2 over an earlier version keeps your jobs, settings, presets and keys.

Known limitations

• Unsigned installers: Windows SmartScreen will warn. Choose More info, then Run anyway,
  or install from the Store.
• The Mac app is not signed with an Apple Developer ID, so the first time macOS refuses to
  open it: open System Settings → Privacy & Security, choose Open Anyway beside Kalamu,
  and confirm. macOS may also ask once, and again after an update, to let Kalamu use the key
  saved in your Keychain.
• The Mac version is new: it is built and tested on Macs in GitHub Actions, but has not yet
  been used on a real Mac. Please report what does not work with
  Report a problem. Intel Macs are not
  supported, and the application menu is in English only.
• Gemini 3.5 Transcribe follows Google's documentation and has not yet run against the live
  service: try a short recording first, and check its estimated cost against your Google bill.
• Gemini, offline Whisper, economy mode, tables of contents, corrections and the dictionary's
  proposed merges have had fewer runs against the live services than OpenAI, Claude and Mistral
  text extraction: try a small job first.
• A workflow run again starts from the models' text, not the corrections made in the last run.
• English interface only; French and German will follow.

Included software

Kalamu bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD). The
exact source of the bundled FFmpeg, its build configuration and licences are in the
ffmpeg-9.0.2-1 release. Offline
transcription uses whisper.cpp (MIT). The fonts are under the SIL Open Font License. Every other
component is listed with its licence in licenses/THIRD-PARTY-NOTICES.txt.

See the roadmap for what comes
next.

Kalamu 1.1.1

Kalamu 1.1.1 keeps your place: the glossary, correcting a text and zooming a page no longer
send you back to the start, and each job in the list has its own menu. For what 1.1.0 added
(projects, a project dictionary, workflows), see the
1.1.0 release.

What's new since 1.1.0

• A menu on each job in the list: right-click a job, or press the keyboard's menu key, to
  rename it in place, add it to one of your projects, or delete it.
• Words explained where they are: the ? beside a word opens its definition in a bubble beside
  it, so a job being set up in New job stays as it was. A link in the bubble opens the whole
  glossary, now in alphabetical order.
• Correct where you are reading: Correct opens the editor on the word you clicked or
  selected, or on the first line in view, and keeps that line where it stood on screen. The
  editor is as tall as the text, so it scrolls as the text did.
• Large windows: the Workflows page sets its examples side by side, as many as fit.

Fixed

• A page zoomed wider than its pane can be scrolled to its left edge again.
• New job no longer scrolls sideways when an estimate line is long, and its cost column reaches
  the foot of the page however long the form.
• Thinking lists each level once, the suggested one marked, rather than the suggested level
  twice.
• kalamu-samples.zip includes a whole 16-page newspaper issue of 1915, scans only, to try a
  table of contents with.

Download

• Microsoft Store: the Store copy installs without a warning and updates itself:
  Kalamu in the Microsoft Store.
• Kalamu_1.1.1_x64-setup.exe: Windows 11 on Intel and AMD PCs.
• Kalamu_1.1.1_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops).
• kalamu-samples.zip: a short recording, scanned pages, a newspaper issue and a text to try
  the app with.
• Kalamu_1.1.1.msixbundle and the two .msix packages are the Store submission and its
  per-architecture packages, for testing; install Kalamu from the Store or the installers.

Every download bundles FFmpeg and offline Whisper; nothing else needs to be installed.
Installing 1.1.1 over 1.1.0 or 1.0.0 keeps your jobs, settings, presets and keys.

Known limitations

• Unsigned installers: Windows SmartScreen will warn. Choose More info, then Run anyway,
  or install from the Store.
• Gemini, offline Whisper, economy mode, tables of contents, corrections and the dictionary's
  proposed merges have had fewer runs against the live services than OpenAI, Claude and Mistral
  text extraction: try a small job first.
• A workflow run again starts from the models' text, not the corrections made in the last run.
• English interface only; French and German will follow.

Included software

Kalamu bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD). The
exact source of the bundled FFmpeg, its build configuration and licences are in the
ffmpeg-9.0.2-1 release. Offline
transcription uses whisper.cpp (MIT). The fonts are under the SIL Open Font License. Every other
component is listed with its licence in licenses/THIRD-PARTY-NOTICES.txt.

