Fix Qt worker lifecycle crash

CHANGELOG.md

@@ -24,6 +24,14 @@ Alle wesentlichen Projektänderungen werden hier in verständlicher Form dokumen
 - Strikter Nullkostenmodus, lokale Schutzlimits und inhaltsfreies Provider-Audit.
 - Bereinigte CLI-Befehle `/provider`, `/privacy` und `/status`.
 - Sicheres PowerShell-Startskript ohne Systemänderungen.
+- Gemeinsames Streaming für Ollama und Gemini mit atomarer Speicherung,
+  Gedankenfilter, lesbarer Terminalausgabe und sicherem Strg+C-Abbruch.
+- Automatische lokale Sitzungstitel, `/rename`, nummerierte `/sessions`,
+  `/load <NR|ID>` und begrenzte `/search`-Suche.
+- PySide6-Desktop-MVP mit Sitzungsleiste, Markdownchat, Qt-Streamingworker,
+  sicherem Abbruch, Datenschutzdialog und bereinigter Statusspalte.
+- Sicherer Startweg `scripts/start-javis-gui.ps1`, der die vorhandene
+  Loopback-/Prozesslogik des CLI-Starters wiederverwendet.
 
 ### Changed
 
@@ -39,6 +47,11 @@ Alle wesentlichen Projektänderungen werden hier in verständlicher Form dokumen
   bleiben ladbar.
 - Nicht geheime Providerkonfiguration liegt lokal außerhalb von Git; ein Key wird
   erst unmittelbar vor einem zulässigen Cloudaufruf geladen.
+- Allgemeine unpersönliche Technikfragen werden präziser als `allowed`
+  klassifiziert; persönliche, interne und geheime Inhalte behalten Vorrang.
+- Privacy-Ausgabe unterscheidet lokalen Modus von technischem Fallback.
+- SQLite-Sitzungen werden verlustfrei und idempotent um Titel und zuletzt
+  verwendeten Provider migriert.
 
 ### Security
 
@@ -48,3 +61,7 @@ Alle wesentlichen Projektänderungen werden hier in verständlicher Form dokumen
   Nicht-Billing deaktiviert; Javis aktiviert niemals Billing.
 - Cloudkontext wird lokal klassifiziert und minimiert; 429, Netzwerk-, Timeout-,
   Key- und Modellfehler fallen auf Ollama zurück.
+- Gesundheits-, Symptom-, Verletzungs-, Medikamenten-, Diagnose-, Behandlungs-
+  und medizinische Notfallfragen werden vor Cloudclient-Erstellung strikt lokal behandelt.
+- Gemini-/Datenschutzablauf wurde am 30.07.2026 zusätzlich von Pascal manuell
+  erfolgreich abgenommen; kein Keywert erschien in der Ausgabe.

README.md

@@ -5,9 +5,9 @@ Javis ist ein langfristiges, lokal-first ausgerichtetes Projekt für einen pers
 ## Aktueller Stand
 
 Der lokale Textchat funktioniert auf Windows mit Ollama und `qwen3:8b`.
-Ein optionaler Gemini-Free-Provider ist auf einem Feature-Branch vorbereitet;
-Datenschutzrouter, Nullkostenmodus und lokales Fallback sind automatisiert
-getestet, aber noch nicht von Pascal mit einem echten Key abgenommen.
+Gemini-Free-Hybridrouting, Datenschutz, Streaming und Sitzungsnavigation sind
+implementiert. Zusätzlich steht ein erster PySide6-Desktopclient zur manuellen
+Abnahme bereit.
 
 Noch nicht implementiert sind Obsidian-Zugriff, Serverzugriff, Sprache, Mobile
 Client und Living Mind.
@@ -33,6 +33,15 @@ entsprechenden `JAVIS_*`-Umgebungsvariablen möglich.
 Im Chat zeigt `/help` alle Befehle. Ohne `JAVIS_DATA_DIR` nutzt Javis einen
 sicheren benutzerspezifischen Plattformstandard.
 
+Desktopoberfläche:
+
+```powershell
+powershell -ExecutionPolicy Bypass -File .\scripts\start-javis-gui.ps1
+```
+
+Alternativ: `uv run javis gui`. Die [Desktop-Anleitung](docs/DESKTOP_UI.md)
+beschreibt Tastenkürzel, Datenschutzdialog und den manuellen Abnahmetest.
+
 Gemini bleibt standardmäßig aus. Die spätere lokale Einrichtung erfolgt ohne
 Key in der Konfigurationsdatei:
 

docs/CHATGPT_HANDOFF.md

@@ -58,8 +58,8 @@ Rootserver:
 - Modelle liegen außerhalb von Git unter `D:\Javis-Data\ollama-models`.
 - Der lokale Ollama-Dienst bindet nur an `127.0.0.1:11434`; PATH, Registry,
   Autostart und Windows-Dienste blieben unverändert.
-- CLI-Befehle: `/new`, `/sessions`, `/load <ID>`, `/clear`, `/provider`,
-  `/privacy`, `/status`, `/help`, `/exit`.
+- CLI-Befehle: `/new`, `/sessions`, `/load <NR|ID>`, `/search <Text>`,
+  `/rename <Titel>`, `/clear`, `/provider`, `/privacy`, `/status`, `/help`, `/exit`.
 - Sitzungsdaten liegen über `JAVIS_DATA_DIR` außerhalb von Git; der bestätigte
   Smoke-Test nutzte `D:\Javis-Data\runtime\core-chat-smoke`.
 - Kein Obsidian-, Laptop- oder Rootserverzugriff wurde implementiert.
@@ -99,6 +99,24 @@ Rootserver:
 - `scripts/start-javis.ps1` verifiziert das Repository, setzt nur
   Prozessvariablen, prüft Ollama auf Loopback, startet es bei Bedarf versteckt
   und wartet begrenzt. Pfade sind per Parameter oder `JAVIS_*` überschreibbar.
+- Allgemeine unpersönliche Technikfragen zu RAM/SSD, SQLite, HTTP, Git und
+  ähnlichen Begriffen werden gezielter `allowed`; `never` und `ask` behalten
+  Vorrang bei Secrets, interner Infrastruktur und persönlichem Kontext.
+- Gesundheits-, Symptom-, Blutungs-, Verletzungs-, Medikamenten-, Diagnose-,
+  Behandlungs- und medizinische Notfallfragen sind zwingend lokal `never`.
+- `/privacy` trennt Inhaltsklassifikation, tatsächlichen Provider, Unterdrückung
+  durch Modus `local` und technischen Provider-Fallback.
+- Ollama, Gemini, Hybridrouter, Chatservice und CLI streamen sichtbare Antwortteile.
+  Verdeckte Gedanken werden nicht angefordert oder ausgegeben; Terminal-Markdown
+  wird lesbar bereinigt. Strg+C verwirft Teilantworten atomar und kehrt zum Chat zurück.
+- Desktop-MVP auf `feat/desktop-ui` begonnen: PySide6, gemeinsamer CLI-/GUI-
+  Runtime-Builder, Qt-freier Controller und ruhiges Drei-Spalten-Grundgerüst vorhanden.
+- Sitzungen, Suche, Umbenennen, Markdownanzeige, Qt-Streamingworker, sicherer
+  Abbruch, Modus und bereinigte Statusanzeige sind im Desktopclient verdrahtet.
+- `ask` unterstützt `Einmal erlauben`, `Lokal beantworten` und `Abbrechen`;
+  `never` erreicht weiterhin keinen Cloudfreigabedialog.
+- Startwege `uv run javis gui` und `scripts/start-javis-gui.ps1` funktionieren;
+  der GUI-Starter verwendet die sichere Loopback-/Prozesslogik des CLI-Starters.
 
 ## Aktuelle Architektur
 
@@ -106,7 +124,10 @@ Rootserver:
 - plattformunabhängige Pfade und zentrale Konfiguration
 - Modell-Provider als kleine austauschbare Schnittstelle
 - Ollama-Provider akzeptiert nur lokale HTTP-Loopback-Adressen
-- SQLite-Sitzungsspeicher mit atomaren Benutzer-/Assistentenpaaren
+- SQLite-Sitzungsspeicher mit atomaren Benutzer-/Assistentenpaaren, neutralem
+  Starttitel, lokaler sicherer Titelerzeugung und zuletzt genutztem Provider
+- Streaming wird erst nach vollständigem Abschluss atomar gespeichert; bei Abbruch
+  bleibt weder die Benutzerfrage noch eine unvollständige Antwort im Verlauf.
 - Laufzeitdaten außerhalb von Git über `JAVIS_DATA_DIR` oder sicheren Plattformstandard
 - Hybridprovider bevorzugt Gemini für cloudgeeignete Inhalte und verwendet
   Ollama bei Datenschutz, Ablehnung, Offline-, Quota- und Providerfehlern
@@ -157,6 +178,11 @@ Rootserver:
 - Lesender Ollama-Status prüft nur Loopback-Erreichbarkeit und Modellliste.
 - Sicheres PowerShell-Startskript und kompakte Startdokumentation ergänzt; keine
   PATH-, Registry-, Autostart-, Dienst- oder Richtlinienänderung.
+- Medizinische Inhalte vor Cloudclient-Erstellung strikt auf `never` gesetzt;
+  bestehende SQLite-Daten transaktional und wiederholbar um Titel/Provider migriert.
+- Echtes Ollama-/Gemini-Streaming, providerübergreifendes Fallback vor der ersten
+  Ausgabe und sicheren Streaming-Abbruch nach Teilausgabe implementiert.
+- CLI-Streaming mit lesbarer Markdown-Bereinigung und sicherem Strg+C ergänzt.
 
 ## Aktuelle Tests
 
@@ -171,12 +197,19 @@ Letzter bestätigter Projektstand:
 - uv-Lock und `uv sync --dev`: bestanden
 - Python in `.venv`: 3.12.13
 - Ruff in `.venv`: 0.16.0
-- Unit-Tests: 62 bestanden; Timeout und fehlendes Modell fallen nachweislich lokal zurück
+- Tests: 101 bestanden; zusätzlich Qt-Offscreen-Fenster, Dialogzuordnung und
+  sichere Wiederverwendung des Startskripts abgedeckt
+- `uv run javis gui` im Offscreen-Modus: erfolgreich, Exitcode 0
+- echter lokaler GUI-Smoke: erfolgreich; 1 Sitzung, 2 Nachrichten, Provider Ollama
+- GUI-Startskript mit prozesslokalem ExecutionPolicy-Bypass: erfolgreich
+- manuelle Desktop-Abnahme durch Pascal: noch offen
 - PowerShell-Syntax des Startskripts: erfolgreich geparst
 - Ruff Lint: bestanden
 - Ruff Formatprüfung: bestanden
-- lokaler nachrichtenfreier CLI-Smoke: bestanden; Nullkostenmodus aktiv,
-  Gemini nicht konfiguriert, Schlüssel nicht vorhanden, 0 Cloudanfragen
+- lokaler CLI-Smoke ohne Secretzugriff: bestanden
+- echter Ollama-CLI-Streaming-Smoke mit Neustart/Laden: bestanden
+- echter Gemini-Streaming-Smoke mit öffentlichem Inhalt: bestanden; ein Abschnitt,
+  20 sichtbare Zeichen, 14 Eingabe- und 4 Ausgabetoken; kein Keywert ausgegeben
 - Obsidian-Vault unverändert: 1 Markdown-, 11 Gesamtdateien, letzter
   Schreibzeitpunkt weiterhin 30.07.2026 15:42:31
 - Rootserver und Ubuntu-Laptop wurden nicht kontaktiert
@@ -203,53 +236,35 @@ Abnahmestatus:
 - manuell durch Pascal getestet: am 30.07.2026 erfolgreich; CLI-Start, echte
   deutsche Antworten, Sitzungskontext, `/sessions`, `/exit`, Neustart,
   `/load <UUID>`, vollständiger Verlauf und Fortsetzung
-- Gemini/Datenschutz/Fallback automatisiert durch Codex getestet: ja, nur mit
-  Test-Doubles und lokalem nachrichtenfreiem CLI-Smoke
-- Gemini/Datenschutz/Fallback manuell durch Pascal getestet: noch nicht
-- von Pascal für produktive Verwendung freigegeben: noch nicht ausdrücklich erklärt
+- Gemini/Datenschutz/Fallback automatisiert durch Codex getestet: ja
+- Gemini/Datenschutz/Fallback manuell durch Pascal getestet: am 30.07.2026;
+  verdeckte Key-Speicherung ohne Wertausgabe, vollständiger `/status`, echter
+  Gemini-Erfolg, Cloudzähler, `ask`-Ablehnung, `never`, `/provider local` und
+  Startskript erfolgreich
+- für Merge freigegeben: ja, nach den dokumentierten kleinen Korrekturen
+- für produktive Nutzung freigegeben: nicht separat erklärt; keine Freigabe für
+  spätere Tools, Obsidian oder autonome Aktionen
 
 ## Git-Stand
 
-- aktueller Branch: `feat/gemini-privacy-router`
-- stabiler Branch: `main`
-- Grundgerüstcommit: `6d04171`
-- letzter Commit vor der Handoff-Verdichtung: `0e82748`
-- Handoff-Verdichtung auf `main`: `e6eb995`
-- Python-Entwicklungsumgebung auf dem Feature-Branch: `fa90a5c`
-- Laufzeit-/Modelldokumentation: `10d7cd7`
-- Chatkern und CLI: `ec9b2f9`
-- Unit-Tests: `b153daa`
-- Feature-Dokumentation: `5d567e5`
-- lokaler Merge nach `main`: `f3d6dfb`
-- geprüfter und gepushter Abschlussstand vor diesem Handoff-Update: `0a550e6`
-- Secret-Provider: `570e7fa`
-- Datenschutz/Nullkosten: `286d96d`
-- isolierter Gemini-Provider: `f3e1183`
-- Hybridrouter: `e57186f`
-- lokale Konfiguration: `ab6ea48`
-- CLI-Verdrahtung: `af9426e`
-- Startskript: `141ff2e`
-- Resttests: `a0af15a`
+- aktueller Arbeitsbranch: `feat/desktop-ui`
+- stabiler Ausgangsstand: `main` bei `d3e85f2`
+- GUI-Grundgerüst und Kernadapter: `1658e04`
+- Sitzungen, Streaming, Abbruch und Datenschutz: `cf35148`
+- Tests, Startskript und Desktopdokumentation: `571969b`
+- `feat/desktop-ui` bis `571969b` gepusht und mit Remote synchron verifiziert
+- Chatkomfort-Merge nach `main`: `cbba102`
+- 91 Unit-Tests, Ruff und Strukturprüfung auf dem gemergten `main` bestanden
+- Feature-Branch erfolgreich zu `origin/feat/chat-comfort` gepusht
+- `main` erfolgreich bis `3361eae` zu `origin/main` gepusht und synchron verifiziert
 - `origin` verwendet HTTPS
 - kein Force-Push und keine umgeschriebene Historie
 
-## Kompakter Changelog
-
-- `49a1dd3`: erster Gitea-Verbindungstest
-- `6d04171`: sicheres Javis-Grundgerüst
-- `0e82748`: Werkzeug-, Token- und Handoff-Regeln
-- `e6eb995`: einmalige Archivierung und Verdichtung des Handoffs
-- aktuell: isolierte Python-Toolchain und reproduzierbare Entwicklungsumgebung
-- aktuell: portable Ollama-Laufzeit und `qwen3:8b` außerhalb von Git verifiziert
-- aktuell: lokaler CLI-Chat und persistente SQLite-Sitzungen vollständig abgenommen
-- aktuell: Gemini-Free-Hybridrouting, Datenschutz und Nullkostenmodus
-  automatisiert vollständig geprüft; manueller Live-Test steht aus
-
 ## Offene Entscheidungen und Fehler
 
 - endgültiger Produkt-/Repositoryname bleibt offen
 - Startskript startet Ollama bei Bedarf pro Prozess; kein Autostart oder Dienst
-- Antworten werden noch nicht gestreamt; Sitzungen besitzen noch keine Titel oder Suche
+- Antworten werden gestreamt; Titel, Navigation und Suche sind implementiert
 - genauer späterer Obsidian-Schreibbereich ist nicht freigegeben
 - normaler Secret-Provider ist festgelegt: Betriebssystem-Keyring; eine
   Klartext-XML wird nicht für API-Schlüssel verwendet
@@ -273,9 +288,10 @@ Abnahmestatus:
 
 ## Nächster sinnvoller Auftrag
 
-Pascal führt anhand `docs/NEXT_SESSION.md` den echten Gemini-, Ask- und
-Never-Test mit ausschließlich künstlichen beziehungsweise unpersönlichen Daten
-durch. Bis zu seiner Bestätigung nicht nach `main` mergen.
+Desktop-MVP nach `docs/DESKTOP_UI.md` manuell durch Pascal abnehmen. Danach
+stoppen; vor dieser Abnahme nicht nach `main` mergen und keine Folgephase beginnen.
+Medizinische Antwortqualität bleibt ein späteres Sicherheits-/Systemprompt-Thema.
+Noch keine Obsidian-Integration oder Tools beginnen.
 
 Keinen API-Key in Chat, Git, Obsidian, Logs oder Dokumentation übernehmen.
 

docs/DECISIONS.md

@@ -9,6 +9,7 @@
 | ADR-0005 | Laufenden Handoff kompakt halten und Vollstände nur zu Meilensteinen archivieren | angenommen |
 | ADR-0006 | Ollama und qwen3:8b für den ersten Windows-Chat verwenden | angenommen |
 | ADR-0007 | Gemini kostenlos bevorzugen, sensible Inhalte lokal halten und niemals Billing aktivieren | angenommen |
+| UI-0001 | Desktop-MVP verwendet PySide6 und denselben Javis-Kern wie die CLI | angenommen |
 
 Der normale Runtime-Secret-Speicher ist mit dem Betriebssystem-Keyring festgelegt;
 keine XML-Secret-Datei. Offen bleiben insbesondere der genaue

docs/DESKTOP_UI.md

@@ -0,0 +1,87 @@
+# Javis Desktop
+
+## Installation und Start
+
+Die Desktopoberfläche verwendet PySide6 und denselben Python-Kern wie der CLI-Chat.
+
+```powershell
+& 'D:\Javis-Tools\uv\uv.exe' sync --dev
+& 'D:\Javis-Tools\uv\uv.exe' run javis gui
+```
+
+Der empfohlene Windows-Start prüft zusätzlich Repository, lokale Werkzeuge und
+Ollama und setzt ausschließlich Prozessvariablen:
+
+```powershell
+.\scripts\start-javis-gui.ps1
+```
+
+Falls Windows den direkten Skriptaufruf per Richtlinie blockiert, funktioniert
+ohne systemweite Änderung der prozesslokale Start:
+
+```powershell
+powershell -NoProfile -ExecutionPolicy Bypass -File .\scripts\start-javis-gui.ps1
+```
+
+Der bisherige Start über `.\scripts\start-javis.ps1` beziehungsweise
+`uv run javis chat` bleibt unverändert verfügbar.
+
+## Bedienung
+
+- `Enter`: Nachricht senden
+- `Shift+Enter`: neue Zeile
+- `Ctrl+N`: neue Sitzung
+- `Ctrl+K`: Sitzungssuche fokussieren
+- `Esc`: laufende Antwort abbrechen
+- Doppelklick: Sitzung laden
+- Kontextmenü einer Sitzung: umbenennen
+
+Die linke Spalte verwaltet Sitzungen. In der Mitte erscheinen Verlauf,
+Markdownantwort und Eingabe. Rechts stehen nur bereinigte technische Metadaten:
+Zustand, Modus, Provider, Ollama-Erreichbarkeit, Modell, Datenschutzentscheidung,
+Cloudzähler, Fallback und Datenverzeichnis.
+
+## Provider und Datenschutz
+
+`Hybrid/Auto` verwendet weiterhin den vorhandenen lokalen Datenschutzrouter.
+`Lokal` verhindert Cloudnutzung. Bei `ask` stehen drei Entscheidungen bereit:
+
+- `Einmal erlauben`: minimierten notwendigen Kontext einmalig an Gemini senden
+- `Lokal beantworten`: ohne Cloud mit Ollama antworten
+- `Abbrechen`: Anfrage ohne Speicherung beenden
+
+`never` wird ausschließlich lokal verarbeitet und zeigt keinen Freigabedialog.
+Der Gemini-Key bleibt im Betriebssystem-Keyring und wird in der Oberfläche weder
+ausgegeben noch dokumentiert.
+
+## Streaming und Abbruch
+
+Die Modellgenerierung läuft in einem Qt-Worker. Es kann nur eine Antwort
+gleichzeitig laufen. `Stoppen`, `Esc` oder kontrolliertes Schließen fordern den
+Abbruch an. Ein unvollständiges Benutzer-/Assistentenpaar wird nicht gespeichert.
+
+## Bekannte Grenzen
+
+- keine Sprache, kein Mikrofon und kein Wakeword
+- keine Betriebssystem-, App-, Obsidian-, Server- oder Laptopsteuerung
+- keine autonomen Aktionen und keine neuen Cloudanbieter
+- Schließen kann kurz warten, bis der lokale Netzwerkstream kontrolliert reagiert
+- medizinische Antwortqualität ist ein späteres Sicherheits-/Systemprompt-Thema
+
+## Manueller Abnahmetest für Pascal
+
+1. `.\scripts\start-javis-gui.ps1`
+2. Prüfen, dass sich das Fenster öffnet.
+3. Eine lokale Frage stellen.
+4. Die gestreamte Antwort prüfen.
+5. Eine neue Sitzung anlegen.
+6. Die alte Sitzung laden.
+7. Eine Antwort mit `Esc` oder `Stoppen` abbrechen.
+8. Den Modus `Lokal` testen.
+9. Eine öffentliche Technikfrage im Hybridmodus testen.
+10. Eine `ask`-Anfrage lokal beantworten lassen.
+11. Anwendung schließen und neu starten.
+12. Die gespeicherte Sitzung erneut laden.
+
+Abnahmestatus: automatisierte Tests und echter lokaler GUI-Smoke durch Codex
+bestanden; manuelle Abnahme durch Pascal steht noch aus.

docs/NEXT_SESSION.md

@@ -1,45 +1,26 @@
 # Nächste Sitzung
 
-## Ziel
+## Ausgangsstand
 
-Pascal nimmt den Gemini-Free-Provider und die lokalen Fallbacks manuell ab.
-Bis dahin bleibt `feat/gemini-privacy-router` ungemergt.
+Der Core-Chat, Gemini-/Datenschutz-Meilenstein und das Chatkomfort-Paket sind
+automatisiert geprüft. Pascal hat Core-Chat sowie Gemini/Datenschutz am
+30.07.2026 manuell abgenommen.
 
-## Vorbereitung durch Pascal
+Chatkomfort umfasst sicheres Ollama-/Gemini-Streaming, Strg+C ohne Teilpersistenz,
+medizinisches `never`, automatische und manuelle Sitzungstitel, nummerierte
+Listen, Laden per Nummer/ID-Präfix und begrenzte Suche in Titeln/Benutzertexten.
 
