# Skriver

Skriver is a local-first transcription CLI for agents and humans.

## Core idea

Skriver turns meeting audio or video into an evidence bundle:

- transcript.md
- manifest.json
- workflow.md
- summary_draft.json
- low_confidence_segments.json
- speaker_diarization.json
- context_artifacts.json
- screen_ocr.tsv for video

## What it does

- transcribes audio
- transcribes video
- extracts OCR from screenshots
- merges extra context files
- applies glossary corrections
- flags uncertain transcript segments
- optionally runs local diarization

## What it does not do

- no built-in hosted assistant
- no hidden LLM summarizer
- no replacing agent reasoning

## Recommended agent workflow

1. Run `skriver transcribe ...`
2. Read `manifest.json`
3. Read `transcript.md`
4. Review `summary_draft.json`
5. Review `low_confidence_segments.json`
6. Review `context_artifacts.json`
7. Review `speaker_diarization.json` if present
8. Review `screen_ocr.tsv` and `screens/` for video

## Install

macOS/Linux:

`curl -fsSL https://skriver.ansund.com/install.sh | bash`

Windows PowerShell:

`irm https://skriver.ansund.com/install.ps1 | iex`