See the roadmap for what comes
next.

Kalamu 1.1.0

Kalamu 1.1.0 keeps a researcher's material together across jobs: projects with their
collections and documents, a dictionary of the names your analyses found, and workflows that
run one step after another, stopping where you want to check.

What's new since 1.0.0

• Projects. Gather documents in collections and tags, a document in several without being
  copied. Each document shows every result read from it, whatever job made it, with how much
  was reviewed and what it cost; mark the one you prefer. Add files before reading them, add
  jobs you ran before, start a new job or a workflow from a selection, and see the project's
  activity filtered by task, model or date, costed together.
• A project dictionary of the people, places and terms your analyses listed, across the
  project or one collection: merge the spellings that are one person or place under a label of
  your choosing, take them apart again, keep two apart, mark them reviewed, and export the
  result as JSON and CSV. A model can propose merges for a field; nothing is merged until you
  accept a proposal.
• Workflows: pipelines run one after another on the same files, each step reading the files
  or an earlier step's corrected text. A step can stop for your review, and the next waits until
  you continue. Then, at the end of New job, analyses the text once you have checked it, with
  no workflow to build, and three ready-made examples show the rest.
• Use this text in a new job: send any finished extraction, transcription or correction, as
  you corrected it, into an analysis, a correction or a workflow, without reading the originals
  again. An analysis of text corrected since is marked out of date.
• Several pages per call for text extraction, so a model sees an article that runs over the
  next page; each page is still reviewed and exported on its own.
• Thinking for models that reason, set to a suggestion for each task, and every answer
  allowed the whole output its model is documented to give.

Fixed

• A job reading several pages per call no longer sends pages already read again after the model
  catalogue is updated, and a call whose answer could not be split into pages now counts in what
  the job cost.
• A workflow step started at launch is tied to its project's documents; a step is never made
  twice, nor started after its run is cancelled, and a step that could not start for want of
  text starts once the text is read.
• A searchable PDF is written only by a model that reports where text sits on the page, whatever
  a pipeline asked for.
• Values not known yet (a cost, an error rate) are read out as such by screen readers rather
  than as a dash.

Download

• Microsoft Store: the Store copy installs without a warning and updates itself:
  Kalamu in the Microsoft Store.
• Kalamu_1.1.0_x64-setup.exe: Windows 11 on Intel and AMD PCs.
• Kalamu_1.1.0_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops).
• kalamu-samples.zip: a short recording, a scanned page and a text to try the app with.
• Kalamu_1.1.0.msixbundle and the two .msix packages are the Store submission and its
  per-architecture packages, for testing; install Kalamu from the Store or the installers.

Every download bundles FFmpeg and offline Whisper; nothing else needs to be installed.
Installing 1.1.0 over 1.0.0 keeps your jobs, settings, presets and keys.

Known limitations

• Unsigned installers: Windows SmartScreen will warn. Choose More info, then Run anyway,
  or install from the Store.
• Gemini, offline Whisper, economy mode, tables of contents, corrections and the dictionary's
  proposed merges have had fewer runs against the live services than OpenAI, Claude and Mistral
  text extraction: try a small job first.
• A workflow run again starts from the models' text, not the corrections made in the last run.
• English interface only; French and German will follow.

Included software

Kalamu bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD). The
exact source of the bundled FFmpeg, its build configuration and licences are in the
ffmpeg-9.0.2-1 release. Offline
transcription uses whisper.cpp (MIT). The fonts are under the SIL Open Font License. Every other
component is listed with its licence in licenses/THIRD-PARTY-NOTICES.txt.

See the roadmap for what comes
next.

Kalamu 1.0.0

The first version of Kalamu for the Microsoft Store, and the first that is not a test build:
read scanned and handwritten documents, transcribe interviews and analyse texts with the AI
provider of your choice, or offline, and check every result against its source.

