Herunterladen
Laden Sie Naia herunter und beginnen Sie mit Ihrem KI-Avatar zu chatten.
v0.2.0 stellt signierte automatische Aktualisierungen wieder her. Der veraltete latest.json-Endpunkt von nextain/naia-os wird durch nextain/naia-shell ersetzt, die Signaturkompatibilität mit naia-tauri.key bleibt erhalten, die Sprach-payload-Korrekturen aus v0.1.9 sind enthalten und die main-Rolle wird nach der Anmeldung in den Einstellungen synchronisiert. Die Upgrade-Pfade von v0.1.8 und v0.1.9 wurden per E2E geprüft.
- Ship persistent, workspace-isolated Naia Memory through the pinned Shell-Agent runtime, with Korean-capable offline semantic retrieval and safe keyword-only fallback.
- Configure a custom OpenAI-compatible Base URL for local and self-hosted servers.
- Use consistent keyboard shortcuts to approve or reject tool permission requests exactly once.
- Separate reasoning from final answers and prevent hidden reasoning from reaching speech output.
Windows 10 / 11 x86_64. Während der Installation werden Administratorrechte angefordert. Die Registrierung im Microsoft Store steht derzeit noch aus.
Systemanforderungen
- Windows 10/11 x86_64
- Optionale Funktion — Naia Host Audio: NVIDIA-GPU mit mindestens 6GB VRAM
Prüfsummen überprüfen
Um die Integrität der heruntergeladenen Dateien zu überprüfen:
sha256sum -c SHA256SUMSVersionshinweise
v0.2.2 (2026-08-23)
Chat, local-provider, workspace, permission, BGM, localization, and CLI reliability improvements. VoxCPM2 audio distortion remains tracked separately and is not part of this release note.
- feature(memory)Ship persistent, workspace-isolated Naia Memory through the pinned Shell-Agent runtime, with Korean-capable offline semantic retrieval and safe keyword-only fallback. #174
- feature(providers)Configure a custom OpenAI-compatible Base URL for local and self-hosted servers. #475
- feature(permissions)Use consistent keyboard shortcuts to approve or reject tool permission requests exactly once. #477
- fix(chat)Separate reasoning from final answers and prevent hidden reasoning from reaching speech output. #479
- refactor(voice)Apply one shared text-normalization pipeline before LLM replies reach TTS. #480
- feature(workspace)Preview MP3, WAV, and MP4 files with native audio and video controls. #482
- feature(chat)Render GFM, syntax-highlighted code blocks, Mermaid diagrams, and workspace file output in chat. #483
- feature(cli)Register the naia Shell command and open a supplied file while renaming the managed Discord launcher to naia-gateway. #484
- fix(bgm)Accept delayed playback acknowledgements and report blocked embeds without false failure or silence. #430
- fix(workspace)Use the platform primary modifier when opening terminal file paths and preserve ordinary selection behavior. #432
- fix(i18n)Apply the saved or OS locale before first render and keep recovery UI localized. #437
- fix(apps)Use one canonical app directory for install, listing, removal, and safe legacy migration. #472
- fix(workspace)Load the workspace file tree independently when the Herdr server is stopped or unavailable. #492
- feature(bgm)Optionally hide the YouTube background video while its audio keeps playing. #493
- fix(bgm)Animate the musical-note indicator from authoritative YouTube playback events. #494
- fix(bgm)Show Stop while playback is active and Play after playback stops. #495