-1. In Google AI Studio einen neuen, nur für Javis verwendeten Gemini-Auth-Key
-   erstellen.
-2. Für das zugehörige Google-Projekt kein Billing verbinden beziehungsweise den
-   externen Billingstatus selbst prüfen.
-3. Den Key niemals in Chat, Codex, Git, Obsidian oder einen sichtbaren
-   Kommandozeilenparameter kopieren.
-4. Im Repository ausführen:
+Das Desktop-MVP auf `feat/desktop-ui` verwendet denselben Kern wie die CLI.
+101 Tests, Qt-Offscreen-Start und echter lokaler Ollama-GUI-Smoke sind erfolgreich.
+Die manuelle Abnahme durch Pascal steht noch aus.
 
-```powershell
-& 'D:\Javis-Tools\uv\uv.exe' run javis configure gemini
-& 'D:\Javis-Tools\uv\uv.exe' run javis secrets set gemini
-powershell -ExecutionPolicy Bypass -File .\scripts\start-javis.ps1
-```
+## Nächster sinnvoller Schritt
 
-Die erste Einrichtung verlangt zwei lokale Bestätigungen. Der zweite Befehl liest
-den Key verdeckt ein und speichert ihn im Windows Credential Manager.
-
-## Manueller Test
-
-1. `/status`: Key und Konfiguration `ja`, Nullkostenmodus `aktiv`.
-2. Unpersönlich: `Erkläre in zwei Sätzen, was eine SQLite-Transaktion ist.`
-3. `/privacy`: `allowed`, Provider `gemini`, kein Fallback.
-4. Ask ohne private Details: `Ich habe eine private Alltagssituation. Nenne nur
-   allgemeine Möglichkeiten.` Bei der Freigabe Enter drücken; Antwort muss lokal
-   über Ollama kommen.
-5. Never mit Kunstbeispiel: `Nur lokal: Der Beispiel-API-Key ist absichtlich
-   erfunden.` Es darf keine Cloudfreigabe angeboten werden; Antwort bleibt lokal.
-6. `/status`: bereinigte Zähler prüfen, danach `/exit`.
-
-Erst wenn Pascal diese Punkte bestätigt, darf über den Merge nach `main`
-entschieden werden. Google-Quoten sind projektabhängig und in AI Studio zu prüfen.
+Nur den Desktop-MVP nach `docs/DESKTOP_UI.md` manuell abnehmen. Erst danach mit
+Pascal ein einzelnes nächstes Paket festlegen; keine Folgephase automatisch beginnen.
 
 ## Weiterhin nicht beginnen
 
-Kein Obsidian-Zugriff, SSH-/Serverzugriff, Sprache, Mobile Client, Tools oder
-Living Mind. Keine echten persönlichen Daten für Cloudtests verwenden.
+Noch keine Obsidian-Integration, Tools, SSH-/Serversteuerung, Sprache, Mobile
+Client oder Living Mind. Der vorhandene Gemini-Key bleibt im Betriebssystem-
+Keyring und darf nicht erneut in Chat, Logs oder Dokumentation erscheinen.

docs/PROJECT_STATUS.md

@@ -1,6 +1,6 @@
 # Projektstatus
 
-Stand: Gemini-Free-Feature automatisiert geprüft, 30.07.2026.
+Stand: Desktop-MVP automatisiert und mit lokalem Modell geprüft, 30.07.2026.
 
 ## Erreicht
 
@@ -20,10 +20,15 @@ Stand: Gemini-Free-Feature automatisiert geprüft, 30.07.2026.
 - zwei echte deutsche Modellantworten erfolgreich vollständig auf der RTX 3060 erzeugt
 - kalter Test 36,19 s, warmer Kurztest 0,56 s; geladene GPU-Belegung 6.842 MiB
 - lokaler Ollama-Provider, zentrale Konfiguration und SQLite-Sitzungsspeicher implementiert
-- CLI mit `/new`, `/sessions`, `/load`, `/clear`, `/help` und `/exit` implementiert
+- CLI streamt Antworten und unterstützt `/new`, nummerierte `/sessions`,
+  `/load <NR|ID>`, `/search`, `/rename`, `/clear`, `/help` und `/exit`
+- bestehende SQLite-Datenbanken werden automatisch, transaktional und
+  wiederholbar um Sitzungstitel und zuletzt verwendeten Provider migriert
+- lokale Titelbildung ohne Modellaufruf; offensichtliche Secrets und medizinische
+  Inhalte werden nicht wörtlich in automatische Titel übernommen
 - echte CLI-Modellantwort sowie Laden und Fortsetzen derselben Sitzung in einem
   neuen Prozess bestätigt
-- 13 Unit-Tests, Ruff und Strukturprüfung erfolgreich
+- 101 Tests, Ruff, Struktur-, Gitignore-, Secret- und Auditprüfung erfolgreich
 - lokaler Core-Chat am 30.07.2026 zusätzlich von Pascal manuell erfolgreich
   geprüft: deutsche Antworten, Sitzungskontext, `/sessions`, `/exit`, Neustart,
   `/load <UUID>`, vollständiger Verlauf und Fortsetzung
@@ -33,10 +38,28 @@ Stand: Gemini-Free-Feature automatisiert geprüft, 30.07.2026.
 - Core-Chat automatisiert durch Codex getestet: ja
 - Core-Chat manuell durch Pascal getestet: ja, am 30.07.2026
 - Gemini/Datenschutz/Fallback automatisiert durch Codex getestet: ja
-- Gemini/Datenschutz/Fallback manuell durch Pascal getestet: noch nicht
-- von Pascal für produktive Verwendung freigegeben: noch nicht ausdrücklich erklärt
+- Gemini/Datenschutz/Fallback manuell durch Pascal getestet: ja, am 30.07.2026
+- für Merge freigegeben: ja, nach den zwei kleinen Korrekturen
+- für produktive Nutzung freigegeben: nicht separat erklärt; keine spätere Tool-,
+  Obsidian- oder Autonomiefunktion
 
-## Aktuelles Featurepaket
+## Chatkomfort-Meilenstein
+
+- Branch `feat/chat-comfort`
+- Ollama und Gemini verwenden denselben optionalen Streamingvertrag; Provider
+  ohne Streaming bleiben kompatibel
+- keine versteckten Gedanken; Terminal-Markdown wird lesbar bereinigt
+- vollständige Antwort wird genau einmal gespeichert; Strg+C und Streamfehler
+  hinterlassen keine unvollständige Nachricht
+- medizinische Fragen werden vor jeder Cloudclient-Erstellung zwingend `never`
+- Sitzungsliste zeigt Nummer, kurze ID, Titel, Datum, Nachrichtenanzahl und Provider
+- Laden per letzter Listennummer oder eindeutigem ID-Präfix; Mehrdeutigkeit wird erklärt
+- Suche ist auf 20 Ergebnisse begrenzt, parametrisiert und durchsucht nur Titel
+  sowie Benutzernachrichten
+- echter Ollama-CLI-Streaming-/Neustart-Smoke und echter kleiner
+  Gemini-Streaming-Smoke mit öffentlichem Inhalt erfolgreich
+
+## Gemini-/Datenschutz-Meilenstein
 
 - Branch `feat/gemini-privacy-router`
 - offizielle Gemini-Rahmenbedingungen geprüft; Standardmodell `gemini-3.6-flash`
@@ -60,9 +83,12 @@ Stand: Gemini-Free-Feature automatisiert geprüft, 30.07.2026.
 - `javis configure gemini` mit doppelter ausdrücklicher Bestätigung
 - echter Chat auf Hybridprovider verdrahtet; Key wird verzögert geladen
 - `/provider`, `/privacy` und `/status` geben keine Inhalte oder Keywerte aus
+- allgemeine unpersönliche Technikbegriffe werden gezielter `allowed`, während
+  persönliche, interne oder geheime technische Inhalte `ask`/`never` bleiben
+- `/privacy` trennt lokalen Modus von einem technischen Provider-Fallback
 - sicheres `scripts/start-javis.ps1` mit relativer Repositoryprüfung,
   Loopback-Ollama und überschreibbaren externen Pfaden
-- 62 Unit-Tests aktuell erfolgreich; PowerShell-Syntax, Lockfile, Sync,
+- 91 Unit-Tests aktuell erfolgreich; PowerShell-Syntax, Lockfile, Sync,
   Strukturprüfung, Secret-Scan, Gitignore, Ruff-Format und Lint sauber
 
 ## Nicht implementiert
@@ -70,24 +96,42 @@ Stand: Gemini-Free-Feature automatisiert geprüft, 30.07.2026.
 Obsidian-Adapter, Tools, Serverzugriff, Sprache, Mobile Client,
 Living-Mind-Frontend.
 
+## Desktop-MVP
+
+- PySide6-Client verwendet Runtime, Chatservice, Hybridprovider, Datenschutz,
+  SQLite, Audit und Secret-Provider des bestehenden Kerns
+- Sitzungen, Suche, Umbenennen, Markdown, Streaming, Abbruch, Modus,
+  Dreifach-Freigabedialog und bereinigter Status implementiert
+- `uv run javis gui` und `scripts/start-javis-gui.ps1` vorhanden
+- Qt-Offscreen-Start und echter lokaler Ollama-GUI-Smoke bestanden
+- automatisiert durch Codex getestet: ja
+- echter lokaler GUI-Smoke durch Codex: ja; 1 Sitzung, 2 Nachrichten, Ollama
+- manuell durch Pascal abgenommen: noch nicht
+
 ## Git
 
-- stabiler Branch: `main`
-- aktueller Branch: `feat/gemini-privacy-router`
-- Grundgerüstcommit: `6d04171`
-- Basis des Feature-Branches: `e6eb995`
-- Toolchain: `fa90a5c`
-- Laufzeit/Modell: `10d7cd7`
-- Chatkern: `ec9b2f9`
-- Tests: `b153daa`
-- Dokumentation: `5d567e5`
-- Merge nach `main`: `f3d6dfb`
+- aktueller Branch: `feat/desktop-ui`
+- Desktop-Basis: `main` bei `d3e85f2`
+- GUI-Grundgerüst: `1658e04`
+- Sitzungen/Streaming/Datenschutz: `cf35148`
+- Tests/Startskript/Dokumentation: `571969b`
+- `feat/desktop-ui` zu Gitea gepusht und synchron verifiziert
+- Basis: `main` bei `77b510b`
+- medizinisches Routing: `9148193`
+- Streaming: `00a5a6c`
+- Titelmigration: `112ae6d`
+- Sitzungsnavigation: `43c1f76`
+- Dokumentation: `328dcd9`
+- Feature-Branch zu `origin/feat/chat-comfort` gepusht
+- Merge nach `main`: `cbba102`; 91 Tests und Ruff auf `main` erneut bestanden
+- `main` bis `3361eae` zu `origin/main` gepusht und synchron verifiziert
+- vorheriger Gemini-/Datenschutz-Merge auf `main`: `564a3fd`
 
 ## Bekannte Einschränkung
 
-Ollama wird bewusst manuell in einem eigenen Terminal gestartet. Antworten werden
-noch nicht gestreamt; Sitzungen haben noch keine Titel, Suche oder Löschfunktion.
+Das Startskript startet Ollama bei Bedarf lokal. Eine Löschfunktion ist nicht
+Bestandteil dieses Pakets.
 `qwen3:8b` ist nicht automatisch für die GTX 1050 des späteren Zielhosts geeignet.
-Der echte Gemini-/Fallback-Test durch Pascal und danach die Mergeentscheidung
-fehlen. Es wurde kein echter API-Key angefordert, angezeigt oder konfiguriert und
-keine Cloudanfrage ausgeführt.
+Der Gemini-Key liegt ausschließlich im Betriebssystem-Keyring; sein Wert wurde
+von Codex weder gelesen noch ausgegeben. Pascal hat den echten Cloud- und
+Fallbackablauf erfolgreich geprüft.

javis-gui-crash.log

@@ -0,0 +1,13 @@
+Fatal Python error: Aborted
+
+Current thread 0x000066dc (most recent call first):
+  File "D:\Javis-Projekt\src\javis\ui\main_window.py", line 301 in _generation_complete
+  File "D:\Javis-Projekt\src\javis\ui\app.py", line 44 in run_gui
+  File "D:\Javis-Projekt\src\javis\interface\cli.py", line 429 in main
+  File "D:\Javis-Projekt\.venv\Scripts\javis.exe\__main__.py", line 10 in <module>
+  File "<frozen runpy>", line 88 in _run_code
+  File "<frozen runpy>", line 198 in _run_module_as_main
+
+Extension modules: charset_normalizer.md, charset_normalizer.cd, requests.packages.charset_normalizer.md, requests.packages.chardet.md, requests.packages.charset_normalizer.cd, requests.packages.chardet.cd, _cffi_backend, websockets.speedups, shiboken6.Shiboken, PySide6.QtCore, PySide6.QtGui, PySide6.QtWidgets (total: 12)
+QThreadStorage: entry 1 destroyed before end of thread 0x28a793bea70
+QThreadStorage: entry 0 destroyed before end of thread 0x28a793bea70

pyproject.toml

@@ -11,6 +11,7 @@ requires-python = ">=3.11"
 dependencies = [
     "google-genai>=2.6,<3",
     "keyring>=25.7,<26",
+    "pyside6>=6.8,<7",
 ]
 
 [project.scripts]
@@ -18,6 +19,7 @@ javis = "javis.interface.cli:main"
 
 [dependency-groups]
 dev = [
+    "pytest>=8,<9",
     "ruff>=0.12,<1",
 ]
 

scripts/start-javis-gui.ps1

@@ -0,0 +1,20 @@
+[CmdletBinding()]
+param(
+    [string]$UvPath,
+    [string]$OllamaPath,
+    [string]$OllamaModels,
+    [string]$DataDir,
+    [string]$UvCacheDir,
+    [string]$PythonInstallDir
+)
+
+$ErrorActionPreference = 'Stop'
+$env:JAVIS_START_INTERFACE = 'gui'
+
+$sharedStarter = Join-Path $PSScriptRoot 'start-javis.ps1'
+if (-not (Test-Path -LiteralPath $sharedStarter -PathType Leaf)) {
+    throw "Das gemeinsame sichere Javis-Startskript wurde nicht gefunden."
+}
+
+& $sharedStarter @PSBoundParameters
+exit $LASTEXITCODE

scripts/start-javis.ps1

@@ -113,7 +113,12 @@ if (-not (Test-OllamaReady)) {
 
 Push-Location $repositoryRoot
 try {
+    if ($env:JAVIS_START_INTERFACE -eq 'gui') {
+        & $resolvedUv run javis gui
+    }
+    else {
         & $resolvedUv run javis chat
+    }
     exit $LASTEXITCODE
 }
 finally {

src/javis/core/chat_service.py

@@ -2,16 +2,55 @@
 
 from __future__ import annotations
 
+import re
+from collections.abc import Iterator
 from dataclasses import dataclass
 
 from javis.memory.sqlite_store import ChatSession, SQLiteSessionStore
-from javis.providers.base import ChatMessage, LocalModelProvider
+from javis.providers.base import (
+    ChatMessage,
+    InvalidProviderResponseError,
+    LocalModelProvider,
+)
 
 
 class SessionProviderMismatchError(RuntimeError):
     """The active provider cannot safely continue the stored session."""
 
 
+_SENSITIVE_TITLE_PATTERN = re.compile(
+    r"(?ix)"
+    r"\b(?:api[-_ ]?key|passwort|password|token|secret|iban|kreditkarte|"
+    r"private[-_ ]?key|ssh[-_ ]?key)\b|"
+    r"-----BEGIN [A-Z ]+PRIVATE KEY-----|"
+    r"\b[A-Z]{2}\d{2}(?:[ ]?[A-Z0-9]){11,30}\b|"
+    r"\b[\w.+-]+@[\w.-]+\.[A-Z]{2,}\b"
+)
+_MEDICAL_TITLE_PATTERN = re.compile(
+    r"(?ix)\b(?:arzt|ärzt|blut(?:e|en|ung)?|diagnos|gesundheit|krank|"
+    r"medikament|notfall|schmerz|symptom|therap|verletz|wunde)\w*\b"
+)
+
+
+def derive_session_title(text: str, *, limit: int = 60) -> str:
+    """Create a short local title without retaining obvious sensitive values."""
+
+    normalized = " ".join(text.split())
+    if _SENSITIVE_TITLE_PATTERN.search(normalized):
+        return "Sensible Anfrage"
+    if _MEDICAL_TITLE_PATTERN.search(normalized):
+        return "Gesundheitsfrage"
+    cleaned = re.sub(r"[`*_#~]+", "", normalized).strip(" -:;,.!?")
+    if not cleaned:
+        return "Neue Unterhaltung"
+    if len(cleaned) <= limit:
+        return cleaned
+    shortened = cleaned[: limit - 1].rsplit(" ", 1)[0].rstrip(" -:;,.!?")
+    if not shortened:
+        shortened = cleaned[: limit - 1].rstrip()
+    return f"{shortened}…"
+
+
 @dataclass(frozen=True, slots=True)
 class LoadedSession:
     session: ChatSession
@@ -33,6 +72,24 @@ class ChatService:
     def list_sessions(self) -> list[ChatSession]:
         return self.store.list_sessions()
 
+    def search_sessions(self, text: str) -> list[ChatSession]:
+        normalized = text.strip()
+        if not normalized:
+            raise ValueError("Für die Suche fehlt ein Text.")
+        return self.store.search_sessions(normalized)
+
+    def resolve_session(self, reference: str) -> LoadedSession:
+        session = self.store.resolve_unique_prefix(reference)
+        return self.load_session(session.id)
+
+    def rename_session(self, session_id: str, title: str) -> ChatSession:
+        normalized = " ".join(title.split()).strip()
+        if not normalized:
+            raise ValueError("Der Sitzungstitel darf nicht leer sein.")
+        if len(normalized) > 60:
+            normalized = f"{normalized[:59].rstrip()}…"
+        return self.store.rename_session(session_id, normalized)
+
     def load_session(self, session_id: str) -> LoadedSession:
         session = self.store.get_session(session_id)
         supports_session = getattr(self.provider, "supports_session", None)
@@ -56,8 +113,63 @@ class ChatService:
         loaded = self.load_session(session_id)
         messages = [*loaded.messages, ChatMessage("user", normalized)]
         response = self.provider.chat(messages)
-        self.store.append_exchange(session_id, normalized, response)
+        self.store.append_exchange(
+            session_id,
+            normalized,
+            response,
+            title_if_first=derive_session_title(normalized),
+            last_provider=self._last_provider_name(),
+        )
         return response
 
+    def stream_send(self, session_id: str, text: str) -> Iterator[str]:
+        normalized = text.strip()
+        if not normalized:
+            raise ValueError("Eine leere Nachricht wird nicht gesendet.")
+        loaded = self.load_session(session_id)
+        messages = [*loaded.messages, ChatMessage("user", normalized)]
+        stream_method = getattr(self.provider, "stream_chat", None)
+        stream = (
+            stream_method(messages)
+            if callable(stream_method)
+            else iter((self.provider.chat(messages),))
+        )
+        chunks: list[str] = []
+        completed = False
+        try:
+            for chunk in stream:
+                if not isinstance(chunk, str):
+                    raise InvalidProviderResponseError(
+                        "Der Provider lieferte einen ungültigen Streaming-Abschnitt."
+                    )
+                if not chunk:
+                    continue
+                chunks.append(chunk)
+                yield chunk
+            completed = True
+        finally:
+            if not completed:
+                close_stream = getattr(stream, "close", None)
+                if callable(close_stream):
+                    close_stream()
+
+        response = "".join(chunks).strip()
+        if not response:
+            raise InvalidProviderResponseError(
+                "Der Provider lieferte keine verwendbare Streaming-Antwort."
+            )
+        self.store.append_exchange(
+            session_id,
+            normalized,
+            response,
+            title_if_first=derive_session_title(normalized),
+            last_provider=self._last_provider_name(),
+        )
+
     def clear_session(self, session_id: str) -> None:
         self.store.clear_messages(session_id)
+
+    def _last_provider_name(self) -> str:
+        route = getattr(self.provider, "last_route", None)
+        provider_name = getattr(route, "provider", None)
+        return provider_name if isinstance(provider_name, str) else self.provider.name

src/javis/core/provider_router.py

@@ -2,21 +2,32 @@
 
 from __future__ import annotations
 
-from collections.abc import Callable
+import threading
+from collections.abc import Callable, Iterator
 from dataclasses import dataclass
+from enum import StrEnum
 
 from javis.memory.usage_store import ProviderEvent, SQLiteUsageStore, UsageStoreError
 from javis.providers.base import (
     ChatMessage,
     InvalidApiKeyError,
+    InvalidProviderResponseError,
     LocalModelProvider,
     MissingApiKeyError,
     ProviderError,
     ProviderUsage,
+    ResponseAbortedError,
 )
 from javis.security.privacy import CloudPolicy, PrivacyDecision, PrivacyRouter
 
-ApprovalCallback = Callable[[PrivacyDecision], bool]
+
+class ApprovalChoice(StrEnum):
+    ALLOW = "allow"
+    LOCAL = "local"
+    CANCEL = "cancel"
+
+
+ApprovalCallback = Callable[[PrivacyDecision], ApprovalChoice | bool]
 NoticeCallback = Callable[[str], None]
 CloudProviderFactory = Callable[[], LocalModelProvider]
 
@@ -26,6 +37,8 @@ class RouteStatus:
     provider: str
     privacy_policy: CloudPolicy
     fallback: bool
+    cloud_suppressed_by_local_mode: bool
+    technical_fallback: bool
     reason: str
 
 
@@ -72,6 +85,8 @@ class HybridProvider:
         self._max_cloud_input_chars = max_cloud_input_chars
         self._max_cloud_context_messages = max_cloud_context_messages
         self._cloud_session_disabled = False
+        self._approval_lock = threading.Lock()
+        self._next_approval: ApprovalChoice | None = None
         self.last_route: RouteStatus | None = None
 
     def supports_session(self, provider: str, model: str) -> bool:
@@ -85,6 +100,33 @@ class HybridProvider:
             raise ValueError("Provider-Modus muss auto, local oder gemini sein.")
         self.mode = mode
 
