# Kalamu

- **Years:** Since 2026
- **Application:** Windows 11 (x64, ARM64)
- **Status:** Released
- **Licence:** MIT
- **Interface:** English
- **Methods:** Rust, Tauri, SvelteKit, Svelte 5, TypeScript, LLM, Prompt Engineering, OCR, HTR, Audio transcription, Whisper, NER, Sentiment analysis, Reproducible Research
- **Web page:** https://www.frederickmadore.com/digital-humanities/kalamu

*A free, open-source Windows app for researchers: read scanned and handwritten documents, transcribe interviews and analyse texts with Gemini, OpenAI, Claude or Mistral, or offline with Whisper, and check every result against its source. In the Microsoft Store.*

More and more colleagues ask me how to use AI to transcribe their recordings and read their documents. My [Colab notebooks](https://www.frederickmadore.com/digital-humanities/zmo-ai-pipelines.md) and the [Python pipelines](https://www.frederickmadore.com/digital-humanities/iwac-ai-pipelines.md) behind the Islam West Africa Collection already do this, but they assume some technical comfort: files uploaded to Colab and steps followed in order, or Python, a terminal and a configuration file. Kalamu packages the same workflows as an ordinary Windows app. Install it, paste an API key, pick a file or a folder and press Start.

*Kalamu* is Swahili for pen, from the Arabic *qalam*. It is in the [Microsoft Store](https://apps.microsoft.com/detail/9MXT5555BZ1H) for Windows 11.

## What it does

- **Extract text** from PDFs and images, printed or handwritten, with the provider of your choice, keeping the lines as on the page or joining them into running text
- **Transcribe audio and video** with speakers and timestamps, in the cloud or offline with Whisper; public YouTube videos are transcribed from their link
- **Analyse texts** with fields you define: a summary, the people, places and organisations a document names, its tone with a reason, themes from your own list, or the question, argument, methods and findings of a scholarly publication
- **Make a detailed table of contents** of a newspaper issue, an archival file scanned in one piece, a book or a report
- **Correct existing OCR** page by page, keeping the spelling of the time, with every change shown against the original
- **Compare models** on the same documents, their differences highlighted and error rates measured against corrected pages

## Checked against the source

AI output can be fluent and wrong. Every result opens beside its source, page by page or line by line of a recording, and the places where models disagree are highlighted. Corrections are kept apart from the AI's text, which is never overwritten. Each export, from plain text and Word to subtitles and searchable PDF, comes with a `provenance.json` recording the model, the prompt and the settings, and a job's settings can be saved as a pipeline and exported to cite in a methods section.

## What it costs to run

The app is free, and every feature stays free. Offline transcription with Whisper costs nothing. The cloud providers bill their own usage to your own key; work is saved piece by piece so nothing is paid for twice, and large collections can go through the providers' batch services at half the price.

## How it's built

A [Tauri](https://tauri.app/) app with a Rust engine and a [SvelteKit](https://svelte.dev/docs/kit) front end, native on x64 and ARM64 Windows. The prompts, provider clients and provenance format are ported from the Python pipelines, and the app's plumbing (API keys in Windows Credential Manager, the local Whisper engine, MSIX packaging) is shared with [Live Translation and Subtitles](https://www.frederickmadore.com/digital-humanities/live-translation.md). Files go straight from the computer to the chosen provider; offline, nothing leaves it. The code is [MIT](https://github.com/fmadore/Kalamu/blob/main/LICENSE).

## Addresses

- [Microsoft Store](https://apps.microsoft.com/detail/9MXT5555BZ1H)
- [GitHub releases](https://github.com/fmadore/Kalamu/releases)
- [Kalamu](https://github.com/fmadore/Kalamu)