# YazSes v1.4.1 — Stable baseline

**Released:** 2026-07-01 · PyPI · Snap (stable + edge) · APT · GitHub

v1.4.1 is the **stable baseline** of YazSes as a focused, offline, hold-to-talk
voice-dictation and voice-command daemon. Everything in the "dictation era" of the
project works well and ships cleanly across all channels from the canonical home
[`MSKazemi/yazses`](https://github.com/MSKazemi/yazses).

This release is deliberately marked as a checkpoint: the next line of work opens a
**new area** for YazSes (a broader voice-first human–computer interaction layer),
which will be developed and released separately. v1.4.1 is the version to fall back
to for a rock-solid, single-purpose dictation experience.

## What works well as of v1.4.1

- **Hold-to-talk dictation** into any focused app on Linux, macOS, and Windows,
  fully offline via faster-whisper (CPU int8) — no cloud, no API key.
- **Voice commands & basic keystrokes** (Tier-1 regex grammar, optional SLM router).
- **Voice punctuation & formatting** (opt-in): "comma", "period", "new line", …
- **Selectable injection backend** (`auto` / `type` / `clipboard` / `wtype`) with a
  Wayland flood guard for Ubuntu 26+ compositors.
- **Opt-in encrypted on-device learning** (`yazses tune`, `mark-wrong`, corpus) and
  the dormant v2 cognitive-layer scaffolding — all off by default, privacy-first.
- **Turnkey ops**: `yazses doctor`, `setup`, `features`, `vocab`, `hotkey`, `restart`.

## Fixed in v1.4.1

- **Cross-platform imports** — `system/setup`, `inject/auto`, `inject/clipboard`
  now import cleanly on Windows/macOS (lazy `grp`/`pwd`, guarded `os.getuid`),
  restoring a green test matrix on Linux × macOS × Windows for Python 3.11 & 3.12.
- **Snap release reliability** — a tagged release on the canonical repo now fails
  loudly when the Snap Store credential is missing instead of silently skipping
  the publish; PyPI publishing is idempotent (`skip-existing`).

## Install

```bash
pip install yazses            # or: uv tool install yazses
sudo snap install yazses      # stable channel
```

## Looking ahead

The v2 effort explores a set of novel, privacy-first, on-device features that extend
YazSes from "dictation" toward a broader voice-first interaction layer. Design
artifacts (ADRs, specs, architecture, roadmap) are tracked in `design/` and the
active plan in `.claude/plans/`. v1.4.1 remains the recommended stable release until
v2 features graduate from experimental.