+    def privacy_decision(self, text: str) -> PrivacyDecision:
+        """Classify a prospective request without contacting a provider."""
+        return self._privacy_router.classify(
+            text,
+            cloud_requested=self.mode == "gemini",
+        )
+
+    def queue_approval(self, choice: ApprovalChoice) -> None:
+        """Provide the GUI decision for exactly the next ASK request."""
+        if choice not in {ApprovalChoice.ALLOW, ApprovalChoice.LOCAL}:
+            raise ValueError("Nur Freigabe oder lokale Antwort kann vorgemerkt werden.")
+        with self._approval_lock:
+            self._next_approval = choice
+
+    def _approval_choice(self, decision: PrivacyDecision) -> ApprovalChoice:
+        with self._approval_lock:
+            queued = self._next_approval
+            self._next_approval = None
+        if queued is not None:
+            return queued
+        result = self._approval_callback(decision)
+        if isinstance(result, bool):
+            return ApprovalChoice.ALLOW if result else ApprovalChoice.LOCAL
+        if isinstance(result, ApprovalChoice):
+            return result
+        raise ValueError("Ungültiges Ergebnis der Cloudfreigabe.")
+
     def chat(self, messages: list[ChatMessage]) -> str:
         current_text = messages[-1].content if messages else ""
         cloud_requested = self.mode == "gemini"
@@ -94,14 +136,23 @@ class HybridProvider:
         )
 
         if self.mode == "local":
-            return self._local(messages, decision, "lokaler Modus", fallback=False)
+            return self._local(
+                messages,
+                decision,
+                "lokaler Modus",
+                fallback=False,
+                cloud_suppressed_by_local_mode=True,
+            )
         if decision.policy is CloudPolicy.NEVER:
             return self._local(messages, decision, decision.reason, fallback=True)
 
         approved = decision.policy is CloudPolicy.ALLOWED
         if decision.policy is CloudPolicy.ASK:
-            approved = self._approval_callback(decision)
-            if not approved:
+            choice = self._approval_choice(decision)
+            if choice is ApprovalChoice.CANCEL:
+                raise ResponseAbortedError("Anfrage vor der Cloudfreigabe abgebrochen.")
+            approved = choice is ApprovalChoice.ALLOW
+            if choice is ApprovalChoice.LOCAL:
                 return self._local(messages, decision, "Cloudfreigabe abgelehnt", fallback=True)
 
         unavailable_reason = self._cloud_unavailable_reason()
@@ -148,13 +199,257 @@ class HybridProvider:
             )
         )
         self.last_route = RouteStatus(
-            cloud_provider.name,
-            decision.policy,
-            False,
-            "Cloudaufruf erfolgreich",
+            provider=cloud_provider.name,
+            privacy_policy=decision.policy,
+            fallback=False,
+            cloud_suppressed_by_local_mode=False,
+            technical_fallback=False,
+            reason="Cloudaufruf erfolgreich",
         )
         return response
 
+    def stream_chat(self, messages: list[ChatMessage]) -> Iterator[str]:
+        current_text = messages[-1].content if messages else ""
+        decision = self._privacy_router.classify(
+            current_text,
+            cloud_requested=self.mode == "gemini",
+        )
+
+        if self.mode == "local":
+            yield from self._stream_local(
+                messages,
+                decision,
+                "lokaler Modus",
+                fallback=False,
+                cloud_suppressed_by_local_mode=True,
+            )
+            return
+        if decision.policy is CloudPolicy.NEVER:
+            yield from self._stream_local(
+                messages,
+                decision,
+                decision.reason,
+                fallback=True,
+            )
+            return
+
+        approved = decision.policy is CloudPolicy.ALLOWED
+        if decision.policy is CloudPolicy.ASK:
+            choice = self._approval_choice(decision)
+            if choice is ApprovalChoice.CANCEL:
+                raise ResponseAbortedError("Anfrage vor der Cloudfreigabe abgebrochen.")
+            approved = choice is ApprovalChoice.ALLOW
+            if choice is ApprovalChoice.LOCAL:
+                yield from self._stream_local(
+                    messages,
+                    decision,
+                    "Cloudfreigabe abgelehnt",
+                    fallback=True,
+                )
+                return
+
+        unavailable_reason = self._cloud_unavailable_reason()
+        if unavailable_reason:
+            yield from self._stream_local(
+                messages,
+                decision,
+                unavailable_reason,
+                fallback=True,
+            )
+            return
+        if len(current_text) > self._max_cloud_input_chars:
+            yield from self._stream_local(
+                messages,
+                decision,
+                "Nachricht überschreitet das lokale Cloud-Größenlimit",
+                fallback=True,
+            )
+            return
+
+        cloud_messages = self._privacy_router.minimal_context(
+            messages,
+            current_policy=decision.policy,
+            approved=approved,
+            max_chars=self._max_cloud_input_chars,
+            max_messages=self._max_cloud_context_messages,
+        )
+        if not cloud_messages:
+            yield from self._stream_local(
+                messages,
+                decision,
+                "kein freigegebener Cloudkontext",
+                fallback=True,
+            )
+            return
+
+        emitted = False
+        try:
+            cloud_provider = self._cloud_provider_factory()
+            for chunk in self._provider_chunks(cloud_provider, cloud_messages):
+                emitted = True
+                yield chunk
+        except InvalidApiKeyError as exc:
+            self._cloud_session_disabled = True
+            if emitted:
+                self._raise_interrupted_cloud(decision, exc)
+            yield from self._stream_cloud_failure(
+                messages,
+                decision,
+                exc,
+                disable_notice=True,
+            )
+            return
+        except (MissingApiKeyError, ProviderError) as exc:
+            if emitted:
+                self._raise_interrupted_cloud(decision, exc)
+            yield from self._stream_cloud_failure(messages, decision, exc)
+            return
+
+        if not emitted:
+            error = InvalidProviderResponseError(
+                "Gemini hat keine verwendbare Streaming-Antwort geliefert."
+            )
+            yield from self._stream_cloud_failure(messages, decision, error)
+            return
+
+        usage = getattr(cloud_provider, "last_usage", ProviderUsage())
+        self._record(
+            ProviderEvent(
+                provider=cloud_provider.name,
+                model=cloud_provider.model,
+                success=True,
+                error_category=None,
+                fallback=False,
+                privacy_policy=decision.policy.value,
+                input_tokens=usage.input_tokens,
+                output_tokens=usage.output_tokens,
+            )
+        )
+        self.last_route = RouteStatus(
+            provider=cloud_provider.name,
+            privacy_policy=decision.policy,
+            fallback=False,
+            cloud_suppressed_by_local_mode=False,
+            technical_fallback=False,
+            reason="Cloudstream erfolgreich",
+        )
+
+    @staticmethod
+    def _provider_chunks(
+        provider: LocalModelProvider,
+        messages: list[ChatMessage],
+    ) -> Iterator[str]:
+        stream_method = getattr(provider, "stream_chat", None)
+        chunks = (
+            stream_method(messages) if callable(stream_method) else iter((provider.chat(messages),))
+        )
+        completed = False
+        try:
+            for chunk in chunks:
+                if not isinstance(chunk, str):
+                    raise InvalidProviderResponseError(
+                        "Der Provider lieferte einen ungültigen Streaming-Abschnitt."
+                    )
+                if chunk:
+                    yield chunk
+            completed = True
+        finally:
+            if not completed:
+                close = getattr(chunks, "close", None)
+                if callable(close):
+                    close()
+
+    def _stream_cloud_failure(
+        self,
+        messages: list[ChatMessage],
+        decision: PrivacyDecision,
+        error: ProviderError,
+        *,
+        disable_notice: bool = False,
+    ) -> Iterator[str]:
+        self._record_cloud_error(decision, error)
+        reason = "Gemini nicht verfügbar"
+        if disable_notice:
+            reason = "Gemini-Schlüssel abgelehnt; Cloud für diese Sitzung deaktiviert"
+        yield from self._stream_local(
+            messages,
+            decision,
+            reason,
+            fallback=True,
+            technical_fallback=True,
+        )
+
+    def _raise_interrupted_cloud(
+        self,
+        decision: PrivacyDecision,
+        error: ProviderError,
+    ) -> None:
+        self._record_cloud_error(decision, error)
+        self.last_route = RouteStatus(
+            provider="gemini",
+            privacy_policy=decision.policy,
+            fallback=False,
+            cloud_suppressed_by_local_mode=False,
+            technical_fallback=True,
+            reason="Cloudstream nach Teilausgabe abgebrochen",
+        )
+        raise ResponseAbortedError(
+            "Der Gemini-Stream wurde nach einer Teilausgabe abgebrochen; "
+            "es wurde kein lokaler Ersatz angehängt."
+        ) from error
+
+    def _stream_local(
+        self,
+        messages: list[ChatMessage],
+        decision: PrivacyDecision,
+        reason: str,
+        *,
+        fallback: bool,
+        cloud_suppressed_by_local_mode: bool = False,
+        technical_fallback: bool = False,
+    ) -> Iterator[str]:
+        if fallback:
+            self._notice_callback(f"{reason} – lokale Antwort mit Ollama.")
+        emitted = False
+        try:
+            for chunk in self._provider_chunks(self.local_provider, messages):
+                emitted = True
+                yield chunk
+        except ProviderError as exc:
+            self._record(
+                ProviderEvent(
+                    provider=self.local_provider.name,
+                    model=self.local_provider.model,
+                    success=False,
+                    error_category=type(exc).__name__,
+                    fallback=fallback,
+                    privacy_policy=decision.policy.value,
+                )
+            )
+            raise
+        if not emitted:
+            raise InvalidProviderResponseError(
+                "Ollama hat keine verwendbare Streaming-Antwort geliefert."
+            )
+        self._record(
+            ProviderEvent(
+                provider=self.local_provider.name,
+                model=self.local_provider.model,
+                success=True,
+                error_category=None,
+                fallback=fallback,
+                privacy_policy=decision.policy.value,
+            )
+        )
+        self.last_route = RouteStatus(
+            provider=self.local_provider.name,
+            privacy_policy=decision.policy,
+            fallback=fallback,
+            cloud_suppressed_by_local_mode=cloud_suppressed_by_local_mode,
+            technical_fallback=technical_fallback,
+            reason=reason,
+        )
+
     def _cloud_unavailable_reason(self) -> str | None:
         if not self._cloud_enabled:
             return "Gemini ist lokal nicht aktiviert"
@@ -178,21 +473,33 @@ class HybridProvider:
         *,
         disable_notice: bool = False,
     ) -> str:
-        error_category = type(error).__name__
+        self._record_cloud_error(decision, error)
+        reason = "Gemini nicht verfügbar"
+        if disable_notice:
+            reason = "Gemini-Schlüssel abgelehnt; Cloud für diese Sitzung deaktiviert"
+        return self._local(
+            messages,
+            decision,
+            reason,
+            fallback=True,
+            technical_fallback=True,
+        )
+
+    def _record_cloud_error(
+        self,
+        decision: PrivacyDecision,
+        error: ProviderError,
+    ) -> None:
         self._record(
             ProviderEvent(
                 provider="gemini",
                 model=self._cloud_model,
                 success=False,
-                error_category=error_category,
+                error_category=type(error).__name__,
                 fallback=False,
                 privacy_policy=decision.policy.value,
             )
         )
-        reason = "Gemini nicht verfügbar"
-        if disable_notice:
-            reason = "Gemini-Schlüssel abgelehnt; Cloud für diese Sitzung deaktiviert"
-        return self._local(messages, decision, reason, fallback=True)
 
     def _local(
         self,
@@ -201,6 +508,8 @@ class HybridProvider:
         reason: str,
         *,
         fallback: bool,
+        cloud_suppressed_by_local_mode: bool = False,
+        technical_fallback: bool = False,
     ) -> str:
         if fallback:
             self._notice_callback(f"{reason} – lokale Antwort mit Ollama.")
@@ -216,10 +525,12 @@ class HybridProvider:
             )
         )
         self.last_route = RouteStatus(
-            self.local_provider.name,
-            decision.policy,
-            fallback,
-            reason,
+            provider=self.local_provider.name,
+            privacy_policy=decision.policy,
+            fallback=fallback,
+            cloud_suppressed_by_local_mode=cloud_suppressed_by_local_mode,
+            technical_fallback=technical_fallback,
+            reason=reason,
         )
         return response
 

src/javis/core/runtime.py

@@ -0,0 +1,145 @@
+"""Shared runtime assembly for CLI and desktop interfaces."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from javis.config.settings import Settings
+from javis.core.chat_service import ChatService
+from javis.core.provider_router import ApprovalCallback, HybridProvider, NoticeCallback
+from javis.memory.sqlite_store import SQLiteSessionStore
+from javis.memory.usage_store import SQLiteUsageStore, UsageStoreError
+from javis.providers.base import MissingApiKeyError
+from javis.providers.gemini import GeminiProvider
+from javis.providers.ollama import OllamaProvider
+from javis.security.privacy import PrivacyRouter
+from javis.security.secrets import SecretProvider, SecretStoreError
+
+
+@dataclass(frozen=True, slots=True)
+class RuntimeStatus:
+    ollama_reachable: bool
+    local_model_available: bool
+    local_model: str
+    gemini_key_available: bool
+    gemini_configured: bool
+    free_only: bool
+    provider_mode: str
+    cloud_requests_today: int | None
+    local_fallbacks_today: int | None
+    data_dir: str
+
+    def cli_lines(self) -> list[str]:
+        cloud_requests = (
+            self.cloud_requests_today if self.cloud_requests_today is not None else "unbekannt"
+        )
+        local_fallbacks = (
+            self.local_fallbacks_today if self.local_fallbacks_today is not None else "unbekannt"
+        )
+        return [
+            f"Ollama erreichbar: {'ja' if self.ollama_reachable else 'nein'}",
+            f"Lokales Modell vorhanden: {'ja' if self.local_model_available else 'nein'}",
+            f"Gemini-Schlüssel vorhanden: {'ja' if self.gemini_key_available else 'nein'}",
+            f"Gemini konfiguriert: {'ja' if self.gemini_configured else 'nein'}",
+            f"Nullkostenmodus: {'aktiv' if self.free_only else 'inaktiv'}",
+            f"Aktiver Providermodus: {self.provider_mode}",
+            f"Heutige Cloudanfragen: {cloud_requests}",
+            f"Heutige lokale Fallbacks: {local_fallbacks}",
+        ]
+
+
+@dataclass(slots=True)
+class JavisRuntime:
+    settings: Settings
+    service: ChatService
+    local_provider: OllamaProvider
+    hybrid_provider: HybridProvider
+    usage_store: SQLiteUsageStore
+    secrets: SecretProvider
+
+    def status(self) -> RuntimeStatus:
+        ollama = self.local_provider.probe()
+        try:
+            key_available = self.secrets.gemini_key_available()
+        except SecretStoreError:
+            key_available = False
+        try:
+            cloud_requests: int | None = self.usage_store.cloud_requests_on()
+            local_fallbacks: int | None = self.usage_store.local_fallbacks_on()
+        except UsageStoreError:
+            cloud_requests = local_fallbacks = None
+        return RuntimeStatus(
+            ollama_reachable=ollama.reachable,
+            local_model_available=ollama.model_available,
+            local_model=self.local_provider.model,
+            gemini_key_available=key_available,
+            gemini_configured=(
+                self.settings.gemini_enabled
+                and self.settings.free_only
+                and self.settings.billing_confirmed_disabled
+            ),
+            free_only=self.settings.free_only,
+            provider_mode=self.hybrid_provider.mode,
+            cloud_requests_today=cloud_requests,
+            local_fallbacks_today=local_fallbacks,
+            data_dir=str(self.settings.data_dir),
+        )
+
+
+def build_javis_runtime(
+    settings: Settings,
+    *,
+    approval_callback: ApprovalCallback,
+    notice_callback: NoticeCallback,
+    secret_provider: SecretProvider | None = None,
+) -> JavisRuntime:
+    """Assemble the existing core once for any user interface."""
+
+    store = SQLiteSessionStore(settings.database_path)
+    usage_store = SQLiteUsageStore(settings.usage_database_path)
+    secrets = secret_provider or SecretProvider()
+    local = OllamaProvider(
+        settings.model,
+        settings.ollama_base_url,
+        settings.timeout_seconds,
+    )
+
+    def create_cloud_provider() -> GeminiProvider:
+        try:
+            key = secrets.get_gemini_key()
+        except SecretStoreError as exc:
+            raise MissingApiKeyError(
+                "Der Gemini-Schlüssel konnte nicht sicher geladen werden."
+            ) from exc
+        return GeminiProvider(
+            model=settings.gemini_model,
+            api_key=key,
+            timeout_seconds=settings.gemini_timeout_seconds,
+            max_output_tokens=settings.max_cloud_output_tokens,
+            max_retries=settings.gemini_max_retries,
+        )
+
+    hybrid = HybridProvider(
+        local_provider=local,
+        cloud_provider_factory=create_cloud_provider,
+        cloud_model=settings.gemini_model,
+        privacy_router=PrivacyRouter(),
+        usage_store=usage_store,
+        approval_callback=approval_callback,
+        notice_callback=notice_callback,
+        mode=settings.provider_mode,
+        cloud_enabled=settings.gemini_enabled,
+        billing_confirmed_disabled=settings.billing_confirmed_disabled,
+        free_only=settings.free_only,
+        max_cloud_requests_per_day=settings.max_cloud_requests_per_day,
+        max_cloud_input_chars=settings.max_cloud_input_chars,
+        max_cloud_context_messages=settings.max_cloud_context_messages,
+    )
+    return JavisRuntime(
+        settings=settings,
+        service=ChatService(store, hybrid),
+        local_provider=local,
+        hybrid_provider=hybrid,
+        usage_store=usage_store,
+        secrets=secrets,
+    )

src/javis/interface/cli.py

@@ -4,6 +4,7 @@ from __future__ import annotations
 
 import argparse
 import getpass
+import re
 import sys
 from collections.abc import Callable, Sequence
 from pathlib import Path
@@ -11,28 +12,28 @@ from pathlib import Path
 from javis.config.local_config import LocalConfigError, write_gemini_activation
 from javis.config.settings import ConfigurationError, Settings
 from javis.core.chat_service import ChatService, SessionProviderMismatchError
-from javis.core.provider_router import HybridProvider
+from javis.core.runtime import build_javis_runtime
 from javis.memory.sqlite_store import (
+    AmbiguousSessionReferenceError,
     ChatSession,
     SessionNotFoundError,
     SessionStoreError,
-    SQLiteSessionStore,
 )
-from javis.memory.usage_store import SQLiteUsageStore, UsageStoreError
-from javis.providers.base import MissingApiKeyError, ProviderError
-from javis.providers.gemini import GeminiProvider
-from javis.providers.ollama import OllamaProvider
-from javis.security.privacy import PrivacyDecision, PrivacyRouter
+from javis.memory.usage_store import UsageStoreError
+from javis.providers.base import ProviderError
 from javis.security.secrets import SecretProvider, SecretStoreError
 
 InputFunction = Callable[[str], str]
 OutputFunction = Callable[[str], None]
+StreamOutputFunction = Callable[[str], None]
 StatusFunction = Callable[[], list[str]]
 
 HELP_TEXT = """Befehle:
   /new              neue Sitzung beginnen
   /sessions         gespeicherte Sitzungen auflisten
-  /load <ID>        frühere Sitzung laden und fortsetzen
+  /load <NR|ID>     Sitzung per Listennummer oder eindeutigem ID-Anfang laden
+  /search <TEXT>    Titel und eigene Nachrichten durchsuchen
+  /rename <TITEL>   aktive Sitzung umbenennen
   /clear            Nachrichten der aktiven Sitzung leeren
   /provider         aktiven Providermodus anzeigen
   /provider <MODUS> Modus auto, local oder gemini setzen
@@ -47,6 +48,7 @@ def build_parser() -> argparse.ArgumentParser:
     subparsers = parser.add_subparsers(dest="command", required=True)
     chat_parser = subparsers.add_parser("chat", help="lokalen Textchat starten")
     chat_parser.add_argument("--session", help="vorhandene Sitzungs-ID laden")
+    subparsers.add_parser("gui", help="Javis-Desktopoberfläche starten")
     configure_parser = subparsers.add_parser(
         "configure",
         help="nicht geheime lokale Einstellungen einrichten",
@@ -137,19 +139,56 @@ def _show_sessions(sessions: list[ChatSession], output: OutputFunction) -> None:
     if not sessions:
         output("Keine gespeicherten Sitzungen vorhanden.")
         return
-    for session in sessions:
+    for number, session in enumerate(sessions, start=1):
+        date = session.updated_at[:10]
         output(
-            f"{session.id} | {session.updated_at} | "
-            f"{session.provider}/{session.model} | {session.message_count} Nachrichten"
+            f"{number}. {session.id[:8]} | {session.title} | {date} | "
+            f"{session.message_count} Nachrichten | {session.last_provider}"
         )
 
 
+class TerminalMarkdownRenderer:
+    """Turn common Markdown markers into readable incremental terminal text."""
+
+    _tail_size = 3
+
+    def __init__(self) -> None:
+        self._pending = ""
+
+    @staticmethod
+    def _clean(text: str) -> str:
+        text = re.sub(r"(?m)^[ \t]{0,3}#{1,6}[ \t]+", "", text)
+        for marker in ("```", "**", "__", "~~", "`"):
+            text = text.replace(marker, "")
+        return text
+
+    def feed(self, chunk: str) -> str:
+        combined = self._pending + chunk
+        if len(combined) <= self._tail_size:
+            self._pending = combined
+            return ""
+        visible = combined[: -self._tail_size]
+        self._pending = combined[-self._tail_size :]
+        return self._clean(visible)
+
+    def finish(self) -> str:
+        visible = self._clean(self._pending)
+        self._pending = ""
+        return visible
+
+
+def _terminal_write(text: str) -> None:
+    sys.stdout.write(text)
+    sys.stdout.flush()
+
+
 def run_chat(
     service: ChatService,
     *,
     session_id: str | None = None,
     input_fn: InputFunction = input,
     output: OutputFunction = print,
+    stream_output: StreamOutputFunction | None = None,
     status_fn: StatusFunction | None = None,
 ) -> int:
     try:
@@ -169,6 +208,7 @@ def run_chat(
         return 2
 
     output("Lokaler Javis-Chat. /help zeigt die Befehle.")
+    shown_sessions: list[ChatSession] = []
 
     while True:
         try:
@@ -194,10 +234,33 @@ def run_chat(
             continue
         if entered == "/sessions":
             try:
-                _show_sessions(service.list_sessions(), output)
+                shown_sessions = service.list_sessions()
+                _show_sessions(shown_sessions, output)
             except SessionStoreError as exc:
                 output(f"Fehler: {exc}")
             continue
+        if entered.startswith("/search"):
+            parts = entered.split(maxsplit=1)
+            if len(parts) != 2 or not parts[1].strip():
+                output("Verwendung: /search <Text>")
+                continue
+            try:
+                shown_sessions = service.search_sessions(parts[1])
+                _show_sessions(shown_sessions, output)
+            except (SessionStoreError, ValueError) as exc:
+                output(f"Fehler: {exc}")
+            continue
+        if entered.startswith("/rename"):
+            parts = entered.split(maxsplit=1)
+            if len(parts) != 2 or not parts[1].strip():
+                output("Verwendung: /rename <neuer Titel>")
+                continue
+            try:
+                renamed = service.rename_session(active_id, parts[1])
+                output(f"Sitzung umbenannt: {renamed.title}")
+            except (SessionNotFoundError, SessionStoreError, ValueError) as exc:
+                output(f"Fehler: {exc}")
+            continue
         if entered == "/clear":
             try:
                 service.clear_session(active_id)
@@ -227,9 +290,11 @@ def run_chat(
                 output("Noch keine Datenschutzentscheidung in dieser Sitzung.")
             else:
                 output(
-                    "Letzte Datenschutzentscheidung: "
+                    "Datenschutzentscheidung: "
                     f"{route.privacy_policy.value}; Provider: {route.provider}; "
-                    f"Fallback: {'ja' if route.fallback else 'nein'}"
+                    "Cloud durch Modus local unterdrückt: "
+                    f"{'ja' if route.cloud_suppressed_by_local_mode else 'nein'}; "
+                    f"Technischer Fallback: {'ja' if route.technical_fallback else 'nein'}"
                 )
             continue
         if entered == "/status":
@@ -245,12 +310,24 @@ def run_chat(
                 output("Verwendung: /load <Sitzungs-ID>")
                 continue
             try:
-                loaded = service.load_session(parts[1].strip())
+                reference = parts[1].strip()
+                if reference.isdecimal():
+                    number = int(reference)
+                    if not shown_sessions:
+                        output("Zuerst /sessions oder /search anzeigen.")
+                        continue
+                    if number < 1 or number > len(shown_sessions):
+                        output("Diese Listennummer ist nicht vorhanden.")
+                        continue
+                    loaded = service.load_session(shown_sessions[number - 1].id)
+                else:
+                    loaded = service.resolve_session(reference)
                 active_id = loaded.session.id
                 output(f"Sitzung geladen: {active_id}")
                 _show_history(service, active_id, output)
             except (
                 SessionNotFoundError,
+                AmbiguousSessionReferenceError,
                 SessionProviderMismatchError,
                 SessionStoreError,
             ) as exc:
@@ -260,12 +337,41 @@ def run_chat(
             output("Unbekannter Befehl. /help zeigt die verfügbaren Befehle.")
             continue
 
