Releases: AlexanderAbramovPav/scriba
Releases · AlexanderAbramovPav/scriba
scriba v0.1.0
First public release of scriba — free, local, speaker-diarized meeting transcription for your AI agent or second brain.
Highlights
- 100% on-device (whisperX large-v3 + pyannote
community-1), no cloud, no uploads. - Word-level speaker attribution + honest per-word confidence (overlap / ASR / attribution) in a schema-versioned
transcript.jsonsidecar for downstream AI. - Glossary biasing, known-speaker enrollment (
--enroll), and an optional GigaAM-RU backend (--asr gigaam, ~−50% WER on Russian). - Portable per-recording layout + hidden
.scriba/corpus index; live progress (statusline /--watchTUI / macOS notification / Shell-details heartbeat). - Agent-agnostic — Claude Code, Codex, Cursor, Aider via
AGENTS.md, or a plain CLI with no AI at all. - CI on Python 3.11 / 3.12.
See CHANGELOG.md for the full list.