What's new since 0.4.0

• Search in the review (Ctrl F): find a name or a phrase across every page of a document
  or every piece of a recording, corrections included, and go to each match, a recording
  playing from its line. Capitals, accents and line breaks are forgiven. In an analysis, the
  fields whose answers or reasons hold the words are kept, and in a text document the words are
  marked in the document too; in a table of contents, the entries are filtered.
• A glossary, beside Settings, for colleagues new to these tools: OCR and HTR, LLM,
  tokens, reasoning effort, ground truth, error rates, hallucination, provenance and more, in
  plain words, and what each export format suits. A small help mark beside the headings that
  use these words opens it on the right one.
• Report a problem. A flag in every review's header, and a button in Settings › About,
  open a GitHub form to report output that is inappropriate, harmful or invented, or a bug. The
  form opens in your browser; Kalamu sends nothing itself.
• One frame for the three reviews: the text, analysis and table of contents reviews share
  their header, source pane, handles, strip of pages and keys (Ctrl Enter to mark reviewed,
  Page Up and Page Down to turn pages).
• Results that cannot be wrong unnoticed: an answer cut off at the model's output limit is
  marked and listed in provenance; a file changed on disk since it was added is refused.
• Work neither lost nor paid twice: quitting waits for the pieces already sent, and a job
  shows what it cost over all its runs.
• Results that reach you: a document's files are written even when one page keeps failing,
  that page marked [Not read: reason].
• Error rates after review, for any job, against every page you corrected or marked as
  reviewed.
• Provenance of each request: provenance.json records the settings each request carried
  (reasoning effort, output limit, media resolution), for both steps of a table of contents.
• Tables of contents: the review follows the page shown, and the entries can be written in
  the language you choose.
• Links open in your browser in the installed app: YouTube links, and the links in
  Settings, did nothing in 0.4.0.
• New job offers only what will run: which files, models and formats suit each task now
  comes from the engine, which checks them again when a job is created, so a pipeline saved
  for another task cannot start a job a model cannot do.
• Icons centred in their buttons everywhere: an icon-only button could draw its icon off to
  the left.
• Third-party notices: every crate and package built into the app, with its licence, in the
  licenses folder beside FFmpeg's.

Download

• Microsoft Store: the Store copy installs without a warning and updates itself (link once
  the listing is live).
• Kalamu_1.0.0_x64-setup.exe: Windows 11 on Intel and AMD PCs.
• Kalamu_1.0.0_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops).
• kalamu-samples.zip: a short recording, a scanned page and a text to try the app with.
• Kalamu_1.0.0.msixbundle and the two .msix packages are the Store submission and its
  per-architecture packages, for testing; install Kalamu from the Store or the installers.

Every download bundles FFmpeg and offline Whisper; nothing else needs to be installed.
Installing the 1.0.0 installer over 0.4.0 keeps your jobs, settings, presets and keys.

Known limitations

• Unsigned installers: Windows SmartScreen will warn. Choose More info, then Run anyway,
  or install from the Store.
• Gemini, offline Whisper, economy mode, tables of contents and corrections have had fewer runs
  against the live services than OpenAI, Claude and Mistral text extraction: try a small job
  first.
• English interface only; French and German will follow.

Included software

Kalamu bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD). The
exact source of the bundled FFmpeg, its build configuration and licences are in the
ffmpeg-9.0.2-1 release. Offline
transcription uses whisper.cpp (MIT). The fonts are under the SIL Open Font License. Every other
component is listed with its licence in licenses/THIRD-PARTY-NOTICES.txt.

See the roadmap for what comes
next.

Kalamu 0.4.0

Two new tasks for researchers who work with long documents and with text that libraries have
already machine-read. This is still an early pre-release: expect rough edges, and check
results against your sources.

What's new

• Detailed table of contents of a long document scanned as one file: a newspaper or
  magazine issue, an archival file, a book or report. One model reads each page for the
  articles or items on it; another, which can be a cheaper model from any provider (Gemini Pro
  reading, Mistral combining, say), puts the pages together into one entry per article or item,
  with every page it is on and the fields you choose. Ready-made kinds of document:

  • a newspaper or magazine issue: section, title, authors, pages and a summary of each article;
  • an archival file, item by item: kind of document, date, sender, addressee, summary and
    names, as an archivist's inventory gives them;
  • a book, report or thesis: parts, chapters and sections with a summary of each.

  The review shows the scan beside the entries: click an entry to see its first page, and
  correct, add or remove entries. Results come as Markdown, Word, a spreadsheet with one row per
  entry, text and JSON.

• Correct existing text: the noisy text layer of a PDF, or a text or Word file downloaded
  from a digital library (Gallica, Europeana, Internet Archive…), corrected page by page with
  its spelling of the time kept. The review highlights every change against the original, and
  the original reading can be taken back in one click. A newspaper preset follows the
  OCR-correction prompt of iwac-ai-pipelines.