+        stream = service.stream_send(active_id, entered)
+        writer = stream_output or (_terminal_write if output is print else output)
+        renderer = TerminalMarkdownRenderer()
+        started = False
         try:
-            response = service.send(active_id, entered)
-            output(f"Javis: {response}")
+            for chunk in stream:
+                visible = renderer.feed(chunk)
+                if not visible:
+                    continue
+                if not started:
+                    writer("Javis: ")
+                    started = True
+                writer(visible)
+            remaining = renderer.finish()
+            if remaining:
+                if not started:
+                    writer("Javis: ")
+                    started = True
+                writer(remaining)
+            if started:
+                writer("\n")
+        except KeyboardInterrupt:
+            close = getattr(stream, "close", None)
+            if callable(close):
+                close()
+            if started:
+                writer("\n")
+            output("Generierung abgebrochen. Die unvollständige Antwort wurde nicht gespeichert.")
         except (ProviderError, SessionStoreError, SessionProviderMismatchError) as exc:
+            if started:
+                writer("\n")
             output(f"Fehler: {exc}")
         except ValueError as exc:
+            if started:
+                writer("\n")
             output(f"Fehler: {exc}")
 
 
@@ -276,79 +382,21 @@ def build_runtime(
     output: OutputFunction,
     secret_provider: SecretProvider | None = None,
 ) -> tuple[ChatService, StatusFunction]:
-    store = SQLiteSessionStore(settings.database_path)
-    usage_store = SQLiteUsageStore(settings.usage_database_path)
-    secrets = secret_provider or SecretProvider()
-    local = OllamaProvider(
-        settings.model,
-        settings.ollama_base_url,
-        settings.timeout_seconds,
-    )
-
-    def approve(decision: PrivacyDecision) -> bool:
-        answer = input_fn(
-            f"Cloudfreigabe erforderlich ({decision.reason}). Einmalig erlauben? [j/N] "
-        )
+    def approve(_decision: object) -> bool:
+        answer = input_fn("Cloudfreigabe erforderlich. Einmalig erlauben? [j/N] ")
         return _confirmed(answer)
 
-    def create_cloud_provider() -> GeminiProvider:
-        try:
-            key = secrets.get_gemini_key()
-        except SecretStoreError as exc:
-            raise MissingApiKeyError(
-                "Der Gemini-Schlüssel konnte nicht sicher geladen werden."
-            ) from exc
-        return GeminiProvider(
-            model=settings.gemini_model,
-            api_key=key,
-            timeout_seconds=settings.gemini_timeout_seconds,
-            max_output_tokens=settings.max_cloud_output_tokens,
-            max_retries=settings.gemini_max_retries,
-        )
-
-    hybrid = HybridProvider(
-        local_provider=local,
-        cloud_provider_factory=create_cloud_provider,
-        cloud_model=settings.gemini_model,
-        privacy_router=PrivacyRouter(),
-        usage_store=usage_store,
+    runtime = build_javis_runtime(
+        settings,
         approval_callback=approve,
         notice_callback=output,
-        mode=settings.provider_mode,
-        cloud_enabled=settings.gemini_enabled,
-        billing_confirmed_disabled=settings.billing_confirmed_disabled,
-        free_only=settings.free_only,
-        max_cloud_requests_per_day=settings.max_cloud_requests_per_day,
-        max_cloud_input_chars=settings.max_cloud_input_chars,
-        max_cloud_context_messages=settings.max_cloud_context_messages,
+        secret_provider=secret_provider,
     )
 
     def status() -> list[str]:
-        ollama = local.probe()
-        try:
-            key_available = secrets.gemini_key_available()
-        except SecretStoreError:
-            key_available = False
-        gemini_configured = (
-            settings.gemini_enabled and settings.free_only and settings.billing_confirmed_disabled
-        )
-        try:
-            cloud_requests = usage_store.cloud_requests_on()
-            local_fallbacks = usage_store.local_fallbacks_on()
-        except UsageStoreError:
-            cloud_requests = local_fallbacks = -1
-        return [
-            f"Ollama erreichbar: {'ja' if ollama.reachable else 'nein'}",
-            f"Lokales Modell vorhanden: {'ja' if ollama.model_available else 'nein'}",
-            f"Gemini-Schlüssel vorhanden: {'ja' if key_available else 'nein'}",
-            f"Gemini konfiguriert: {'ja' if gemini_configured else 'nein'}",
-            f"Nullkostenmodus: {'aktiv' if settings.free_only else 'inaktiv'}",
-            f"Aktiver Providermodus: {hybrid.mode}",
-            f"Heutige Cloudanfragen: {cloud_requests if cloud_requests >= 0 else 'unbekannt'}",
-            f"Heutige lokale Fallbacks: {local_fallbacks if local_fallbacks >= 0 else 'unbekannt'}",
-        ]
+        return runtime.status().cli_lines()
 
-    return ChatService(store, hybrid), status
+    return runtime.service, status
 
 
 def main(
@@ -375,6 +423,10 @@ def main(
             input_fn=input_fn,
             output=output,
         )
+    if args.command == "gui":
+        from javis.ui.app import run_gui
+
+        return run_gui([])
     if args.command != "chat":
         return 2
 

src/javis/memory/sqlite_store.py

@@ -21,6 +21,10 @@ class SessionNotFoundError(SessionStoreError):
     """A requested session ID does not exist."""
 
 
+class AmbiguousSessionReferenceError(SessionStoreError):
+    """A short session reference matches multiple sessions."""
+
+
 @dataclass(frozen=True, slots=True)
 class ChatSession:
     id: str
@@ -28,9 +32,14 @@ class ChatSession:
     updated_at: str
     provider: str
     model: str
+    title: str
+    last_provider: str
     message_count: int = 0
 
 
+DEFAULT_SESSION_TITLE = "Neue Sitzung"
+
+
 def _utc_now() -> str:
     return datetime.now(UTC).isoformat(timespec="seconds")
 
@@ -63,16 +72,21 @@ class SQLiteSessionStore:
 
     def _initialize(self) -> None:
         with self._connection() as connection:
-            connection.executescript(
+            connection.execute(
                 """
                 CREATE TABLE IF NOT EXISTS sessions (
                     id TEXT PRIMARY KEY,
                     created_at TEXT NOT NULL,
                     updated_at TEXT NOT NULL,
                     provider TEXT NOT NULL,
-                    model TEXT NOT NULL
-                );
-
+                    model TEXT NOT NULL,
+                    title TEXT NOT NULL DEFAULT 'Neue Sitzung',
+                    last_provider TEXT NOT NULL DEFAULT ''
+                )
+                """
+            )
+            connection.execute(
+                """
                 CREATE TABLE IF NOT EXISTS messages (
                     id INTEGER PRIMARY KEY AUTOINCREMENT,
                     session_id TEXT NOT NULL,
@@ -82,10 +96,31 @@ class SQLiteSessionStore:
                     created_at TEXT NOT NULL,
                     FOREIGN KEY (session_id) REFERENCES sessions(id) ON DELETE CASCADE,
                     UNIQUE (session_id, position)
-                );
-
+                )
+                """
+            )
+            columns = {
+                row["name"] for row in connection.execute("PRAGMA table_info(sessions)").fetchall()
+            }
+            if "title" not in columns:
+                connection.execute(
+                    "ALTER TABLE sessions ADD COLUMN title TEXT NOT NULL DEFAULT 'Neue Sitzung'"
+                )
+            if "last_provider" not in columns:
+                connection.execute(
+                    "ALTER TABLE sessions ADD COLUMN last_provider TEXT NOT NULL DEFAULT ''"
+                )
+            connection.execute(
+                """
+                UPDATE sessions
+                SET last_provider = provider
+                WHERE last_provider = ''
+                """
+            )
+            connection.execute(
+                """
                 CREATE INDEX IF NOT EXISTS idx_messages_session_position
-                    ON messages(session_id, position);
+                    ON messages(session_id, position)
                 """
             )
 
@@ -96,14 +131,31 @@ class SQLiteSessionStore:
             with self._connection() as connection:
                 connection.execute(
                     """
-                    INSERT INTO sessions (id, created_at, updated_at, provider, model)
-                    VALUES (?, ?, ?, ?, ?)
+                    INSERT INTO sessions
+                        (id, created_at, updated_at, provider, model, title, last_provider)
+                    VALUES (?, ?, ?, ?, ?, ?, ?)
                     """,
-                    (session_id, now, now, provider, model),
+                    (
+                        session_id,
+                        now,
+                        now,
+                        provider,
+                        model,
+                        DEFAULT_SESSION_TITLE,
+                        provider,
+                    ),
                 )
         except sqlite3.Error as exc:
             raise SessionStoreError("Die Sitzung konnte nicht gespeichert werden.") from exc
-        return ChatSession(session_id, now, now, provider, model)
+        return ChatSession(
+            session_id,
+            now,
+            now,
+            provider,
+            model,
+            DEFAULT_SESSION_TITLE,
+            provider,
+        )
 
     def get_session(self, session_id: str) -> ChatSession:
         try:
@@ -140,6 +192,84 @@ class SQLiteSessionStore:
             raise SessionStoreError("Die Sitzungen konnten nicht aufgelistet werden.") from exc
         return [self._session_from_row(row) for row in rows]
 
+    def search_sessions(self, text: str, *, limit: int = 20) -> list[ChatSession]:
+        normalized = text.strip()
+        if not normalized:
+            return []
+        escaped = normalized.replace("\\", "\\\\").replace("%", "\\%").replace("_", "\\_")
+        pattern = f"%{escaped}%"
+        try:
+            with self._connection() as connection:
+                rows = connection.execute(
+                    """
+                    SELECT s.*, COUNT(m.id) AS message_count
+                    FROM sessions AS s
+                    LEFT JOIN messages AS m ON m.session_id = s.id
+                    WHERE s.title LIKE ? ESCAPE '\\' COLLATE NOCASE
+                       OR EXISTS (
+                            SELECT 1
+                            FROM messages AS searched
+                            WHERE searched.session_id = s.id
+                              AND searched.role = 'user'
+                              AND searched.content LIKE ? ESCAPE '\\' COLLATE NOCASE
+                       )
+                    GROUP BY s.id
+                    ORDER BY s.updated_at DESC, s.created_at DESC
+                    LIMIT ?
+                    """,
+                    (pattern, pattern, max(1, min(limit, 100))),
+                ).fetchall()
+        except sqlite3.Error as exc:
+            raise SessionStoreError("Die Sitzungssuche ist fehlgeschlagen.") from exc
+        return [self._session_from_row(row) for row in rows]
+
+    def resolve_unique_prefix(self, prefix: str) -> ChatSession:
+        normalized = prefix.strip()
+        if not normalized:
+            raise SessionNotFoundError("Leerer Sitzungsverweis.")
+        escaped = normalized.replace("\\", "\\\\").replace("%", "\\%").replace("_", "\\_")
+        try:
+            with self._connection() as connection:
+                rows = connection.execute(
+                    """
+                    SELECT id
+                    FROM sessions
+                    WHERE id LIKE ? ESCAPE '\\'
+                    ORDER BY id
+                    LIMIT 2
+                    """,
+                    (f"{escaped}%",),
+                ).fetchall()
+        except sqlite3.Error as exc:
+            raise SessionStoreError("Der Sitzungsverweis konnte nicht geprüft werden.") from exc
+        if not rows:
+            raise SessionNotFoundError(f"Keine Sitzung beginnt mit '{normalized}'.")
+        if len(rows) > 1:
+            raise AmbiguousSessionReferenceError(
+                f"Der Sitzungsanfang '{normalized}' ist nicht eindeutig."
+            )
+        return self.get_session(rows[0]["id"])
+
+    def rename_session(self, session_id: str, title: str) -> ChatSession:
+        now = _utc_now()
+        try:
+            with self._connection() as connection:
+                cursor = connection.execute(
+                    """
+                    UPDATE sessions
+                    SET title = ?, updated_at = ?
+                    WHERE id = ?
+                    """,
+                    (title, now, session_id),
+                )
+                if cursor.rowcount == 0:
+                    raise SessionNotFoundError(f"Sitzung '{session_id}' wurde nicht gefunden.")
+        except SessionNotFoundError:
+            raise
+        except sqlite3.Error as exc:
+            raise SessionStoreError("Die Sitzung konnte nicht umbenannt werden.") from exc
+        return self.get_session(session_id)
+
     def get_messages(self, session_id: str) -> list[ChatMessage]:
         self.get_session(session_id)
         try:
@@ -162,6 +292,9 @@ class SQLiteSessionStore:
         session_id: str,
         user_content: str,
         assistant_content: str,
+        *,
+        title_if_first: str | None = None,
+        last_provider: str | None = None,
     ) -> None:
         now = _utc_now()
         try:
@@ -194,8 +327,24 @@ class SQLiteSessionStore:
                     ),
                 )
                 connection.execute(
-                    "UPDATE sessions SET updated_at = ? WHERE id = ?",
-                    (now, session_id),
+                    """
+                    UPDATE sessions
+                    SET updated_at = ?,
+                        title = CASE
+                            WHEN ? = 0 AND title = ? THEN COALESCE(?, title)
+                            ELSE title
+                        END,
+                        last_provider = COALESCE(?, last_provider)
+                    WHERE id = ?
+                    """,
+                    (
+                        now,
+                        next_position,
+                        DEFAULT_SESSION_TITLE,
+                        title_if_first,
+                        last_provider,
+                        session_id,
+                    ),
                 )
         except SessionNotFoundError:
             raise
@@ -232,5 +381,7 @@ class SQLiteSessionStore:
             updated_at=row["updated_at"],
             provider=row["provider"],
             model=row["model"],
+            title=row["title"],
+            last_provider=row["last_provider"],
             message_count=row["message_count"],
         )

src/javis/providers/base.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+from collections.abc import Iterator
 from dataclasses import dataclass
 from typing import Protocol, runtime_checkable
 
@@ -70,3 +71,13 @@ class LocalModelProvider(Protocol):
     def chat(self, messages: list[ChatMessage]) -> str:
         """Return one complete assistant response."""
         ...
+
+
+@runtime_checkable
+class StreamingModelProvider(Protocol):
+    name: str
+    model: str
+
+    def stream_chat(self, messages: list[ChatMessage]) -> Iterator[str]:
+        """Yield only visible assistant text chunks."""
+        ...

src/javis/providers/gemini.py

@@ -2,7 +2,7 @@
 
 from __future__ import annotations
 
-from collections.abc import Callable
+from collections.abc import Callable, Iterator
 from typing import Any
 
 import httpx
@@ -49,6 +49,7 @@ class GeminiProvider:
         self._config = types.GenerateContentConfig(
             candidate_count=1,
             max_output_tokens=max_output_tokens,
+            thinking_config=types.ThinkingConfig(include_thoughts=False),
         )
         self._client = client_factory(
             api_key=api_key,
@@ -104,6 +105,73 @@ class GeminiProvider:
         )
         return text.strip()
 
+    def stream_chat(self, messages: list[ChatMessage]) -> Iterator[str]:
+        contents = [
+            types.Content(
+                role="model" if message.role == "assistant" else "user",
+                parts=[types.Part.from_text(text=message.content)],
+            )
+            for message in messages
+        ]
+        self.last_usage = ProviderUsage()
+        visible_text_received = False
+        try:
+            responses = self._client.models.generate_content_stream(
+                model=self.model,
+                contents=contents,
+                config=self._config,
+            )
+            for response in responses:
+                usage = getattr(response, "usage_metadata", None)
+                if usage is not None:
+                    self.last_usage = ProviderUsage(
+                        input_tokens=self._token_count(usage, "prompt_token_count"),
+                        output_tokens=self._token_count(
+                            usage,
+                            "candidates_token_count",
+                        ),
+                    )
+                for text in self._visible_text_parts(response):
+                    visible_text_received = True
+                    yield text
+        except errors.APIError as exc:
+            self._raise_api_error(exc)
+        except httpx.TimeoutException as exc:
+            raise ProviderTimeoutError("Gemini hat nicht rechtzeitig geantwortet.") from exc
+        except httpx.NetworkError as exc:
+            raise CloudNetworkError(
+                "Gemini ist wegen eines Netzwerkfehlers nicht erreichbar."
+            ) from exc
+        except (OSError, ConnectionError) as exc:
+            raise CloudNetworkError(
+                "Gemini ist wegen eines Netzwerkfehlers nicht erreichbar."
+            ) from exc
+
+        if not visible_text_received:
+            raise InvalidProviderResponseError(
+                "Gemini hat keine verwendbare Streaming-Antwort geliefert."
+            )
+
+    @staticmethod
+    def _visible_text_parts(response: object) -> Iterator[str]:
+        candidates = getattr(response, "candidates", None)
+        if isinstance(candidates, list) and candidates:
+            for candidate in candidates:
+                content = getattr(candidate, "content", None)
+                parts = getattr(content, "parts", None)
+                if not isinstance(parts, list):
+                    continue
+                for part in parts:
+                    if getattr(part, "thought", False):
+                        continue
+                    text = getattr(part, "text", None)
+                    if isinstance(text, str) and text:
+                        yield text
+            return
+        text = getattr(response, "text", None)
+        if isinstance(text, str) and text:
+            yield text
+
     @staticmethod
     def _token_count(usage: object, name: str) -> int | None:
         value = getattr(usage, name, None)

src/javis/providers/ollama.py

@@ -4,6 +4,7 @@ from __future__ import annotations
 
 import json
 import socket
+from collections.abc import Iterator
 from dataclasses import dataclass
 from urllib.error import HTTPError, URLError
 from urllib.request import Request, urlopen
@@ -107,3 +108,76 @@ class OllamaProvider:
                 "Ollama hat keine verwendbare Textantwort geliefert."
             )
         return content.strip()
+
+    def stream_chat(self, messages: list[ChatMessage]) -> Iterator[str]:
+        payload = {
+            "model": self.model,
+            "messages": [
+                {"role": message.role, "content": message.content} for message in messages
+            ],
+            "stream": True,
+            "think": False,
+        }
+        request = Request(
+            self._endpoint,
+            data=json.dumps(payload).encode("utf-8"),
+            headers={"Content-Type": "application/json"},
+            method="POST",
+        )
+        completed = False
+        visible_text_received = False
+        try:
+            with urlopen(request, timeout=self._timeout_seconds) as response:
+                for raw_line in response:
+                    if not raw_line.strip():
+                        continue
+                    try:
+                        result = json.loads(raw_line)
+                    except (json.JSONDecodeError, UnicodeDecodeError) as exc:
+                        raise InvalidProviderResponseError(
+                            "Ollama hat einen ungültigen Streaming-Abschnitt geliefert."
+                        ) from exc
+                    error = result.get("error")
+                    if isinstance(error, str) and error:
+                        if "not found" in error.lower():
+                            raise ModelNotInstalledError(
+                                f"Das lokale Modell '{self.model}' ist nicht installiert."
+                            )
+                        raise ProviderUnavailableError("Ollama hat den Stream abgelehnt.")
+                    content = result.get("message", {}).get("content")
+                    if content is not None and not isinstance(content, str):
+                        raise InvalidProviderResponseError(
+                            "Ollama hat ungültigen sichtbaren Text geliefert."
+                        )
+                    if content:
+                        visible_text_received = True
+                        yield content
+                    if result.get("done") is True:
+                        completed = True
+                        break
+        except HTTPError as exc:
+            details = exc.read().decode("utf-8", errors="replace")
+            if exc.code == 404 or "not found" in details.lower():
+                raise ModelNotInstalledError(
+                    f"Das lokale Modell '{self.model}' ist nicht installiert."
+                ) from exc
+            raise ProviderUnavailableError(f"Ollama meldet HTTP-Fehler {exc.code}.") from exc
+        except TimeoutError as exc:
+            raise ProviderTimeoutError(
+                "Die Modellantwort hat das Zeitlimit überschritten."
+            ) from exc
+        except URLError as exc:
+            if isinstance(exc.reason, (TimeoutError, socket.timeout)):
+                raise ProviderTimeoutError(
+                    "Die Modellantwort hat das Zeitlimit überschritten."
+                ) from exc
+            raise ProviderUnavailableError(
+                "Ollama ist unter der konfigurierten lokalen Adresse nicht erreichbar."
+            ) from exc
+
+        if not completed:
+            raise ResponseAbortedError("Der Ollama-Stream wurde vorzeitig beendet.")
+        if not visible_text_received:
+            raise InvalidProviderResponseError(
+                "Ollama hat keine verwendbare Streaming-Antwort geliefert."
+            )

src/javis/security/privacy.py

@@ -42,8 +42,13 @@ _NEVER_PATTERNS: tuple[tuple[re.Pattern[str], str], ...] = (
     ),
     (
         re.compile(
-            r"(?i)\b(diagnose|gesundheit|krankheit|medikament|patient|arztbericht|"
-            r"psychotherapie|blutwert)\b"
+            r"(?i)\b(symptom(?:e|en)?|blut(?:e|est|et|en|ung|ungen|wert|werte)?|"
+            r"verletz(?:ung|ungen|t)|wunde[n]?|medikament(?:e|en)?|diagnos(?:e|en|tiziert)|"
+            r"behandl(?:ung|ungen|e|en)|medizinisch(?:e[rsn]?|er)?\s+notfall|"
+            r"notaufnahme|rettungsdienst|gesundheit|krankheit(?:en)?|patient(?:en)?|"
+            r"arzt|ärzt(?:in|lich(?:e[rsn]?|er)?)|arztbericht|krankenhaus|"
+            r"psychotherapie|schmerz(?:en)?|fieber|atemnot|bewusstlos|ohnmacht|"
+            r"vergiftung|allergisch(?:e[rsn]?|er)?\s+reaktion)\b"
         ),
         "Gesundheitsinformation",
     ),
@@ -75,7 +80,8 @@ _ASK_PATTERNS: tuple[tuple[re.Pattern[str], str], ...] = (
     ),
     (
         re.compile(
-            r"(?i)\b(intern(?:e[rsn]?|er)? hostname|interne? domain|rootserver|"
+            r"(?i)\b(intern(?:e|er|es|en)?\s+"
+            r"(?:hostname|server|domain|infrastruktur)|rootserver|"
             r"privates? projekt|nicht öffentlich|systeminformation|"
             r"private[rs]? termin)\b"
         ),
@@ -90,7 +96,9 @@ _ASK_PATTERNS: tuple[tuple[re.Pattern[str], str], ...] = (
 _ALLOWED_PATTERNS: tuple[re.Pattern[str], ...] = (
     re.compile(
         r"(?i)\b(python|programmier|quellcode|algorithmus|sqlite|linux|windows|"
-        r"öffentliche dokumentation|allgemeine wissensfrage|was ist|wie funktioniert)\b"
+        r"git|ram|arbeitsspeicher|ssd|solid state drive|http|statuscode|"
+        r"transaktion|öffentliche dokumentation|allgemeine wissensfrage|"
+        r"was ist|wie funktioniert)\b"
     ),
 )
 

src/javis/ui/__init__.py

@@ -0,0 +1 @@
+"""PySide6 desktop interface for the existing Javis core."""

