Repository navigation
Funktionen
michael edited this page Nov 2, 2026
·
2 revisions
English Version below.
Überblick über alles, was HermiVoice kann.
Sprachsteuerung (Push-to-Talk)
- Taste halten, sprechen, loslassen – mit Live-Zustandsfeedback: Bereit → Hört → Erkennt → Sendet → Gesendet/Fehler.
- Alternativ Tastatureingabe, Antworten per TTS vorlesen.
Offline-Spracherkennung
- sherpa-onnx mit Zipformer-Kroko, Deutsch und Englisch parallel, ~57 MB einmaliger Download je Modell, Wechsel ohne Download und ohne Neustart.
- Läuft komplett auf dem Gerät – kein Audio wird übertragen.
Chat-Sessions
- Mehrere Sessions, Favoriten (Stern), durchsuchbare Historie mit Snippet und Trefferzahl.
- Tokenweises Streaming (SSE) mit „Denkt…“-Block und Markdown-Rendering; Export und Teilen als Markdown.
Mehrere Hermes-Bots (Profile)
- Eigener API-Key und eigene Modellwahl je Bot, Wechsel über das Topbar-Dropdown, aktiv persistent.
Cloud-KI-Anbieter
- OpenRouter, OpenAI, Groq, Mistral, DeepSeek – verschlüsselter API-Key je Anbieter, Primärmodell mit Fallback-Modell.
- Konfigurierbare Zeitfenster für erste Antwort und Gesamtantwortzeit.
Darstellung & Komfort
- Dunkles UI, 6 Akzentfarben, Chat-Hintergründe, eigene Avatare, Benachrichtigung, Haptik, TTS.
- Konfig-Backup als passphrase-verschlüsseltes JSON (AES-256-GCM).
Sicherheit
- API-Keys verschlüsselt im privaten App-Speicher,
allowBackup=false, Warnung beihttp://-Adressen außerhalb des lokalen Netzes.
An overview of everything HermiVoice can do.
Voice control (push-to-talk)
- Hold the button, speak, release – with live status feedback: Ready → Listening → Recognizing → Sending → Sent/Error.
- Alternatively type your input, read answers aloud via TTS.
Offline speech recognition
- sherpa-onnx with Zipformer-Kroko, German and English in parallel, ~57 MB one-time download per model, switching without download and without restart.
- Runs entirely on the device – no audio is transmitted.
Chat sessions
- Multiple sessions, favorites (star), searchable history with snippet and hit count.
- Token-by-token streaming (SSE) with „Thinking…“ block and Markdown rendering; export and share as Markdown.
Multiple Hermes bots (profiles)
- Own API key and own model choice per bot, switch via the top bar dropdown, active bot stays persistent.
Cloud AI providers
- OpenRouter, OpenAI, Groq, Mistral, DeepSeek – encrypted API key per provider, primary model with fallback model.
- Configurable time windows for the first response and the total response time.
Appearance & comfort
- Dark UI, 6 accent colors, chat backgrounds, custom avatars, notification, haptics, TTS.
- Config backup as passphrase-encrypted JSON (AES-256-GCM).
Security
- API keys encrypted in private app storage,
allowBackup=false, warning forhttp://addresses outside the local network.