First packaged release of TalaX — local, adaptive push-to-talk dictation for developers.
Downloads (Linux x86_64)
| Package | For |
|---|---|
TalaX_1.2.0_amd64.deb |
Debian / Ubuntu (sudo dpkg -i) |
TalaX-1.2.0-1.x86_64.rpm |
Fedora / RHEL (sudo rpm -i) |
TalaX_1.2.0_amd64.AppImage |
Any distro (chmod +x and run) |
Verify downloads against SHA256SUMS.txt. macOS and Windows remain source-build (see the README).
Highlights
- Security: TLS certificate-validation fixes in the model-download path (rustls-webpki 0.103.13 — RUSTSEC-2026-0098/0099/0104); npm advisories cleared (vite/esbuild).
- Safe by default: review-first / clipboard-only delivery restored as the default — auto-paste and keystroke type-out are opt-in. Clipboard paste injection now confirms the text landed and settles before restoring your old clipboard (no more lost-race pastes).
- Robustness: config values are validated on save (hotkey, modes, ranges, profile name); IPC handlers recover from a poisoned state mutex; database read errors surface to the UI instead of being silently swallowed; profile clone is atomic with cleanup on failure.
- Learning pipeline: n-gram model now persists as JSON (unmaintained bincode dependency dropped); end-to-end learning-loop regression test added (save corrections → auto-apply promotion → reload → apply).
- Packaging:
cargo tauri buildfixed (frontend paths were broken); deb/rpm/AppImage bundles are built and smoke-tested; CI runs the full gate battery (fmt, clippy-D warnings, cargo audit, cargo-deny, tests, svelte-check, vitest, npm audit, vite build).
Versioning note
This release unifies the version line with the upstream development repo: v1.2.0 follows v0.3.0 on this repository.