src/javis/ui/app.py

@@ -0,0 +1,53 @@
+"""Desktop application entry point."""
+
+from __future__ import annotations
+
+import os
+import sys
+from collections.abc import Sequence
+
+from PySide6.QtCore import QTimer
+from PySide6.QtWidgets import QApplication, QMessageBox
+
+from javis.config.settings import ConfigurationError, Settings
+from javis.core.provider_router import ApprovalChoice
+from javis.core.runtime import build_javis_runtime
+from javis.memory.sqlite_store import SessionStoreError
+from javis.memory.usage_store import UsageStoreError
+from javis.ui.chat_controller import DesktopController
+from javis.ui.main_window import MainWindow
+from javis.ui.workers import ApprovalBridge
+
+
+def run_gui(argv: Sequence[str] | None = None) -> int:
+    application = QApplication.instance() or QApplication(list(argv or []))
+    application.setApplicationName("Javis Desktop")
+    approval_bridge = ApprovalBridge()
+    try:
+        settings = Settings.from_env()
+        runtime = build_javis_runtime(
+            settings,
+            approval_callback=lambda _decision: ApprovalChoice.CANCEL,
+            notice_callback=approval_bridge.notice_received.emit,
+        )
+    except (ConfigurationError, SessionStoreError, UsageStoreError) as exc:
+        QMessageBox.critical(None, "Javis konnte nicht starten", str(exc))
+        return 2
+    window = MainWindow(DesktopController(runtime), approval_bridge)
+    window.show()
+    smoke_exit = os.environ.get("JAVIS_GUI_SMOKE_EXIT_MS")
+    if smoke_exit:
+        try:
+            delay = max(1, min(int(smoke_exit), 10_000))
+        except ValueError:
+            delay = 500
+        QTimer.singleShot(delay, application.quit)
+    return application.exec()
+
+
+def main() -> int:
+    return run_gui(sys.argv)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

src/javis/ui/chat_controller.py

@@ -0,0 +1,137 @@
+"""Small non-Qt adapter between desktop widgets and the existing Javis core."""
+
+from __future__ import annotations
+
+import re
+import threading
+from collections.abc import Iterator
+
+from javis.core.provider_router import ApprovalChoice
+from javis.core.runtime import JavisRuntime
+from javis.memory.sqlite_store import ChatSession
+from javis.providers.base import ProviderError, ResponseAbortedError
+from javis.security.privacy import CloudPolicy, PrivacyDecision
+from javis.ui.models import (
+    DesktopStatus,
+    GenerationState,
+    StreamEvent,
+    StreamEventKind,
+)
+
+_SECRET_PATTERN = re.compile(
+    r"AIza[\w-]{20,}|sk-[\w-]{16,}|-----BEGIN [A-Z ]*PRIVATE KEY-----",
+    re.IGNORECASE,
+)
+
+
+def sanitized_error(error: BaseException) -> str:
+    text = str(error).strip() or "Unbekannter Fehler."
+    return _SECRET_PATTERN.sub("[SECRET MASKIERT]", text)
+
+
+class DesktopController:
+    """Expose core operations without placing business rules in widgets."""
+
+    def __init__(self, runtime: JavisRuntime) -> None:
+        self.runtime = runtime
+        self.service = runtime.service
+        self.active_session_id: str | None = None
+        self.state = GenerationState.READY
+
+    def ensure_session(self) -> ChatSession:
+        if self.active_session_id is None:
+            session = self.service.new_session()
+            self.active_session_id = session.id
+            return session
+        return self.service.load_session(self.active_session_id).session
+
+    def new_session(self) -> ChatSession:
+        session = self.service.new_session()
+        self.active_session_id = session.id
+        return session
+
+    def list_sessions(self, search: str = "") -> list[ChatSession]:
+        return (
+            self.service.search_sessions(search) if search.strip() else self.service.list_sessions()
+        )
+
+    def load_session(self, session_id: str):
+        loaded = self.service.load_session(session_id)
+        self.active_session_id = loaded.session.id
+        return loaded
+
+    def rename_active_session(self, title: str) -> ChatSession:
+        session = self.ensure_session()
+        return self.service.rename_session(session.id, title)
+
+    def set_mode(self, mode: str) -> None:
+        self.runtime.hybrid_provider.set_mode(mode)
+
+    def approval_decision(self, text: str) -> PrivacyDecision | None:
+        provider = self.runtime.hybrid_provider
+        decision = provider.privacy_decision(text)
+        if provider.mode == "local" or decision.policy is not CloudPolicy.ASK:
+            return None
+        return decision
+
+    def queue_approval(self, choice: ApprovalChoice) -> None:
+        self.runtime.hybrid_provider.queue_approval(choice)
+
+    def stream_message(
+        self,
+        text: str,
+        cancel_event: threading.Event,
+    ) -> Iterator[StreamEvent]:
+        session = self.ensure_session()
+        stream = self.service.stream_send(session.id, text)
+        self.state = GenerationState.THINKING
+        yield StreamEvent(StreamEventKind.STATE, state=self.state)
+        started = False
+        try:
+            for chunk in stream:
+                if cancel_event.is_set():
+                    stream.close()
+                    self.state = GenerationState.ABORTED
+                    yield StreamEvent(StreamEventKind.STATE, state=self.state)
+                    return
+                if not started:
+                    started = True
+                    self.state = GenerationState.ANSWERING
+                    yield StreamEvent(StreamEventKind.STATE, state=self.state)
+                yield StreamEvent(StreamEventKind.CHUNK, text=chunk)
+        except ResponseAbortedError:
+            self.state = GenerationState.ABORTED
+            yield StreamEvent(StreamEventKind.STATE, state=self.state)
+            return
+        except ProviderError as exc:
+            if cancel_event.is_set():
+                self.state = GenerationState.ABORTED
+                yield StreamEvent(StreamEventKind.STATE, state=self.state)
+                return
+            self.state = GenerationState.ERROR
+            yield StreamEvent(StreamEventKind.ERROR, text=sanitized_error(exc), state=self.state)
+            return
+        except Exception as exc:
+            self.state = GenerationState.ERROR
+            yield StreamEvent(StreamEventKind.ERROR, text=sanitized_error(exc), state=self.state)
+            return
+        self.state = GenerationState.READY
+        yield StreamEvent(StreamEventKind.FINISHED, state=self.state)
+
+    def status(self) -> DesktopStatus:
+        runtime_status = self.runtime.status()
+        route = self.runtime.hybrid_provider.last_route
+        state = self.state
+        if state is GenerationState.READY and not runtime_status.ollama_reachable:
+            state = GenerationState.OFFLINE
+        return DesktopStatus(
+            state=state,
+            mode="Lokal" if runtime_status.provider_mode == "local" else "Hybrid/Auto",
+            provider=route.provider if route else "—",
+            ollama_reachable=runtime_status.ollama_reachable,
+            local_model=runtime_status.local_model,
+            privacy_policy=route.privacy_policy.value if route else "—",
+            cloud_requests_today=runtime_status.cloud_requests_today,
+            local_fallback=bool(route and route.fallback),
+            data_dir=runtime_status.data_dir,
+        )

src/javis/ui/main_window.py

@@ -0,0 +1,473 @@
+"""Main PySide6 window for the Javis desktop MVP."""
+
+from __future__ import annotations
+
+from html import escape
+
+from PySide6.QtCore import QSignalBlocker, Qt, QThread, QTimer, Signal
+from PySide6.QtGui import QAction, QCloseEvent, QKeySequence, QShortcut
+from PySide6.QtWidgets import (
+    QComboBox,
+    QFormLayout,
+    QHBoxLayout,
+    QLabel,
+    QLineEdit,
+    QListWidget,
+    QListWidgetItem,
+    QMainWindow,
+    QMenu,
+    QMessageBox,
+    QPushButton,
+    QSplitter,
+    QTextBrowser,
+    QTextEdit,
+    QVBoxLayout,
+    QWidget,
+)
+
+from javis.core.provider_router import ApprovalChoice
+from javis.providers.base import ChatMessage
+from javis.security.privacy import PrivacyDecision
+from javis.ui.chat_controller import DesktopController
+from javis.ui.models import GenerationState, StreamEvent, StreamEventKind
+from javis.ui.workers import ApprovalBridge, StreamWorker
+
+
+class MessageEdit(QTextEdit):
+    send_requested = Signal()
+
+    def keyPressEvent(self, event) -> None:
+        if event.key() in {Qt.Key_Return, Qt.Key_Enter} and not (
+            event.modifiers() & Qt.ShiftModifier
+        ):
+            self.send_requested.emit()
+            event.accept()
+            return
+        super().keyPressEvent(event)
+
+
+class MainWindow(QMainWindow):
+    def __init__(
+        self,
+        controller: DesktopController,
+        approval_bridge: ApprovalBridge,
+    ) -> None:
+        super().__init__()
+        self.controller = controller
+        self.approval_bridge = approval_bridge
+        self._thread: QThread | None = None
+        self._worker: StreamWorker | None = None
+        self._generation_finalized = True
+        self._close_when_finished = False
+        self._messages: list[ChatMessage] = []
+        self._draft_user = ""
+        self._draft_assistant = ""
+        self._thinking_dots = 0
+        self._build_ui()
+        self._connect_signals()
+        self._apply_style()
+        self._open_initial_session()
+        self._animation = QTimer(self)
+        self._animation.timeout.connect(self._animate_state)
+        self._animation.start(450)
+
+    def _build_ui(self) -> None:
+        self.setWindowTitle("Javis Desktop")
+        self.resize(1360, 820)
+        splitter = QSplitter(Qt.Horizontal)
+        splitter.addWidget(self._build_sidebar())
+        splitter.addWidget(self._build_chat())
+        splitter.addWidget(self._build_status_panel())
+        splitter.setSizes([280, 760, 300])
+        self.setCentralWidget(splitter)
+        self.statusBar().showMessage("Bereit")
+
+        QShortcut(QKeySequence("Ctrl+N"), self, activated=self._new_session)
+        QShortcut(QKeySequence("Ctrl+K"), self, activated=self._focus_search)
+        QShortcut(QKeySequence("Esc"), self, activated=self._stop_generation)
+
+    def _build_sidebar(self) -> QWidget:
+        panel = QWidget()
+        layout = QVBoxLayout(panel)
+        self.new_button = QPushButton("Neue Sitzung")
+        self.search_edit = QLineEdit()
+        self.search_edit.setPlaceholderText("Sitzungen suchen …")
+        self.session_list = QListWidget()
+        self.session_list.setContextMenuPolicy(Qt.CustomContextMenu)
+        layout.addWidget(self.new_button)
+        layout.addWidget(self.search_edit)
+        layout.addWidget(self.session_list, 1)
+        return panel
+
+    def _build_chat(self) -> QWidget:
+        panel = QWidget()
+        layout = QVBoxLayout(panel)
+        self.chat_view = QTextBrowser()
+        self.chat_view.setOpenExternalLinks(False)
+        self.message_edit = MessageEdit()
+        self.message_edit.setPlaceholderText("Nachricht an Javis …")
+        self.message_edit.setMaximumHeight(130)
+        buttons = QHBoxLayout()
+        buttons.addStretch(1)
+        self.stop_button = QPushButton("Stoppen")
+        self.stop_button.setEnabled(False)
+        self.send_button = QPushButton("Senden")
+        buttons.addWidget(self.stop_button)
+        buttons.addWidget(self.send_button)
+        layout.addWidget(self.chat_view, 1)
+        layout.addWidget(self.message_edit)
+        layout.addLayout(buttons)
+        return panel
+
+    def _build_status_panel(self) -> QWidget:
+        panel = QWidget()
+        layout = QVBoxLayout(panel)
+        title = QLabel("SYSTEMSTATUS")
+        title.setObjectName("panelTitle")
+        layout.addWidget(title)
+        self.state_label = QLabel("Bereit")
+        self.state_label.setObjectName("stateLabel")
+        layout.addWidget(self.state_label)
+        self.mode_combo = QComboBox()
+        self.mode_combo.addItem("Hybrid/Auto", "auto")
+        self.mode_combo.addItem("Lokal", "local")
+        form = QFormLayout()
+        form.addRow("Modus", self.mode_combo)
+        self.status_labels: dict[str, QLabel] = {}
+        for key, label in (
+            ("provider", "Provider"),
+            ("ollama", "Ollama erreichbar"),
+            ("model", "Lokales Modell"),
+            ("privacy", "Datenschutz"),
+            ("cloud", "Cloudaufrufe heute"),
+            ("fallback", "Lokales Fallback"),
+            ("data", "Datenverzeichnis"),
+        ):
+            value = QLabel("—")
+            value.setWordWrap(True)
+            self.status_labels[key] = value
+            form.addRow(label, value)
+        layout.addLayout(form)
+        layout.addStretch(1)
+        return panel
+
+    def _connect_signals(self) -> None:
+        self.new_button.clicked.connect(self._new_session)
+        self.search_edit.textChanged.connect(self._refresh_sessions)
+        self.session_list.itemDoubleClicked.connect(self._load_item)
+        self.session_list.customContextMenuRequested.connect(self._session_menu)
+        self.send_button.clicked.connect(self._send)
+        self.stop_button.clicked.connect(self._stop_generation)
+        self.message_edit.send_requested.connect(self._send)
+        self.mode_combo.currentIndexChanged.connect(self._mode_changed)
+        self.approval_bridge.notice_received.connect(self._show_notice)
+
+    def _apply_style(self) -> None:
+        self.setStyleSheet(
+            """
+            QMainWindow, QWidget { background: #171c21; color: #e8edf2; }
+            QLineEdit, QTextEdit, QTextBrowser, QListWidget, QComboBox {
+                background: #202b35; border: 1px solid #344553;
+                border-radius: 6px; padding: 7px; selection-background-color: #b97828;
+            }
+            QPushButton { background: #2b3945; border: 1px solid #485b69;
+                border-radius: 6px; padding: 8px 14px; }
+            QPushButton:hover { border-color: #d3943f; }
+            QPushButton:disabled { color: #68747d; }
+            #panelTitle { color: #d3943f; font-weight: 700; letter-spacing: 1px; }
+            #stateLabel { background: #24342e; color: #86d49b; padding: 10px;
+                border-radius: 6px; font-weight: 700; }
+            QStatusBar { background: #11161a; color: #b8c2ca; }
+            """
+        )
+
+    def _new_session(self) -> None:
+        if self._worker is not None:
+            return
+        session = self.controller.new_session()
+        self._messages = []
+        self._draft_user = self._draft_assistant = ""
+        self._render_chat()
+        self._refresh_sessions()
+        self.statusBar().showMessage(f"Neue Sitzung: {session.title}", 3000)
+        self.message_edit.setFocus()
+        self._refresh_status()
+
+    def _open_initial_session(self) -> None:
+        try:
+            sessions = self.controller.list_sessions()
+            if sessions:
+                loaded = self.controller.load_session(sessions[0].id)
+                self._messages = list(loaded.messages)
+                self._render_chat()
+                self._refresh_sessions()
+                self._refresh_status()
+                return
+        except Exception as exc:
+            self._show_error(str(exc))
+        self._new_session()
+
+    def _refresh_sessions(self) -> None:
+        try:
+            sessions = self.controller.list_sessions(self.search_edit.text())
+        except Exception as exc:
+            self._show_error(str(exc))
+            return
+        active = self.controller.active_session_id
+        self.session_list.clear()
+        for session in sessions:
+            item = QListWidgetItem(
+                f"{session.title}\n"
+                f"{session.last_provider} · {session.message_count} Nachrichten · "
+                f"{session.updated_at[:16].replace('T', ' ')}"
+            )
+            item.setData(Qt.UserRole, session.id)
+            self.session_list.addItem(item)
+            if session.id == active:
+                self.session_list.setCurrentItem(item)
+
+    def _load_item(self, item: QListWidgetItem) -> None:
+        if self._worker is not None:
+            return
+        try:
+            loaded = self.controller.load_session(item.data(Qt.UserRole))
+        except Exception as exc:
+            self._show_error(str(exc))
+            return
+        self._messages = list(loaded.messages)
+        self._draft_user = self._draft_assistant = ""
+        self._render_chat()
+        self.statusBar().showMessage(f"Sitzung geladen: {loaded.session.title}", 3000)
+
+    def _session_menu(self, position) -> None:
+        item = self.session_list.itemAt(position)
+        if item is None:
+            return
+        menu = QMenu(self)
+        rename_action = QAction("Umbenennen", menu)
+        rename_action.triggered.connect(lambda: self._rename_item(item))
+        menu.addAction(rename_action)
+        menu.exec(self.session_list.mapToGlobal(position))
+
+    def _rename_item(self, item: QListWidgetItem) -> None:
+        if item.data(Qt.UserRole) != self.controller.active_session_id:
+            self._load_item(item)
+        from PySide6.QtWidgets import QInputDialog
+
+        title, accepted = QInputDialog.getText(self, "Sitzung umbenennen", "Neuer Titel:")
+        if not accepted:
+            return
+        try:
+            self.controller.rename_active_session(title)
+            self._refresh_sessions()
+        except Exception as exc:
+            self._show_error(str(exc))
+
+    def _send(self) -> None:
+        text = self.message_edit.toPlainText().strip()
+        if not text or self._worker is not None:
+            return
+        try:
+            decision = self.controller.approval_decision(text)
+            if decision is not None:
+                choice = self._show_approval(decision)
+                if choice is ApprovalChoice.CANCEL:
+                    self.statusBar().showMessage("Anfrage abgebrochen.", 3000)
+                    return
+                self.controller.queue_approval(choice)
+        except Exception as exc:
+            self._show_error(str(exc))
+            return
+        self.message_edit.clear()
+        self._draft_user = text
+        self._draft_assistant = ""
+        self._render_chat()
+        self._set_generating(True)
+        self._generation_finalized = False
+        thread = QThread(self)
+        worker = StreamWorker(self.controller, text)
+        worker.moveToThread(thread)
+        thread.started.connect(worker.run)
+        worker.event_received.connect(self._handle_stream_event)
+        worker.completed.connect(thread.quit)
+        worker.completed.connect(worker.deleteLater)
+        worker.completed.connect(self._generation_complete)
+        thread.finished.connect(self._thread_finished)
+        thread.finished.connect(thread.deleteLater)
+        self._thread = thread
+        self._worker = worker
+        thread.start()
+
+    def _handle_stream_event(self, event: StreamEvent) -> None:
+        if self._generation_finalized:
+            return
+        if event.kind is StreamEventKind.CHUNK:
+            self._draft_assistant += event.text
+            self._render_chat()
+        if event.state is not None:
+            self._set_state(event.state)
+        if event.kind is StreamEventKind.ERROR:
+            self._show_error(event.text)
+
+    def _generation_complete(self) -> None:
+        if self._generation_finalized:
+            return
+        self._generation_finalized = True
+        failed = self.controller.state in {
+            GenerationState.ABORTED,
+            GenerationState.ERROR,
+        }
+        self._set_generating(False)
+        try:
+            loaded = self.controller.load_session(self.controller.active_session_id or "")
+            self._messages = list(loaded.messages)
+        except Exception as exc:
+            self._show_error(str(exc))
+        self._draft_user = self._draft_assistant = ""
+        self._render_chat()
+        self._refresh_sessions()
+        self._refresh_status()
+        if failed:
+            self.statusBar().showMessage("Antwort abgebrochen oder fehlgeschlagen.", 5000)
+        else:
+            self.statusBar().showMessage("Antwort vollständig gespeichert.", 3000)
+
+    def _thread_finished(self) -> None:
+        self._worker = None
+        self._thread = None
+        if self._close_when_finished:
+            self._close_when_finished = False
+            QTimer.singleShot(0, self.close)
+
+    def _stop_generation(self) -> None:
+        if self._worker is None:
+            return
+        self._worker.request_cancel()
+        self.statusBar().showMessage("Abbruch angefordert …")
+
+    def _set_generating(self, active: bool) -> None:
+        self.send_button.setEnabled(not active)
+        self.stop_button.setEnabled(active)
+        self.new_button.setEnabled(not active)
+        self.session_list.setEnabled(not active)
+        self.mode_combo.setEnabled(not active)
+
+    def _mode_changed(self) -> None:
+        try:
+            self.controller.set_mode(self.mode_combo.currentData())
+            self._refresh_status()
+        except Exception as exc:
+            self._show_error(str(exc))
+
+    def _show_approval(self, decision: PrivacyDecision) -> ApprovalChoice:
+        box = QMessageBox(self)
+        box.setWindowTitle("Cloudfreigabe")
+        box.setIcon(QMessageBox.Question)
+        box.setText(
+            "Diese Anfrage darf nur nach deiner Freigabe an Gemini gesendet werden.\n\n"
+            "Es wird ausschließlich der minimierte notwendige Kontext übertragen."
+        )
+        box.setInformativeText(f"Lokale Einstufung: {decision.reason}")
+        allow = box.addButton("Einmal erlauben", QMessageBox.AcceptRole)
+        local = box.addButton("Lokal beantworten", QMessageBox.RejectRole)
+        cancel = box.addButton("Abbrechen", QMessageBox.DestructiveRole)
+        box.exec()
+        clicked = box.clickedButton()
+        choice = ApprovalChoice.CANCEL
+        if clicked is allow:
+            choice = ApprovalChoice.ALLOW
+        elif clicked is local:
+            choice = ApprovalChoice.LOCAL
+        elif clicked is cancel:
+            choice = ApprovalChoice.CANCEL
+        if box.clickedButton() is local:
+            self.statusBar().showMessage("Die Anfrage wird lokal beantwortet.", 3000)
+        return choice
+
+    def _show_notice(self, text: str) -> None:
+        self.statusBar().showMessage(text, 5000)
+
+    def _show_error(self, text: str) -> None:
+        QMessageBox.critical(self, "Javis-Fehler", text)
+        self.statusBar().showMessage(f"Fehler: {text}", 5000)
+
+    def _refresh_status(self) -> None:
+        try:
+            status = self.controller.status()
+        except Exception as exc:
+            self._show_error(str(exc))
+            return
+        self._set_state(status.state)
+        target_mode = "local" if status.mode == "Lokal" else "auto"
+        mode_index = self.mode_combo.findData(target_mode)
+        if mode_index >= 0 and mode_index != self.mode_combo.currentIndex():
+            blocker = QSignalBlocker(self.mode_combo)
+            self.mode_combo.setCurrentIndex(mode_index)
+            del blocker
+        self.status_labels["provider"].setText(status.provider)
+        self.status_labels["ollama"].setText("Ja" if status.ollama_reachable else "Nein")
+        self.status_labels["model"].setText(status.local_model)
+        self.status_labels["privacy"].setText(status.privacy_policy)
+        self.status_labels["cloud"].setText(
+            str(status.cloud_requests_today)
+            if status.cloud_requests_today is not None
+            else "Unbekannt"
+        )
+        self.status_labels["fallback"].setText("Ja" if status.local_fallback else "Nein")
+        self.status_labels["data"].setText(status.data_dir)
+
+    def _set_state(self, state: GenerationState) -> None:
+        self.state_label.setText(state.value)
+        colors = {
+            GenerationState.READY: ("#24342e", "#86d49b"),
+            GenerationState.THINKING: ("#3a3020", "#e1a54f"),
+            GenerationState.ANSWERING: ("#3a3020", "#e1a54f"),
+            GenerationState.ABORTED: ("#3a3020", "#e1a54f"),
+            GenerationState.ERROR: ("#3a2424", "#ed7777"),
+            GenerationState.OFFLINE: ("#3a2424", "#ed7777"),
+        }
+        background, foreground = colors[state]
+        self.state_label.setStyleSheet(
+            f"background: {background}; color: {foreground}; padding: 10px; "
+            "border-radius: 6px; font-weight: 700;"
+        )
+
+    def _animate_state(self) -> None:
+        if self.controller.state not in {
+            GenerationState.THINKING,
+            GenerationState.ANSWERING,
+        }:
+            return
+        self._thinking_dots = (self._thinking_dots + 1) % 4
+        self.state_label.setText(f"{self.controller.state.value}{'.' * self._thinking_dots}")
+
+    def _render_chat(self) -> None:
+        blocks: list[str] = []
+        messages = list(self._messages)
+        if self._draft_user:
+            messages.append(ChatMessage("user", self._draft_user))
+        if self._draft_assistant:
+            messages.append(ChatMessage("assistant", self._draft_assistant))
+        for message in messages:
+            label = "Pascal" if message.role == "user" else "Javis"
+            content = message.content if message.role == "assistant" else escape(message.content)
+            blocks.append(f"### {label}\n\n{content}")
+        self.chat_view.setMarkdown("\n\n---\n\n".join(blocks))
+        scrollbar = self.chat_view.verticalScrollBar()
+        scrollbar.setValue(scrollbar.maximum())
+
+    def _focus_search(self) -> None:
+        self.search_edit.setFocus()
+        self.search_edit.selectAll()
+
+    def closeEvent(self, event: QCloseEvent) -> None:
+        if self._thread is not None:
+            self._close_when_finished = True
+            self._stop_generation()
+            event.ignore()
+            self.statusBar().showMessage(
+                "Javis beendet die laufende Antwort noch kontrolliert.",
+                3000,
+            )
+            return
+        event.accept()