• How it works: under the tasks in New job, a short animated diagram shows how the chosen
  task splits and joins the work, drawn with your own files once they are added.

• Jobs: open the files a job wrote from its page, keep that list after a restart, and
  rename jobs.

• DeepSeek V4.1 Flash, hosted by Mistral, for analyses, in economy mode too.

• Recordings play at once in the review, straight from the file, instead of being converted
  first.

Download

• Kalamu_0.4.0_x64-setup.exe: Windows 11 on Intel and AMD PCs.
• Kalamu_0.4.0_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops).

Both bundle FFmpeg and offline Whisper; nothing else needs to be installed. Installing over
0.3.0 keeps your jobs, settings, presets and keys.

Known limitations

• Unsigned installers: Windows SmartScreen will warn. Choose More info, then Run anyway.
• Tables of contents and corrections have not yet run against the live services. Try the first
  pages as a sample before a whole issue or file.
• A table of contents reads PDFs and multi-page images: photographs of one archival file should
  first be put together into a single PDF.
• Gemini, offline Whisper and economy mode have not yet run against the live services; OpenAI,
  Claude and Mistral have. Try a small job first.
• English interface only; French and German will follow.

Included software

Kalamu bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD). The
exact source of the bundled FFmpeg, its build configuration and licences are in the
ffmpeg-9.0.2-1 release. Offline
transcription uses whisper.cpp (MIT). The fonts are under the SIL Open Font License.

See the roadmap for what comes
next.

Kalamu 0.3.0

The first test build under the name Kalamu, formerly Scriptorium, and the first for both x64
and ARM64. This is still an early pre-release: expect rough edges, and check results against
your sources.

What's new

• A new name, Kalamu, Swahili for pen, from the Arabic qalam, with a new icon (a
  handwritten sheet behind its transcription, one line under the highlighter) and warmer,
  paper-toned greys in light and dark. On its first start, Kalamu moves the jobs, settings,
  presets, Whisper models and API keys of Scriptorium 0.1 or 0.2 over to its own name.
• Economy mode: send a job to the provider's batch service, with Gemini, OpenAI, Claude or
  Mistral, at half the price. Answers come within a day, and the app can be closed in the
  meantime. Cached input is billed at its own, lower price, and both costs are shown before
  starting.
• YouTube videos are transcribed from their link with Gemini, without downloading them, and
  reviewed beside the video.
• Pipelines can be exported and imported: a job's models, settings and full prompts as a
  JSON file to share with colleagues, and a Markdown record to cite in a methods section.
• Analyses:
  • A ready-made analysis of scholarly publications, giving their question, argument, methods
    and findings for a literature review.
  • Photographed pages and scanned images can be analysed directly with Gemini, OpenAI or
    Claude, as PDFs already could, without extracting their text first.
• More models: GPT-6.1 Terra, Mistral's chat models for analyses, and the Z.ai GLM models
  hosted by Mistral. A model is sent only the inputs it accepts: a text-only model asks for the
  text to be extracted first.
• The strip of pages shows thumbnails of scans and images in the review.
• About in Settings links each project Kalamu credits, and an optional Buy Me a Coffee link
  sits at the foot of the sidebar. It unlocks nothing: every feature stays free.

Fixes

• Scans compressed with JPEG 2000 or JBIG2, common in library PDFs, now show in the review.
• New job's tips match the task: analyses no longer get the tips for recordings.

Download

• Kalamu_0.3.0_x64-setup.exe: Windows 11 on Intel and AMD PCs.
• Kalamu_0.3.0_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops).

Both bundle FFmpeg and offline Whisper; nothing else needs to be installed.

Coming from Scriptorium 0.1 or 0.2? Windows treats Kalamu as a separate app, so it installs
beside Scriptorium rather than over it. Install Kalamu and start it once, so it moves your jobs,
settings and keys over, then uninstall Scriptorium from Settings → Apps.

Known limitations

• Unsigned installers: Windows SmartScreen will warn. Choose More info, then Run anyway.
• Gemini, offline Whisper and economy mode have not yet run against the live services; OpenAI,
  Claude and Mistral have. Try a small job first.
• The YouTube player inside the app is not yet checked in the installed build; if it stays
  blank, the transcript is still made.
• English interface only; French and German will follow.

