Aquesta és una versió anterior. La versió actual és la 3.1.7. Kivio 3.1.7 · Data de publicació:
Fitxers d’instal·lació 2.8.7
| Arquitectura | Tipus | Àmbit | Mida del fitxer | Baixa | SHA256 |
|---|---|---|---|---|---|
| 64 bits | EXE (NSIS) | Usuari actual | — | github.com | VirusTotal |
L’enllaç i l’SHA256 provenen del manifest comunitari del Gestor de paquets del Windows. Convé comparar el hash del fitxer baixat abans d’executar-lo.
Notes de la versió
- Expanded external-CLI agent support — Claude Code now supports ask-user, plan approval, background tasks, and live subagent progress; subagent calls render as dedicated cards. File-read limits, context compaction, empty-response handling, and usage accounting are further aligned with Pi.
- Pi and pi-btw integration — Pi provider configuration joins the shared external-CLI system, and pi-btw events map into the common run protocol. Mid-stream upstream failures retain retry behavior and are presented separately from ordinary process exits.
- Long-paste attachments — oversized pasted text becomes an editable in-memory attachment, with complete support across persistence, regeneration, external CLIs, export, and steering, so its body is no longer silently lost.
- Chat interaction and rendering fixes — fixes raw <details> tags, adjacent Chinese bold text not rendering, ask-user answers collapsing, image context menus being swallowed by the WebView, and a window-resize race that disabled streaming scroll follow.
- Status motion and dark mode — the message stream gains a persistent run-status row, with more accurate background-task counts and sidebar activity. Status animation issues seen on some machines, especially in dark mode, are fixed, while WebView hover redraw cost is reduced.
- Memory-command filtering — tightens and corrects safety matching for command-like wording so normal content is not mistakenly filtered, and cleans up related composer edge states.
- OpenCode, models, and usage — completes native OpenCode provider configuration; external-CLI context usage now updates per turn, records time to first token and effort level, and fixes model mapping and accounting boundaries.
- More polish — supports queued messages and immediate steering during generation, runs update checks in place, leaves reading space below newly sent messages, tightens the provider/CLI settings layout, and fixes duplicate password-visibility controls in Edge/WebView2.