src/javis/ui/models.py

@@ -0,0 +1,43 @@
+"""UI-facing immutable models without Qt dependencies."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from enum import StrEnum
+
+
+class GenerationState(StrEnum):
+    OFFLINE = "Offline"
+    READY = "Bereit"
+    THINKING = "Denkt"
+    ANSWERING = "Antwortet"
+    ABORTED = "Abgebrochen"
+    ERROR = "Fehler"
+
+
+class StreamEventKind(StrEnum):
+    STATE = "state"
+    CHUNK = "chunk"
+    NOTICE = "notice"
+    ERROR = "error"
+    FINISHED = "finished"
+
+
+@dataclass(frozen=True, slots=True)
+class StreamEvent:
+    kind: StreamEventKind
+    text: str = ""
+    state: GenerationState | None = None
+
+
+@dataclass(frozen=True, slots=True)
+class DesktopStatus:
+    state: GenerationState
+    mode: str
+    provider: str
+    ollama_reachable: bool
+    local_model: str
+    privacy_policy: str
+    cloud_requests_today: int | None
+    local_fallback: bool
+    data_dir: str

src/javis/ui/workers.py

@@ -0,0 +1,76 @@
+"""Qt worker objects for non-blocking model generation and approval."""
+
+from __future__ import annotations
+
+import threading
+
+from PySide6.QtCore import QObject, Signal, Slot
+
+from javis.core.provider_router import ApprovalChoice
+from javis.security.privacy import PrivacyDecision
+from javis.ui.chat_controller import DesktopController
+
+
+class ApprovalBridge(QObject):
+    approval_requested = Signal(object)
+    notice_received = Signal(str)
+
+    def __init__(self) -> None:
+        super().__init__()
+        self._lock = threading.Lock()
+        self._event: threading.Event | None = None
+        self._result = ApprovalChoice.CANCEL
+
+    def request(self, decision: PrivacyDecision) -> ApprovalChoice:
+        event = threading.Event()
+        with self._lock:
+            self._event = event
+            self._result = ApprovalChoice.CANCEL
+        self.approval_requested.emit(decision)
+        event.wait()
+        with self._lock:
+            result = self._result
+            self._event = None
+        return result
+
+    def resolve(self, choice: ApprovalChoice) -> None:
+        with self._lock:
+            self._result = choice
+            event = self._event
+        if event is not None:
+            event.set()
+
+    def cancel_pending(self) -> None:
+        self.resolve(ApprovalChoice.CANCEL)
+
+
+class StreamWorker(QObject):
+    event_received = Signal(object)
+    completed = Signal()
+
+    def __init__(self, controller: DesktopController, text: str) -> None:
+        super().__init__()
+        self._controller = controller
+        self._text = text
+        self._cancel_event = threading.Event()
+        self._started = False
+        self._completed = False
+
+    @Slot()
+    def run(self) -> None:
+        if self._started:
+            return
+        self._started = True
+        try:
+            for event in self._controller.stream_message(
+                self._text,
+                self._cancel_event,
+            ):
+                self.event_received.emit(event)
+        finally:
+            if not self._completed:
+                self._completed = True
+                self.completed.emit()
+
+    def request_cancel(self) -> None:
+        self._cancel_event.set()

tests/unit/test_chat_service.py

@@ -2,9 +2,9 @@ import tempfile
 import unittest
 from pathlib import Path
 
-from javis.core.chat_service import ChatService
+from javis.core.chat_service import ChatService, derive_session_title
 from javis.memory.sqlite_store import SQLiteSessionStore
-from javis.providers.base import ChatMessage, LocalModelProvider
+from javis.providers.base import ChatMessage, LocalModelProvider, ProviderUnavailableError
 
 
 class RecordingProvider:
@@ -19,6 +19,19 @@ class RecordingProvider:
         return f"Antwort {len(self.calls)}"
 
 
+class StreamingProvider(RecordingProvider):
+    def __init__(self, *, fail_after_first: bool = False) -> None:
+        super().__init__()
+        self.fail_after_first = fail_after_first
+
+    def stream_chat(self, messages: list[ChatMessage]):
+        self.calls.append(messages)
+        yield "Teil "
+        if self.fail_after_first:
+            raise ProviderUnavailableError("Streamfehler")
+        yield "Antwort"
+
+
 class ChatServiceTests(unittest.TestCase):
     def setUp(self) -> None:
         self.temporary_directory = tempfile.TemporaryDirectory()
@@ -51,6 +64,67 @@ class ChatServiceTests(unittest.TestCase):
                 "Antwort 2",
             ],
         )
+        self.assertEqual(loaded.session.title, "Hallo")
+
+    def test_title_is_local_short_and_redacts_obvious_sensitive_content(self) -> None:
+        self.assertEqual(
+            derive_session_title("Erkläre SQLite-Transaktionen für Anfänger"),
+            "Erkläre SQLite-Transaktionen für Anfänger",
+        )
+        self.assertLessEqual(derive_session_title("Wort " * 30).__len__(), 60)
+        self.assertEqual(
+            derive_session_title("Mein API-Key ist ABC123 und funktioniert nicht"),
+            "Sensible Anfrage",
+        )
+        self.assertEqual(
+            derive_session_title("Wenn ich blute, sollte ich zum Arzt?"),
+            "Gesundheitsfrage",
+        )
+
+    def test_manual_title_is_normalized_and_limited(self) -> None:
+        session = self.service.new_session()
+
+        renamed = self.service.rename_session(session.id, "  Mein   Titel  ")
+        shortened = self.service.rename_session(session.id, "x" * 80)
+
+        self.assertEqual(renamed.title, "Mein Titel")
+        self.assertEqual(len(shortened.title), 60)
+        self.assertTrue(shortened.title.endswith("…"))
+
+    def test_streaming_response_is_persisted_exactly_once_after_completion(self) -> None:
+        provider = StreamingProvider()
+        service = ChatService(self.service.store, provider)
+        session = service.new_session()
+
+        chunks = list(service.stream_send(session.id, "Hallo"))
+
+        self.assertEqual(chunks, ["Teil ", "Antwort"])
+        self.assertEqual(
+            [message.content for message in service.load_session(session.id).messages],
+            ["Hallo", "Teil Antwort"],
+        )
+        self.assertEqual(service.load_session(session.id).session.message_count, 2)
+
+    def test_closed_stream_does_not_persist_partial_response(self) -> None:
+        provider = StreamingProvider()
+        service = ChatService(self.service.store, provider)
+        session = service.new_session()
+        stream = service.stream_send(session.id, "Hallo")
+
+        self.assertEqual(next(stream), "Teil ")
+        stream.close()
+
+        self.assertEqual(service.load_session(session.id).messages, [])
+
+    def test_provider_error_during_stream_does_not_persist_partial_response(self) -> None:
+        provider = StreamingProvider(fail_after_first=True)
+        service = ChatService(self.service.store, provider)
+        session = service.new_session()
+
+        with self.assertRaises(ProviderUnavailableError):
+            list(service.stream_send(session.id, "Hallo"))
+
+        self.assertEqual(service.load_session(session.id).messages, [])
 
 
 if __name__ == "__main__":

tests/unit/test_cli.py

@@ -1,3 +1,4 @@
+import sqlite3
 import tempfile
 import unittest
 from pathlib import Path
@@ -6,7 +7,12 @@ from types import SimpleNamespace
 from javis.config.settings import Settings
 from javis.core.chat_service import ChatService
 from javis.core.provider_router import HybridProvider
-from javis.interface.cli import build_runtime, run_chat, run_gemini_configuration
+from javis.interface.cli import (
+    build_parser,
+    build_runtime,
+    run_chat,
+    run_gemini_configuration,
+)
 from javis.memory.sqlite_store import SQLiteSessionStore
 from javis.providers.base import ChatMessage
 from javis.security.privacy import CloudPolicy
@@ -19,6 +25,22 @@ class _FakeProvider:
     def chat(self, messages: list[ChatMessage]) -> str:
         return f"Echo: {messages[-1].content}"
 
+    def stream_chat(self, messages: list[ChatMessage]):
+        yield "Echo: "
+        yield messages[-1].content
+
+
+class _MarkdownProvider(_FakeProvider):
+    def stream_chat(self, messages: list[ChatMessage]):
+        yield "## **Ant"
+        yield "wort**\n`SQLite`"
+
+
+class _InterruptingProvider(_FakeProvider):
+    def stream_chat(self, messages: list[ChatMessage]):
+        yield "angefangene Antwort"
+        raise KeyboardInterrupt
+
 
 class _FakeHybridProvider(_FakeProvider):
     name = "hybrid"
@@ -45,6 +67,10 @@ class _LazySecrets:
 
 
 class CliTests(unittest.TestCase):
+    def test_parser_recognizes_gui_and_existing_chat(self) -> None:
+        self.assertEqual(build_parser().parse_args(["gui"]).command, "gui")
+        self.assertEqual(build_parser().parse_args(["chat"]).command, "chat")
+
     def test_basic_chat_commands(self) -> None:
         with tempfile.TemporaryDirectory() as directory:
             store = SQLiteSessionStore(Path(directory) / "sessions.sqlite3")
@@ -59,10 +85,53 @@ class CliTests(unittest.TestCase):
             )
 
         self.assertEqual(result, 0)
-        self.assertTrue(any(line == "Javis: Echo: Hallo" for line in output))
+        self.assertIn("Javis: Echo: Hallo\n", "".join(output))
         self.assertTrue(any("Nachrichten" in line for line in output))
         self.assertEqual(output[-1], "Chat beendet.")
 
+    def test_streaming_cleans_terminal_markdown_but_persists_raw_answer(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            store = SQLiteSessionStore(Path(directory) / "sessions.sqlite3")
+            service = ChatService(store, _MarkdownProvider())
+            inputs = iter(["Frage", "/exit"])
+            output: list[str] = []
+            streamed: list[str] = []
+
+            result = run_chat(
+                service,
+                input_fn=lambda _prompt: next(inputs),
+                output=output.append,
+                stream_output=streamed.append,
+            )
+            session = service.list_sessions()[0]
+            messages = service.load_session(session.id).messages
+
+        self.assertEqual(result, 0)
+        self.assertEqual("".join(streamed), "Javis: Antwort\nSQLite\n")
+        self.assertEqual(messages[-1].content, "## **Antwort**\n`SQLite`")
+
+    def test_keyboard_interrupt_drops_partial_answer_and_returns_to_prompt(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            store = SQLiteSessionStore(Path(directory) / "sessions.sqlite3")
+            service = ChatService(store, _InterruptingProvider())
+            inputs = iter(["Frage", "/exit"])
+            output: list[str] = []
+            streamed: list[str] = []
+
+            result = run_chat(
+                service,
+                input_fn=lambda _prompt: next(inputs),
+                output=output.append,
+                stream_output=streamed.append,
+            )
+            session = service.list_sessions()[0]
+            messages = service.load_session(session.id).messages
+
+        self.assertEqual(result, 0)
+        self.assertEqual(messages, [])
+        self.assertTrue(any("nicht gespeichert" in line for line in output))
+        self.assertEqual(output[-1], "Chat beendet.")
+
     def test_provider_privacy_and_status_commands_are_sanitized(self) -> None:
         with tempfile.TemporaryDirectory() as directory:
             store = SQLiteSessionStore(Path(directory) / "sessions.sqlite3")
@@ -70,7 +139,8 @@ class CliTests(unittest.TestCase):
             provider.last_route = SimpleNamespace(
                 privacy_policy=CloudPolicy.ASK,
                 provider="ollama",
-                fallback=True,
+                cloud_suppressed_by_local_mode=True,
+                technical_fallback=False,
             )
             service = ChatService(store, provider)
             inputs = iter(
@@ -97,9 +167,86 @@ class CliTests(unittest.TestCase):
         self.assertEqual(result, 0)
         self.assertIn("Aktiver Providermodus: auto", output)
         self.assertIn("Providermodus: local", output)
-        self.assertTrue(any("ask; Provider: ollama; Fallback: ja" in line for line in output))
+        self.assertTrue(
+            any(
+                "Datenschutzentscheidung: ask; Provider: ollama; "
+                "Cloud durch Modus local unterdrückt: ja; "
+                "Technischer Fallback: nein" in line
+                for line in output
+            )
+        )
         self.assertIn("Gemini-Schlüssel vorhanden: ja", output)
 
+    def test_numbered_sessions_load_rename_and_search(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            store = SQLiteSessionStore(Path(directory) / "sessions.sqlite3")
+            service = ChatService(store, _FakeProvider())
+            first = service.new_session()
+            service.send(first.id, "Erste Frage")
+            second = service.new_session()
+            service.send(second.id, "Zweite Frage")
+            shown_before = service.list_sessions()
+            target = shown_before[1]
+            inputs = iter(
+                [
+                    "/sessions",
+                    "/load 2",
+                    "/rename Gefundene Sitzung",
+                    "/search Gefundene",
+                    "/exit",
+                ]
+            )
+            output: list[str] = []
+
+            result = run_chat(
+                service,
+                session_id=first.id,
+                input_fn=lambda _prompt: next(inputs),
+                output=output.append,
+            )
+            renamed_title = store.get_session(target.id).title
+
+        self.assertEqual(result, 0)
+        self.assertTrue(any(line.startswith("1. ") and "Frage" in line for line in output))
+        self.assertIn(f"Sitzung geladen: {target.id}", output)
+        self.assertEqual(renamed_title, "Gefundene Sitzung")
+        self.assertTrue(any("Gefundene Sitzung" in line for line in output))
+
+    def test_load_accepts_unique_prefix_and_rejects_ambiguous_prefix(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            database = Path(directory) / "sessions.sqlite3"
+            store = SQLiteSessionStore(database)
+            service = ChatService(store, _FakeProvider())
+            active = service.new_session()
+            with sqlite3.connect(database) as connection:
+                connection.executemany(
+                    """
+                    INSERT INTO sessions
+                        (id, created_at, updated_at, provider, model, title, last_provider)
+                    VALUES (?, '2026-01-01', '2026-01-01',
+                            'ollama', 'test-model', ?, 'ollama')
+                    """,
+                    (
+                        ("shared-111", "Eins"),
+                        ("shared-222", "Zwei"),
+                        ("unique-333", "Drei"),
+                    ),
+                )
+            connection.close()
+            inputs = iter(["/load unique", "/load shared", "/exit"])
+            output: list[str] = []
+
+            result = run_chat(
+                service,
+                session_id=active.id,
+                input_fn=lambda _prompt: next(inputs),
+                output=output.append,
+            )
+
+        self.assertEqual(result, 0)
+        self.assertIn("Sitzung geladen: unique-333", output)
+        self.assertTrue(any("nicht eindeutig" in line for line in output))
+
     def test_gemini_configuration_defaults_to_no(self) -> None:
         with tempfile.TemporaryDirectory() as directory:
             path = Path(directory) / "javis.toml"

tests/unit/test_desktop_controller.py

@@ -0,0 +1,175 @@
+from __future__ import annotations
+
+import tempfile
+import threading
+import unittest
+from pathlib import Path
+from types import SimpleNamespace
+
+from javis.core.chat_service import ChatService
+from javis.core.provider_router import ApprovalChoice
+from javis.memory.sqlite_store import SQLiteSessionStore
+from javis.providers.base import ChatMessage, ResponseAbortedError
+from javis.security.privacy import CloudPolicy, PrivacyDecision
+from javis.ui.chat_controller import DesktopController, sanitized_error
+from javis.ui.models import GenerationState, StreamEventKind
+
+
+class _Service:
+    def __init__(self) -> None:
+        self.sessions = [
+            SimpleNamespace(id="one", title="Eins"),
+            SimpleNamespace(id="two", title="Zwei"),
+        ]
+
+    def new_session(self):
+        return self.sessions[0]
+
+    def list_sessions(self):
+        return self.sessions
+
+    def search_sessions(self, text: str):
+        return [session for session in self.sessions if text in session.title]
+
+    def load_session(self, session_id: str):
+        session = next(item for item in self.sessions if item.id == session_id)
+        return SimpleNamespace(
+            session=session,
+            messages=[ChatMessage("user", "Hallo")],
+        )
+
+    def rename_session(self, session_id: str, title: str):
+        session = next(item for item in self.sessions if item.id == session_id)
+        session.title = title
+        return session
+
+
+class _Hybrid:
+    mode = "auto"
+    last_route = None
+    queued_approval = None
+
+    def set_mode(self, mode: str) -> None:
+        self.mode = mode
+
+    def privacy_decision(self, _text: str) -> PrivacyDecision:
+        return PrivacyDecision(CloudPolicy.ASK, "Test")
+
+    def queue_approval(self, choice: ApprovalChoice) -> None:
+        self.queued_approval = choice
+
+
+class _StreamingProvider:
+    name = "ollama"
+    model = "test"
+    last_route = None
+    mode = "local"
+
+    def stream_chat(self, _messages):
+        yield "Teil "
+        yield "Antwort"
+
+    def set_mode(self, mode: str) -> None:
+        self.mode = mode
+
+
+class _AbortingProvider(_StreamingProvider):
+    def stream_chat(self, _messages):
+        raise ResponseAbortedError("Vom Dialog abgebrochen")
+        yield
+
+
+class DesktopControllerTests(unittest.TestCase):
+    def setUp(self) -> None:
+        self.runtime = SimpleNamespace(
+            service=_Service(),
+            hybrid_provider=_Hybrid(),
+        )
+        self.controller = DesktopController(self.runtime)
+
+    def test_controller_manages_sessions_without_qt_window(self) -> None:
+        created = self.controller.new_session()
+        listed = self.controller.list_sessions()
+        searched = self.controller.list_sessions("Zwei")
+        loaded = self.controller.load_session("two")
+        renamed = self.controller.rename_active_session("Neu")
+
+        self.assertEqual(created.id, "one")
+        self.assertEqual(len(listed), 2)
+        self.assertEqual([session.id for session in searched], ["two"])
+        self.assertEqual(loaded.messages[0].content, "Hallo")
+        self.assertEqual(renamed.title, "Neu")
+
+    def test_controller_preflights_and_queues_gui_approval(self) -> None:
+        decision = self.controller.approval_decision("Private Frage")
+        self.controller.queue_approval(ApprovalChoice.ALLOW)
+
+        self.assertIsNotNone(decision)
+        self.assertIs(decision.policy, CloudPolicy.ASK)
+        self.assertIs(
+            self.runtime.hybrid_provider.queued_approval,
+            ApprovalChoice.ALLOW,
+        )
+
+    def test_error_sanitizer_masks_key_patterns(self) -> None:
+        secret = "AIza" + "x" * 25
+        result = sanitized_error(RuntimeError(f"Fehler {secret}"))
+
+        self.assertNotIn(secret, result)
+        self.assertIn("MASKIERT", result)
+
+    def test_streaming_events_are_forwarded_and_completed(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            provider = _StreamingProvider()
+            service = ChatService(
+                SQLiteSessionStore(Path(directory) / "sessions.sqlite3"),
+                provider,
+            )
+            controller = DesktopController(
+                SimpleNamespace(service=service, hybrid_provider=provider)
+            )
+
+            events = list(controller.stream_message("Hallo", threading.Event()))
+
+            self.assertEqual(
+                [event.text for event in events if event.kind is StreamEventKind.CHUNK],
+                ["Teil ", "Antwort"],
+            )
+            self.assertEqual(events[-1].kind, StreamEventKind.FINISHED)
+            session = service.list_sessions()[0]
+            self.assertEqual(service.load_session(session.id).session.message_count, 2)
+
+    def test_cancel_drops_partial_exchange(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            provider = _StreamingProvider()
+            service = ChatService(
+                SQLiteSessionStore(Path(directory) / "sessions.sqlite3"),
+                provider,
+            )
+            controller = DesktopController(
+                SimpleNamespace(service=service, hybrid_provider=provider)
+            )
+            cancel = threading.Event()
+            cancel.set()
+
+            events = list(controller.stream_message("Hallo", cancel))
+
+            self.assertEqual(events[-1].state, GenerationState.ABORTED)
+            session = service.list_sessions()[0]
+            self.assertEqual(service.load_session(session.id).messages, [])
+
+    def test_provider_abort_is_reported_as_aborted_not_error(self) -> None:
+        with tempfile.TemporaryDirectory() as directory:
+            provider = _AbortingProvider()
+            service = ChatService(
+                SQLiteSessionStore(Path(directory) / "sessions.sqlite3"),
+                provider,
+            )
+            controller = DesktopController(
+                SimpleNamespace(service=service, hybrid_provider=provider)
+            )
+
+            events = list(controller.stream_message("Hallo", threading.Event()))
+
+            self.assertEqual(events[-1].state, GenerationState.ABORTED)
+            self.assertNotIn(StreamEventKind.ERROR, [event.kind for event in events])