Included software

Kalamu bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD). The
exact source of the bundled FFmpeg, its build configuration and licences are in the
ffmpeg-9.0.2-1 release. Offline
transcription uses whisper.cpp (MIT). The fonts are under the SIL Open Font License.

See the roadmap for what comes
next.

Scriptorium 0.2.0 (test pre-release)

The second test build of Scriptorium, with a new interface. This is still an early pre-release: expect rough edges, and check results against your sources. Text extraction, transcription and analyses have now run against the live OpenAI, Claude and Mistral APIs with real keys; Gemini has not yet.

What's new

• A new interface, "Graphite and highlighter": grey frame, white work surface, one black button per screen with its keyboard shortcut, and the highlighter yellow kept for where models disagree. Instrument Sans for the interface, Source Serif 4 for the text being checked, IBM Plex Mono for what a machine produced. All fonts ship with the app, so it works offline.
• The sidebar lists your jobs: running, unfinished and finished, with their progress. New job is always at the top (Ctrl N). The list folds to a narrow rail (Ctrl B) when you need room.
• A new review:
  • The scan sits on a dark lightbox so the page is the brightest thing on screen.
  • Pointing at a highlighted word shows what each model read there; click the correct reading, or type your own, and it is saved as a correction. The model's own text is always kept.
  • Compare opens another model's whole text beside yours, with the same differences marked in both.
  • The panes can be resized by dragging the lines between them.
  • A strip of pages shows what is reviewed. F and Shift F step through differences, Ctrl Enter marks a page reviewed.
• Languages are chosen from a searchable list with favourites, and only the choices the selected models support are offered.
• A new app icon: a page with one line highlighted.

Fixes

• Starting a job no longer crashes the app.
• Transcripts give one turn per speaker, with plain speaker names (found in the first live runs).
• New job uses the window's full width.

Download

• Scriptorium_0.2.0_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops). It installs over 0.1.0. An x64 build will follow.

Known limitations

• Unsigned installer: Windows SmartScreen will warn. Choose More info, then Run anyway.
• English interface only; French and German will follow.
• The name is not final (#1).

Included software

Scriptorium bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD); nothing needs to be installed separately. The exact source of the bundled FFmpeg, its build configuration and licences are in the ffmpeg-9.0.2-1 release. Offline transcription uses whisper.cpp (MIT). The fonts are under the SIL Open Font License.

See the roadmap for what comes next.

Scriptorium 0.1.0 (test pre-release)

The first test build of Scriptorium, for trying the app. This is an early pre-release: every feature is implemented and tested against recorded provider responses, but none has yet been run against the live APIs with real keys. Expect rough edges, and check results against your sources.

What it does

• Extract text from PDFs and images, printed or handwritten, with Gemini, OpenAI, Claude or Mistral. Lines are kept as on the page or joined into running text.
• Transcribe audio and video with speakers and timestamps, in the cloud or offline with Whisper (models download from Settings). Speakers can be given their real names in the review.
• Analyse texts with fields you define: a summary, the people, places and organisations named, a tone with a reason, themes from your own list. Ready-made analyses cover the common needs.
• Review every result beside its source, set as a critical edition, with corrections kept apart from the AI output.
• Compare models on the same documents, with their differences and error rates.
• Resume interrupted jobs without paying twice, and save a job's settings as a pipeline.
• Export to text, Markdown, Word, JSON, CSV, subtitles and searchable PDF (from Mistral OCR), each with a provenance.json.

Download

• Scriptorium_0.1.0_arm64-setup.exe: Windows 11 on ARM64 (Snapdragon laptops). An x64 build will follow.

Known limitations

• Unsigned installer: Windows SmartScreen will warn. Choose More info, then Run anyway.
• English interface only; French and German will follow.
• The app icon and name are not final (#1).

Included software

Scriptorium bundles FFmpeg 9.0.2 (audio only) under the LGPL v2.1 or later, with libopus (BSD); nothing needs to be installed separately. The exact source of the bundled FFmpeg, its build configuration and licences are in the ffmpeg-9.0.2-1 release. Offline transcription uses whisper.cpp (MIT).

See the roadmap for what comes next.