tests/unit/test_desktop_gui.py

@@ -0,0 +1,282 @@
+from __future__ import annotations
+
+import os
+import threading
+import time
+import unittest
+from types import SimpleNamespace
+from unittest.mock import Mock, patch
+
+os.environ.setdefault("QT_QPA_PLATFORM", "offscreen")
+
+from PySide6.QtWidgets import QApplication
+
+from javis.core.provider_router import ApprovalChoice
+from javis.providers.base import ChatMessage
+from javis.security.privacy import CloudPolicy, PrivacyDecision
+from javis.ui.main_window import MainWindow
+from javis.ui.models import DesktopStatus, GenerationState, StreamEvent, StreamEventKind
+from javis.ui.workers import ApprovalBridge, StreamWorker
+
+
+class _Controller:
+    def __init__(self) -> None:
+        self.active_session_id = None
+        self.state = GenerationState.READY
+        self.session = SimpleNamespace(
+            id="session-one",
+            title="Desktop-Test",
+            last_provider="ollama",
+            message_count=2,
+            updated_at="2026-07-30T20:00:00+00:00",
+        )
+        self.messages = [
+            ChatMessage("user", "Hallo"),
+            ChatMessage("assistant", "Guten Tag"),
+        ]
+        self.approval: PrivacyDecision | None = None
+        self.queued_approval: ApprovalChoice | None = None
+        self.started = threading.Event()
+        self.release = threading.Event()
+        self.release.set()
+
+    def list_sessions(self, search: str = ""):
+        return [self.session] if not search or search in self.session.title else []
+
+    def load_session(self, session_id: str):
+        self.active_session_id = session_id
+        return SimpleNamespace(
+            session=self.session,
+            messages=list(self.messages),
+        )
+
+    def new_session(self):
+        self.active_session_id = self.session.id
+        return self.session
+
+    def rename_active_session(self, title: str):
+        self.session.title = title
+        return self.session
+
+    def set_mode(self, _mode: str) -> None:
+        return
+
+    def approval_decision(self, _text: str) -> PrivacyDecision | None:
+        return self.approval
+
+    def queue_approval(self, choice: ApprovalChoice) -> None:
+        self.queued_approval = choice
+
+    def stream_message(self, text: str, cancel_event: threading.Event):
+        self.started.set()
+        self.state = GenerationState.THINKING
+        yield StreamEvent(StreamEventKind.STATE, state=self.state)
+        while not self.release.wait(0.005):
+            if cancel_event.is_set():
+                self.state = GenerationState.ABORTED
+                yield StreamEvent(StreamEventKind.STATE, state=self.state)
+                return
+        if cancel_event.is_set():
+            self.state = GenerationState.ABORTED
+            yield StreamEvent(StreamEventKind.STATE, state=self.state)
+            return
+        self.state = GenerationState.ANSWERING
+        yield StreamEvent(StreamEventKind.STATE, state=self.state)
+        yield StreamEvent(StreamEventKind.CHUNK, text="Testantwort")
+        self.messages.extend(
+            [
+                ChatMessage("user", text),
+                ChatMessage("assistant", "Testantwort"),
+            ]
+        )
+        self.session.message_count = len(self.messages)
+        self.state = GenerationState.READY
+        yield StreamEvent(StreamEventKind.FINISHED, state=self.state)
+
+    def status(self) -> DesktopStatus:
+        return DesktopStatus(
+            state=self.state,
+            mode="Lokal",
+            provider="ollama",
+            ollama_reachable=True,
+            local_model="test-model",
+            privacy_policy="never",
+            cloud_requests_today=0,
+            local_fallback=False,
+            data_dir="D:\\Testdaten",
+        )
+
+
+class _MessageBox:
+    Question = object()
+    AcceptRole = object()
+    RejectRole = object()
+    DestructiveRole = object()
+    selected_label = ""
+
+    def __init__(self, _parent) -> None:
+        self.buttons: dict[str, object] = {}
+
+    def setWindowTitle(self, _text: str) -> None:
+        return
+
+    def setIcon(self, _icon: object) -> None:
+        return
+
+    def setText(self, _text: str) -> None:
+        return
+
+    def setInformativeText(self, _text: str) -> None:
+        return
+
+    def addButton(self, label: str, _role: object) -> object:
+        button = object()
+        self.buttons[label] = button
+        return button
+
+    def exec(self) -> None:
+        return
+
+    def clickedButton(self) -> object | None:
+        return self.buttons.get(type(self).selected_label)
+
+
+class DesktopGuiTests(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls) -> None:
+        cls.application = QApplication.instance() or QApplication([])
+
+    def setUp(self) -> None:
+        self.bridge = ApprovalBridge()
+        self.controller = _Controller()
+        self.window = MainWindow(self.controller, self.bridge)
+        self.window.show()
+        self.application.processEvents()
+
+    def tearDown(self) -> None:
+        self.controller.release.set()
+        self.window._stop_generation()
+        self._process_until(lambda: self.window._thread is None)
+        self.window.close()
+        self.application.processEvents()
+
+    def _process_until(self, condition, timeout: float = 2.0) -> None:
+        deadline = time.monotonic() + timeout
+        while not condition() and time.monotonic() < deadline:
+            self.application.processEvents()
+            time.sleep(0.005)
+        self.application.processEvents()
+        self.assertTrue(condition())
+
+    def test_offscreen_window_loads_session_and_core_widgets(self) -> None:
+        self.assertEqual(self.window.windowTitle(), "Javis Desktop")
+        self.assertEqual(self.window.session_list.count(), 1)
+        self.assertIn("Hallo", self.window.chat_view.toPlainText())
+        self.assertTrue(self.window.send_button.isEnabled())
+        self.assertFalse(self.window.stop_button.isEnabled())
+        self.assertEqual(self.window.status_labels["provider"].text(), "ollama")
+
+    def test_approval_dialog_maps_all_three_choices(self) -> None:
+        decision = PrivacyDecision(CloudPolicy.ASK, "persönliche Information")
+        self.bridge.resolve = Mock()
+        expected = {
+            "Einmal erlauben": ApprovalChoice.ALLOW,
+            "Lokal beantworten": ApprovalChoice.LOCAL,
+            "Abbrechen": ApprovalChoice.CANCEL,
+        }
+
+        with patch("javis.ui.main_window.QMessageBox", _MessageBox):
+            for label, choice in expected.items():
+                with self.subTest(label=label):
+                    _MessageBox.selected_label = label
+                    self.assertIs(self.window._show_approval(decision), choice)
+
+    def test_local_completion_and_second_question_keep_gui_reusable(self) -> None:
+        for text in ("Erste Frage", "Zweite Frage"):
+            self.window.message_edit.setPlainText(text)
+            self.window._send()
+            self._process_until(lambda: self.window._thread is None)
+
+        self.assertEqual(self.controller.session.message_count, 6)
+        self.assertIn("Zweite Frage", self.window.chat_view.toPlainText())
+        self.assertTrue(self.window.send_button.isEnabled())
+
+    def test_approval_is_decided_before_worker_starts(self) -> None:
+        self.controller.approval = PrivacyDecision(CloudPolicy.ASK, "privat")
+        self.window._show_approval = Mock(return_value=ApprovalChoice.ALLOW)
+
+        self.window.message_edit.setPlainText("Private Testfrage")
+        self.window._send()
+
+        self.assertIs(self.controller.queued_approval, ApprovalChoice.ALLOW)
+        self.window._show_approval.assert_called_once()
+        self._process_until(lambda: self.window._thread is None)
+
+    def test_local_approval_choice_is_queued_before_worker_starts(self) -> None:
+        self.controller.approval = PrivacyDecision(CloudPolicy.ASK, "privat")
+        self.window._show_approval = Mock(return_value=ApprovalChoice.LOCAL)
+
+        self.window.message_edit.setPlainText("Lokal beantworten")
+        self.window._send()
+
+        self.assertIs(self.controller.queued_approval, ApprovalChoice.LOCAL)
+        self._process_until(lambda: self.window._thread is None)
+
+    def test_stop_during_streaming_discards_partial_exchange(self) -> None:
+        self.controller.release.clear()
+        before = list(self.controller.messages)
+        self.window.message_edit.setPlainText("Nicht speichern")
+        self.window._send()
+        self._process_until(self.controller.started.is_set)
+
+        self.window._stop_generation()
+        self._process_until(lambda: self.window._thread is None)
+
+        self.assertEqual(self.controller.messages, before)
+        self.assertEqual(self.controller.state, GenerationState.ABORTED)
+
+    def test_stop_immediately_after_start_is_safe(self) -> None:
+        self.controller.release.clear()
+        self.window.message_edit.setPlainText("Sofort stoppen")
+        self.window._send()
+        self.window._stop_generation()
+
+        self._process_until(lambda: self.window._thread is None)
+        self.assertEqual(self.controller.session.message_count, 2)
+
+    def test_completion_is_idempotent_and_retains_qt_references(self) -> None:
+        worker = object()
+        thread = object()
+        self.window._worker = worker
+        self.window._thread = thread
+        self.window._generation_finalized = False
+
+        self.window._generation_complete()
+        self.window._generation_complete()
+
+        self.assertIs(self.window._worker, worker)
+        self.assertIs(self.window._thread, thread)
+        self.window._thread_finished()
+        self.assertIsNone(self.window._worker)
+        self.assertIsNone(self.window._thread)
+
+    def test_worker_emits_completion_only_once(self) -> None:
+        worker = StreamWorker(self.controller, "Nur einmal")
+        completions: list[bool] = []
+        worker.completed.connect(lambda: completions.append(True))
+
+        worker.run()
+        worker.run()
+
+        self.assertEqual(completions, [True])
+
+    def test_close_during_generation_cancels_then_closes(self) -> None:
+        self.controller.release.clear()
+        self.window.message_edit.setPlainText("Beim Schließen stoppen")
+        self.window._send()
+        self._process_until(self.controller.started.is_set)
+
+        self.window.close()
+        self.assertTrue(self.window._close_when_finished)
+        self._process_until(lambda: self.window._thread is None)
+        self._process_until(lambda: not self.window.isVisible())

tests/unit/test_gemini_provider.py

@@ -23,6 +23,7 @@ class _FakeModels:
     def __init__(self, result: object) -> None:
         self.result = result
         self.call: dict[str, Any] | None = None
+        self.stream_call: dict[str, Any] | None = None
 
     def generate_content(self, **kwargs: object) -> object:
         self.call = kwargs
@@ -30,6 +31,14 @@ class _FakeModels:
             raise self.result
         return self.result
 
+    def generate_content_stream(self, **kwargs: object):
+        self.stream_call = kwargs
+        if isinstance(self.result, BaseException):
+            raise self.result
+        if isinstance(self.result, list):
+            return iter(self.result)
+        return iter((self.result,))
+
 
 class _FakeClient:
     def __init__(self, result: object) -> None:
@@ -56,6 +65,23 @@ def _response(text: str = "Cloud-Antwort") -> SimpleNamespace:
     )
 
 
+def _stream_response(
+    visible_text: str,
+    *,
+    thought_text: str | None = None,
+    with_usage: bool = False,
+) -> SimpleNamespace:
+    parts = []
+    if thought_text:
+        parts.append(SimpleNamespace(text=thought_text, thought=True))
+    parts.append(SimpleNamespace(text=visible_text, thought=False))
+    usage = SimpleNamespace(prompt_token_count=12, candidates_token_count=7) if with_usage else None
+    return SimpleNamespace(
+        candidates=[SimpleNamespace(content=SimpleNamespace(parts=parts))],
+        usage_metadata=usage,
+    )
+
+
 class GeminiProviderTests(unittest.TestCase):
     def _provider(
         self, result: object, **overrides: object
@@ -99,6 +125,39 @@ class GeminiProviderTests(unittest.TestCase):
         self.assertEqual([content.role for content in call["contents"]], ["user", "model", "user"])
         self.assertEqual(call["config"].max_output_tokens, 1024)
         self.assertIsNone(call["config"].tools)
+        self.assertFalse(call["config"].thinking_config.include_thoughts)
+
+    def test_streaming_yields_only_visible_text_and_tracks_usage(self) -> None:
+        provider, factory = self._provider(
+            [
+                _stream_response("Cloud ", thought_text="verstecktes Denken"),
+                _stream_response("Antwort", with_usage=True),
+            ]
+        )
+
+        chunks = list(provider.stream_chat([ChatMessage("user", "Was ist SQLite?")]))
+
+        self.assertEqual(chunks, ["Cloud ", "Antwort"])
+        self.assertNotIn("verstecktes Denken", "".join(chunks))
+        self.assertEqual(provider.last_usage.input_tokens, 12)
+        self.assertEqual(provider.last_usage.output_tokens, 7)
+        self.assertEqual(
+            factory.client.models.stream_call["model"],
+            "gemini-3.6-flash",
+        )
+
+    def test_streaming_error_is_sanitized(self) -> None:
+        provider, _ = self._provider(
+            errors.ClientError(
+                429,
+                {"message": "quota test-key-not-a-real-secret"},
+            )
+        )
+
+        with self.assertRaises(ProviderRateLimitError) as raised:
+            list(provider.stream_chat([ChatMessage("user", "Was ist Python?")]))
+
+        self.assertNotIn("test-key-not-a-real-secret", str(raised.exception))
 
     def test_missing_key_stops_before_client_creation(self) -> None:
         factory = _ClientFactory(_response())

tests/unit/test_ollama_provider.py

@@ -4,16 +4,33 @@ import unittest
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from typing import ClassVar
 
-from javis.providers.base import ChatMessage
+from javis.providers.base import ChatMessage, ResponseAbortedError
 from javis.providers.ollama import OllamaProvider
 
 
 class _OllamaHandler(BaseHTTPRequestHandler):
     request_payload: ClassVar[dict[str, object]] = {}
+    abort_stream: ClassVar[bool] = False
 
     def do_POST(self) -> None:
         length = int(self.headers["Content-Length"])
         type(self).request_payload = json.loads(self.rfile.read(length))
+        if type(self).request_payload["stream"]:
+            chunks = [
+                {
+                    "message": {"role": "assistant", "content": "Lokale "},
+                    "done": False,
+                }
+            ]
+            if not type(self).abort_stream:
+                chunks.append(
+                    {
+                        "message": {"role": "assistant", "content": "Antwort"},
+                        "done": True,
+                    }
+                )
+            body = b"".join(json.dumps(chunk).encode() + b"\n" for chunk in chunks)
+        else:
             body = json.dumps(
                 {
                     "message": {"role": "assistant", "content": "Lokale Antwort"},
@@ -39,6 +56,9 @@ class _OllamaHandler(BaseHTTPRequestHandler):
 
 
 class OllamaProviderTests(unittest.TestCase):
+    def setUp(self) -> None:
+        _OllamaHandler.abort_stream = False
+
     def test_provider_uses_local_chat_endpoint(self) -> None:
         server = ThreadingHTTPServer(("127.0.0.1", 0), _OllamaHandler)
         thread = threading.Thread(target=server.serve_forever, daemon=True)
@@ -62,6 +82,44 @@ class OllamaProviderTests(unittest.TestCase):
         self.assertTrue(status.reachable)
         self.assertTrue(status.model_available)
 
+    def test_provider_streams_visible_text_without_thinking(self) -> None:
+        server = ThreadingHTTPServer(("127.0.0.1", 0), _OllamaHandler)
+        thread = threading.Thread(target=server.serve_forever, daemon=True)
+        thread.start()
+        try:
+            provider = OllamaProvider(
+                "test-model",
+                f"http://127.0.0.1:{server.server_port}",
+                2,
+            )
+            chunks = list(provider.stream_chat([ChatMessage("user", "Hallo")]))
+        finally:
+            server.shutdown()
+            server.server_close()
+            thread.join()
+
+        self.assertEqual(chunks, ["Lokale ", "Antwort"])
+        self.assertTrue(_OllamaHandler.request_payload["stream"])
+        self.assertFalse(_OllamaHandler.request_payload["think"])
+
+    def test_provider_rejects_incomplete_stream(self) -> None:
+        _OllamaHandler.abort_stream = True
+        server = ThreadingHTTPServer(("127.0.0.1", 0), _OllamaHandler)
+        thread = threading.Thread(target=server.serve_forever, daemon=True)
+        thread.start()
+        try:
+            provider = OllamaProvider(
+                "test-model",
+                f"http://127.0.0.1:{server.server_port}",
+                2,
+            )
+            with self.assertRaises(ResponseAbortedError):
+                list(provider.stream_chat([ChatMessage("user", "Hallo")]))
+        finally:
+            server.shutdown()
+            server.server_close()
+            thread.join()
+
 
 if __name__ == "__main__":
     unittest.main()

tests/unit/test_privacy.py

@@ -9,11 +9,34 @@ class PrivacyRouterTests(unittest.TestCase):
         self.router = PrivacyRouter()
 
     def test_general_technical_question_is_allowed(self) -> None:
+        for question in (
+            "Erkläre in zwei Sätzen den Unterschied zwischen RAM und SSD.",
+            "Wie funktioniert eine SQLite-Transaktion?",
+            "Was bedeutet HTTP 404?",
+            "Wie funktioniert Git?",
+        ):
+            with self.subTest(question=question):
                 self.assertEqual(
-            self.router.classify("Wie funktioniert SQLite in Python?").policy,
+                    self.router.classify(question).policy,
                     CloudPolicy.ALLOWED,
                 )
 
+    def test_internal_server_question_requires_approval(self) -> None:
+        decision = self.router.classify(
+            "Wie konfiguriere ich den internen Server mit Hostname srv-intern-01?"
+        )
+        self.assertEqual(decision.policy, CloudPolicy.ASK)
+
+    def test_technical_question_with_api_key_is_never(self) -> None:
+        decision = self.router.classify("Wie nutze ich diesen API-Key in Python: AIza" + "x" * 25)
+        self.assertEqual(decision.policy, CloudPolicy.NEVER)
+
+    def test_personal_technical_situation_requires_approval(self) -> None:
+        decision = self.router.classify(
+            "Meine SSD ist voll. Wie verschiebe ich meine persönlichen Dateien?"
+        )
+        self.assertEqual(decision.policy, CloudPolicy.ASK)
+
     def test_personal_information_requires_approval(self) -> None:
         self.assertEqual(
             self.router.classify("Ich möchte meinen Alltag besser planen.").policy,
@@ -40,6 +63,26 @@ class PrivacyRouterTests(unittest.TestCase):
             CloudPolicy.NEVER,
         )
 
+    def test_medical_questions_are_always_never(self) -> None:
+        for question in (
+            "Wenn ich blute, sollte ich dann zum Arzt?",
+            "Welche Symptome hat eine Grippe?",
+            "Wie nehme ich dieses Medikament ein?",
+            "Was bedeutet diese Diagnose?",
+            "Wie sollte eine Verletzung behandelt werden?",
+        ):
+            with self.subTest(question=question):
+                self.assertEqual(
+                    self.router.classify(question).policy,
+                    CloudPolicy.NEVER,
+                )
+
+    def test_technical_terms_cannot_override_medical_never(self) -> None:
+        decision = self.router.classify(
+            "Wie speichere ich meine Symptome und Diagnose mit Python in SQLite?"
+        )
+        self.assertEqual(decision.policy, CloudPolicy.NEVER)
+
     def test_unknown_content_is_conservatively_ask(self) -> None:
         self.assertEqual(
             self.router.classify("Erkläre das bitte genauer.").policy,

tests/unit/test_provider_router.py

@@ -4,7 +4,7 @@ import tempfile
 import unittest
 from pathlib import Path
 
-from javis.core.provider_router import HybridProvider
+from javis.core.provider_router import ApprovalChoice, HybridProvider
 from javis.memory.usage_store import ProviderEvent, SQLiteUsageStore
 from javis.providers.base import (
     ChatMessage,
@@ -16,6 +16,7 @@ from javis.providers.base import (
     ProviderRateLimitError,
     ProviderTimeoutError,
     ProviderUsage,
+    ResponseAbortedError,
 )
 from javis.security.privacy import CloudPolicy, PrivacyRouter
 
@@ -28,11 +29,15 @@ class _RecordingProvider:
         *,
         answer: str = "Antwort",
         error: Exception | None = None,
+        stream_chunks: tuple[str, ...] | None = None,
+        stream_error: Exception | None = None,
     ) -> None:
         self.name = name
         self.model = model
         self.answer = answer
         self.error = error
+        self.stream_chunks = stream_chunks
+        self.stream_error = stream_error
         self.calls: list[list[ChatMessage]] = []
         self.last_usage = ProviderUsage(11, 5)
 
@@ -42,6 +47,17 @@ class _RecordingProvider:
             raise self.error
         return self.answer
 
+    def stream_chat(self, messages: list[ChatMessage]):
+        self.calls.append(messages)
+        if self.stream_chunks is None:
+            if self.error:
+                raise self.error
+            yield self.answer
+        else:
+            yield from self.stream_chunks
+        if self.stream_error:
+            raise self.stream_error
+
 
 class HybridProviderTests(unittest.TestCase):
     def setUp(self) -> None:
@@ -131,6 +147,23 @@ class HybridProviderTests(unittest.TestCase):
         self.assertTrue(router.last_route.fallback)
         self.assertIn("lokale Antwort", self.notices[-1])
 
+    def test_medical_never_routes_locally_before_cloud_construction(self) -> None:
+        constructed = False
+
+        def cloud_factory() -> _RecordingProvider:
+            nonlocal constructed
+            constructed = True
+            return self.cloud
+
+        router = self._router(cloud_provider_factory=cloud_factory, mode="gemini")
+
+        answer = router.chat([ChatMessage("user", "Wenn ich blute, sollte ich dann zum Arzt?")])
+
+        self.assertEqual(answer, "Lokal")
+        self.assertFalse(constructed)
+        self.assertFalse(self.approvals)
+        self.assertEqual(router.last_route.privacy_policy, CloudPolicy.NEVER)
+
     def test_ask_defaults_to_local_when_approval_is_denied(self) -> None:
         router = self._router()
 
@@ -149,6 +182,31 @@ class HybridProviderTests(unittest.TestCase):
         self.assertEqual(answer, "Cloud")
         self.assertEqual(len(self.cloud.calls), 1)
 
+    def test_queued_gui_approval_is_one_shot_and_uses_fake_cloud(self) -> None:
+        router = self._router()
+        question = [ChatMessage("user", "Meine Familie plant Urlaub")]
+        router.queue_approval(ApprovalChoice.ALLOW)
+
+        chunks = list(router.stream_chat(question))
+        second_answer = router.chat(question)
+
+        self.assertEqual(chunks, ["Cloud"])
+        self.assertEqual(second_answer, "Lokal")
+        self.assertEqual(len(self.approvals), 1)
+
+    def test_ask_supports_allow_local_and_cancel_choices(self) -> None:
+        question = [ChatMessage("user", "Meine Familie plant Urlaub")]
+
+        allowed = self._router(approval_callback=lambda _decision: ApprovalChoice.ALLOW)
+        self.assertEqual(allowed.chat(question), "Cloud")
+
+        local = self._router(approval_callback=lambda _decision: ApprovalChoice.LOCAL)
+        self.assertEqual(local.chat(question), "Lokal")
+
+        cancelled = self._router(approval_callback=lambda _decision: ApprovalChoice.CANCEL)
+        with self.assertRaises(ResponseAbortedError):
+            cancelled.chat(question)
+
     def test_429_falls_back_without_losing_local_answer(self) -> None:
         self.cloud.error = ProviderRateLimitError("quota")
         router = self._router()
@@ -253,6 +311,76 @@ class HybridProviderTests(unittest.TestCase):
         self.assertEqual(answer, "Lokal")
         self.assertFalse(self.approvals)
         self.assertFalse(self.cloud.calls)
+        self.assertFalse(router.last_route.fallback)
+        self.assertTrue(router.last_route.cloud_suppressed_by_local_mode)
+        self.assertFalse(router.last_route.technical_fallback)
+
+    def test_network_failure_is_reported_as_technical_fallback(self) -> None:
+        self.cloud.error = CloudNetworkError("offline")
+        router = self._router()
+
+        router.chat([ChatMessage("user", "Wie funktioniert Python?")])
+
+        self.assertFalse(router.last_route.cloud_suppressed_by_local_mode)
+        self.assertTrue(router.last_route.technical_fallback)
+
+    def test_allowed_cloud_response_streams_visible_chunks(self) -> None:
+        self.cloud.stream_chunks = ("Cloud ", "Stream")
+        router = self._router()
+
+        chunks = list(router.stream_chat([ChatMessage("user", "Wie funktioniert SQLite?")]))
+
+        self.assertEqual(chunks, ["Cloud ", "Stream"])
+        self.assertEqual(router.last_route.provider, "gemini")
+        self.assertFalse(router.last_route.fallback)
+
+    def test_cloud_stream_error_before_output_falls_back_locally(self) -> None:
+        self.cloud.stream_chunks = ()
+        self.cloud.stream_error = CloudNetworkError("offline")
+        self.local.stream_chunks = ("Lokaler ", "Ersatz")
+        router = self._router()
+
+        chunks = list(router.stream_chat([ChatMessage("user", "Wie funktioniert Python?")]))
+
+        self.assertEqual(chunks, ["Lokaler ", "Ersatz"])
+        self.assertTrue(router.last_route.fallback)
+        self.assertTrue(router.last_route.technical_fallback)
+        self.assertIn("lokale Antwort", self.notices[-1])
+
+    def test_cloud_stream_error_after_output_never_appends_local_answer(self) -> None:
+        self.cloud.stream_chunks = ("Teilantwort",)
+        self.cloud.stream_error = CloudNetworkError("offline")
+        self.local.stream_chunks = ("Lokaler Ersatz",)
+        router = self._router()
+        stream = router.stream_chat([ChatMessage("user", "Wie funktioniert Python?")])
+
+        self.assertEqual(next(stream), "Teilantwort")
+        with self.assertRaises(ResponseAbortedError):
+            next(stream)
+
+        self.assertFalse(self.local.calls)
+        self.assertFalse(router.last_route.fallback)
+        self.assertTrue(router.last_route.technical_fallback)
+
+    def test_medical_never_streams_locally_before_cloud_construction(self) -> None:
+        constructed = False
+        self.local.stream_chunks = ("Lokal",)
+
+        def cloud_factory() -> _RecordingProvider:
+            nonlocal constructed
+            constructed = True
+            return self.cloud
+
+        router = self._router(cloud_provider_factory=cloud_factory, mode="gemini")
+
+        chunks = list(
+            router.stream_chat([ChatMessage("user", "Welche Diagnose passt zu meinen Schmerzen?")])
+        )
+
+        self.assertEqual(chunks, ["Lokal"])
+        self.assertFalse(constructed)
+        self.assertFalse(self.approvals)
+        self.assertEqual(router.last_route.privacy_policy, CloudPolicy.NEVER)
 
 
 if __name__ == "__main__":

tests/unit/test_sqlite_store.py

@@ -1,15 +1,20 @@
+import sqlite3
 import tempfile
 import unittest
 from pathlib import Path
 
-from javis.memory.sqlite_store import SessionNotFoundError, SQLiteSessionStore
+from javis.memory.sqlite_store import (
+    AmbiguousSessionReferenceError,
+    SessionNotFoundError,
+    SQLiteSessionStore,
+)
 
 
 class SQLiteSessionStoreTests(unittest.TestCase):
     def setUp(self) -> None:
         self.temporary_directory = tempfile.TemporaryDirectory()
-        database = Path(self.temporary_directory.name) / "sessions.sqlite3"
-        self.store = SQLiteSessionStore(database)
+        self.database = Path(self.temporary_directory.name) / "sessions.sqlite3"
+        self.store = SQLiteSessionStore(self.database)
 
     def tearDown(self) -> None:
         self.temporary_directory.cleanup()
@@ -21,6 +26,8 @@ class SQLiteSessionStoreTests(unittest.TestCase):
         self.assertEqual(loaded.id, created.id)
         self.assertEqual(loaded.provider, "ollama")
         self.assertEqual(loaded.model, "test-model")
+        self.assertEqual(loaded.title, "Neue Sitzung")
+        self.assertEqual(loaded.last_provider, "ollama")
         self.assertEqual(self.store.list_sessions()[0].id, created.id)
 
     def test_messages_keep_exchange_order(self) -> None:
@@ -53,6 +60,112 @@ class SQLiteSessionStoreTests(unittest.TestCase):
         self.assertEqual(self.store.get_messages(session.id), [])
         self.assertEqual(self.store.get_session(session.id).message_count, 0)
 
+    def test_first_exchange_sets_title_and_last_provider_atomically(self) -> None:
+        session = self.store.create_session("hybrid", "auto")
+
+        self.store.append_exchange(
+            session.id,
+            "Frage",
+            "Antwort",
+            title_if_first="Kurzer Titel",
+            last_provider="gemini",
+        )
+        self.store.append_exchange(
+            session.id,
+            "Noch eine Frage",
+            "Noch eine Antwort",
+            title_if_first="Darf Titel nicht ersetzen",
+            last_provider="ollama",
+        )
+
+        loaded = self.store.get_session(session.id)
+        self.assertEqual(loaded.title, "Kurzer Titel")
+        self.assertEqual(loaded.last_provider, "ollama")
+
+    def test_old_database_is_migrated_idempotently_without_data_loss(self) -> None:
+        database = Path(self.temporary_directory.name) / "legacy.sqlite3"
+        with sqlite3.connect(database) as connection:
+            connection.executescript(
+                """
+                CREATE TABLE sessions (
+                    id TEXT PRIMARY KEY,
+                    created_at TEXT NOT NULL,
+                    updated_at TEXT NOT NULL,
+                    provider TEXT NOT NULL,
+                    model TEXT NOT NULL
+                );
+                CREATE TABLE messages (
+                    id INTEGER PRIMARY KEY AUTOINCREMENT,
+                    session_id TEXT NOT NULL,
+                    position INTEGER NOT NULL,
+                    role TEXT NOT NULL,
+                    content TEXT NOT NULL,
+                    created_at TEXT NOT NULL,
+                    UNIQUE (session_id, position)
+                );
+                INSERT INTO sessions VALUES
+                    ('legacy', '2026-01-01', '2026-01-02', 'ollama', 'old-model');
+                INSERT INTO messages
+                    (session_id, position, role, content, created_at)
+                VALUES ('legacy', 0, 'user', 'Alte Nachricht', '2026-01-02');
+                """
+            )
+        connection.close()
+
+        migrated = SQLiteSessionStore(database)
+        migrated_again = SQLiteSessionStore(database)
+
+        session = migrated_again.get_session("legacy")
+        self.assertEqual(session.title, "Neue Sitzung")
+        self.assertEqual(session.last_provider, "ollama")
+        self.assertEqual(
+            [message.content for message in migrated.get_messages("legacy")],
+            ["Alte Nachricht"],
+        )
+
+    def test_rename_and_search_only_titles_and_user_messages(self) -> None:
+        first = self.store.create_session("ollama", "test-model")
+        second = self.store.create_session("ollama", "test-model")
+        self.store.rename_session(first.id, "SQLite Hilfe")
+        self.store.append_exchange(first.id, "Transaktion erklären", "Privates Lösungswort")
+        self.store.append_exchange(second.id, "Andere Frage", "Nur SQLite in Antwort")
+
+        self.assertEqual(self.store.get_session(first.id).title, "SQLite Hilfe")
+        self.assertEqual(
+            [session.id for session in self.store.search_sessions("Transaktion")],
+            [first.id],
+        )
+        self.assertEqual(self.store.search_sessions("Lösungswort"), [])
+        self.assertEqual(self.store.search_sessions("SQLite"), [self.store.get_session(first.id)])
+
+    def test_search_treats_sql_wildcards_and_injection_as_plain_text(self) -> None:
+        session = self.store.create_session("ollama", "test-model")
+        self.store.rename_session(session.id, "100% SQLite_Name")
+
+        self.assertEqual(self.store.search_sessions("%")[0].id, session.id)
+        self.assertEqual(self.store.search_sessions("_")[0].id, session.id)
+        self.assertEqual(self.store.search_sessions("' OR 1=1 --"), [])
+
+    def test_unique_and_ambiguous_session_prefixes(self) -> None:
+        with sqlite3.connect(self.database) as connection:
+            connection.executemany(
+                """
+                INSERT INTO sessions
+                    (id, created_at, updated_at, provider, model, title, last_provider)
+                VALUES (?, '2026-01-01', '2026-01-01', 'ollama', 'model', ?, 'ollama')
+                """,
+                (
+                    ("abc111", "Eins"),
+                    ("abc222", "Zwei"),
+                    ("unique333", "Drei"),
+                ),
+            )
+        connection.close()
+
+        self.assertEqual(self.store.resolve_unique_prefix("unique").id, "unique333")
+        with self.assertRaises(AmbiguousSessionReferenceError):
+            self.store.resolve_unique_prefix("abc")
+
 
 if __name__ == "__main__":
     unittest.main()

tests/unit/test_start_script.py

@@ -25,6 +25,24 @@ class StartScriptTests(unittest.TestCase):
         ):
             self.assertNotIn(forbidden, normalized)
 
+    def test_gui_script_reuses_safe_starter_without_system_changes(self) -> None:
+        repository_root = Path(__file__).resolve().parents[2]
+        script = (repository_root / "scripts" / "start-javis-gui.ps1").read_text(encoding="utf-8")
+        normalized = script.lower()
+
+        self.assertIn("start-javis.ps1", normalized)
+        self.assertIn("javis_start_interface", normalized)
+        self.assertIn("'gui'", normalized)
+        self.assertNotIn("c:\\users\\", normalized)
+        for forbidden in (
+            "set-executionpolicy",
+            "setx ",
+            "new-service",
+            "schtasks",
+            "hkey_",
+        ):
+            self.assertNotIn(forbidden, normalized)
+
 
 if __name__ == "__main__":
     unittest.main()

uv.lock

@@ -218,6 +218,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/98/2b/f97f1c193fb855c345d678f5077d6926034db0722df74c8f057020e05a25/charset_normalizer-3.4.9-py3-none-any.whl", hash = "sha256:68e5f26a1ad57ded6d1cfb85331d1c1a195314756471d97758c48498bb4dcdf5", size = 64538, upload-time = "2026-07-07T14:34:56.993Z" },
 ]
 
+[[package]]
+name = "colorama"
+version = "0.4.6"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/d8/53/6f443c9a4a8358a93a6792e2acffb9d9d5cb0a5cfd8802644b7b1c9a02e4/colorama-0.4.6.tar.gz", hash = "sha256:08695f5cb7ed6e0531a20572697297273c47b8cae5a63ffc6d6ed5c201be6e44", size = 27697, upload-time = "2022-10-25T02:36:22.414Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/d1/d6/3965ed04c63042e047cb6a3e6ed1a63a35087b6a609aa3a15ed8ac56c221/colorama-0.4.6-py2.py3-none-any.whl", hash = "sha256:4f1d9991f5acc0ca119f9d443620b77f9d6b33703e51011c16baf57afb285fc6", size = 25335, upload-time = "2022-10-25T02:36:20.889Z" },
+]
+
 [[package]]
 name = "cryptography"
 version = "49.0.0"
@@ -380,6 +389,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/38/3d/2d244233ac4f76e38533cfcb2991c9eb4c7bf688ae0a036d30725b8faafe/importlib_metadata-9.0.0-py3-none-any.whl", hash = "sha256:2d21d1cc5a017bd0559e36150c21c830ab1dc304dedd1b7ea85d20f45ef3edd7", size = 27789, upload-time = "2026-03-20T06:42:55.665Z" },
 ]
 
+[[package]]
+name = "iniconfig"
+version = "2.3.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/72/34/14ca021ce8e5dfedc35312d08ba8bf51fdd999c576889fc2c24cb97f4f10/iniconfig-2.3.0.tar.gz", hash = "sha256:c76315c77db068650d49c5b56314774a7804df16fee4402c1f19d6d15d8c4730", size = 20503, upload-time = "2025-10-18T21:55:43.219Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/cb/b1/3846dd7f199d53cb17f49cba7e651e9ce294d8497c8c150530ed11865bb8/iniconfig-2.3.0-py3-none-any.whl", hash = "sha256:f631c04d2c48c52b84d0d0549c99ff3859c98df65b3101406327ecc7d53fbf12", size = 7484, upload-time = "2025-10-18T21:55:41.639Z" },
+]
+
 [[package]]
 name = "jaraco-classes"
 version = "3.4.0"
@@ -423,10 +441,12 @@ source = { editable = "." }
 dependencies = [
     { name = "google-genai" },
     { name = "keyring" },
+    { name = "pyside6" },
 ]
 
 [package.dev-dependencies]
 dev = [
+    { name = "pytest" },
     { name = "ruff" },
 ]
 
@@ -434,10 +454,14 @@ dev = [
 requires-dist = [
     { name = "google-genai", specifier = ">=2.6,<3" },
     { name = "keyring", specifier = ">=25.7,<26" },
+    { name = "pyside6", specifier = ">=6.8,<7" },
 ]
 
 [package.metadata.requires-dev]
-dev = [{ name = "ruff", specifier = ">=0.12,<1" }]
+dev = [
+    { name = "pytest", specifier = ">=8,<9" },
+    { name = "ruff", specifier = ">=0.12,<1" },
+]
 
 [[package]]
 name = "jeepney"
@@ -475,6 +499,24 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/e8/3d/1087453384dbde46a8c7f9356eead2c58be8a7bf156bca40243377c85715/more_itertools-11.1.0-py3-none-any.whl", hash = "sha256:4b65538ae22f6fed0ce4874efd317463a7489796a0939fa66824dd542125a192", size = 72226, upload-time = "2026-05-22T14:14:28.824Z" },
 ]
 
+[[package]]
+name = "packaging"
+version = "26.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/d7/f1/e7a6dd94a8d4a5626c03e4e99c87f241ba9e350cd9e6d75123f992427270/packaging-26.2.tar.gz", hash = "sha256:ff452ff5a3e828ce110190feff1178bb1f2ea2281fa2075aadb987c2fb221661", size = 228134, upload-time = "2026-04-24T20:15:23.917Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/df/b2/87e62e8c3e2f4b32e5fe99e0b86d576da1312593b39f47d8ceef365e95ed/packaging-26.2-py3-none-any.whl", hash = "sha256:5fc45236b9446107ff2415ce77c807cee2862cb6fac22b8a73826d0693b0980e", size = 100195, upload-time = "2026-04-24T20:15:22.081Z" },
+]
+
+[[package]]
+name = "pluggy"
+version = "1.6.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f9/e2/3e91f31a7d2b083fe6ef3fa267035b518369d9511ffab804f839851d2779/pluggy-1.6.0.tar.gz", hash = "sha256:7dcc130b76258d33b90f61b658791dede3486c3e6bfb003ee5c9bfb396dd22f3", size = 69412, upload-time = "2025-05-15T12:30:07.975Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/54/20/4d324d65cc6d9205fabedc306948156824eb9f0ee1633355a8f7ec5c66bf/pluggy-1.6.0-py3-none-any.whl", hash = "sha256:e920276dd6813095e9377c0bc5566d94c932c33b27a3e3945d8389c374dd4746", size = 20538, upload-time = "2025-05-15T12:30:06.134Z" },
+]
+
 [[package]]
 name = "pyasn1"
 version = "0.6.4"
@@ -622,6 +664,79 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/4b/2d/69abac8f838090bbecd5df894befb2c2619e7996a98ddb949db9f3b93225/pydantic_core-2.46.4-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:d51026d73fcfd93610abc7b27789c26b313920fcfb20e27462d74a7f8b06e983", size = 2193071, upload-time = "2026-05-06T13:38:08.682Z" },
 ]
 
+[[package]]
+name = "pygments"
+version = "2.20.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/c3/b2/bc9c9196916376152d655522fdcebac55e66de6603a76a02bca1b6414f6c/pygments-2.20.0.tar.gz", hash = "sha256:6757cd03768053ff99f3039c1a36d6c0aa0b263438fcab17520b30a303a82b5f", size = 4955991, upload-time = "2026-03-29T13:29:33.898Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f4/7e/a72dd26f3b0f4f2bf1dd8923c85f7ceb43172af56d63c7383eb62b332364/pygments-2.20.0-py3-none-any.whl", hash = "sha256:81a9e26dd42fd28a23a2d169d86d7ac03b46e2f8b59ed4698fb4785f946d0176", size = 1231151, upload-time = "2026-03-29T13:29:30.038Z" },
+]
+
+[[package]]
+name = "pyside6"
+version = "6.11.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pyside6-addons" },
+    { name = "pyside6-essentials" },
+    { name = "shiboken6" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/da/a6/27ba5947ed48918f7b74b7c43a1e280aac069e36f25adeb4c9adfac835c4/pyside6-6.11.1-cp310-abi3-macosx_13_0_universal2.whl", hash = "sha256:537682c3b7530817203e667c1f5a2f00486b37bf52c52eeab438544c7a0917f6", size = 571921, upload-time = "2026-05-13T09:47:36.402Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/de/af89d71410c83b10654d86ff9aff2a4f87c30163658f1cc145242e222526/pyside6-6.11.1-cp310-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:b1fc521ba2bb5109425ab8add06bddbdd524abcad06cfa012cc39a22a189feb2", size = 572102, upload-time = "2026-05-13T09:47:38.249Z" },
+    { url = "https://files.pythonhosted.org/packages/b6/0e/d583bd3f7bf5046a4497b36f3902cfb64aa29554489a5a25c18e6b4ac0ac/pyside6-6.11.1-cp310-abi3-manylinux_2_39_aarch64.whl", hash = "sha256:75f0005c3eb95c07cfb65522ec50d0815ac007a96482c21dc3cb4b4c04895d84", size = 572098, upload-time = "2026-05-13T09:47:39.44Z" },
+    { url = "https://files.pythonhosted.org/packages/57/f2/d9d8ce1373dabb37e5919f63cd18446556079631d3f2eea3ada03c29f6b8/pyside6-6.11.1-cp310-abi3-win_amd64.whl", hash = "sha256:0968877ab1fb4ef3587a284da6fe05e8647ada56a6a3750b6395188e01f4aba6", size = 578377, upload-time = "2026-05-13T09:47:40.76Z" },
+    { url = "https://files.pythonhosted.org/packages/96/02/a6057d8bd2bdb1940820fff2d627fdf4013148c9c57adf69fa40d3452ac3/pyside6-6.11.1-cp310-abi3-win_arm64.whl", hash = "sha256:acee467cb5f256cc47ebb9d815a054c1d8416da380c191b247a76d164aa3f805", size = 561765, upload-time = "2026-05-13T09:47:41.9Z" },
+]
+
+[[package]]
+name = "pyside6-addons"
+version = "6.11.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pyside6-essentials" },
+    { name = "shiboken6" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3f/6b/8bc94aff48b63f788f2d84e5467c12362d68906ba742c0942f46cb04c879/pyside6_addons-6.11.1-cp310-abi3-macosx_13_0_universal2.whl", hash = "sha256:54733c77f789bef5f03c6aff4ad3bec8b2eff021f0cfcbc53d5e6c250ded24f9", size = 331714589, upload-time = "2026-05-13T09:39:12.36Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/62/fb1428a523b2a4541e232aab50d9e789e6b4526f37fd9593452a7ea5b6b3/pyside6_addons-6.11.1-cp310-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:8e6c65fbd73a512d6f72cda8d8277444a85a34dc99dd1dae9c21d35b8671bb1f", size = 175063224, upload-time = "2026-05-13T09:39:34.185Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/9b/2ccd52f66db55c06de65d0501170a1935d04d64d0a230c0d892284a02ce3/pyside6_addons-6.11.1-cp310-abi3-manylinux_2_39_aarch64.whl", hash = "sha256:bf1c6c4e954e5eba3d2a7c661ad4b9689e8f09c7f4a16bdf29713371d11af993", size = 170553429, upload-time = "2026-05-13T09:39:54.424Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/bd/8adc4d350b3b363f3dfc8fccdcf5bfed25f7e36c2fff30c64e106f4f1572/pyside6_addons-6.11.1-cp310-abi3-win_amd64.whl", hash = "sha256:0d13c4dfd671b050a48e4f8d8ddc724b7248f9c0437e7fc47fdf316278572923", size = 168816308, upload-time = "2026-05-13T09:40:13.541Z" },
+    { url = "https://files.pythonhosted.org/packages/65/b7/9a840d97f0f0f04e372a87e205dd30ee285b4e3b021b188459a917c9dc76/pyside6_addons-6.11.1-cp310-abi3-win_arm64.whl", hash = "sha256:3494f480dee92f415be2f2d989c0b3f4755ac332b28045cbf4ba0f5c5a22ba37", size = 35759347, upload-time = "2026-05-13T09:40:21.199Z" },
+]
+
+[[package]]
+name = "pyside6-essentials"
+version = "6.11.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "shiboken6" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b3/da/10d9197e7370eb4fed8df5fc547b7548dec88e5c5949e2d450db4ae96feb/pyside6_essentials-6.11.1-cp310-abi3-macosx_13_0_universal2.whl", hash = "sha256:228de53c2bc26b07e5021fbe3614fc44ca08e4dab9999af08c2b389d2c239957", size = 110352945, upload-time = "2026-05-13T09:43:08.006Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/49/0e1237c4400bec7e335d2c4eeb49bc40d9fd88a9ac44ca9083ce1abdc308/pyside6_essentials-6.11.1-cp310-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:e3ef7027b41e4e55fadb56e3b3257dc8ee92154b639fe67fc4c8e05e9d976c60", size = 79908535, upload-time = "2026-05-13T09:43:24.836Z" },
+    { url = "https://files.pythonhosted.org/packages/4c/c5/da4c5f23c6540ac5211a1f60177c8dee84b1bf40f2719479587ab8c60731/pyside6_essentials-6.11.1-cp310-abi3-manylinux_2_39_aarch64.whl", hash = "sha256:a039b6da68a3a4b9d243217b2b98d475eed3f617159ef6be925badab53c11b0d", size = 78960051, upload-time = "2026-05-13T09:43:35.423Z" },
+    { url = "https://files.pythonhosted.org/packages/64/0e/b663ecc96ca57b5c91b83b6615d6b174380b0faf30338125c26e053d6aa7/pyside6_essentials-6.11.1-cp310-abi3-win_amd64.whl", hash = "sha256:63311bd48e32c584599ab04b9ef7c324082374cd2c9fa533f978fb893bb47e40", size = 77549267, upload-time = "2026-05-13T09:43:44.92Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/12/eb6723faf5cb7fa581145da1c15f40d641b96e080f0491af2f1859fdeedb/pyside6_essentials-6.11.1-cp310-abi3-win_arm64.whl", hash = "sha256:11253ea52aabecefe9febddbbe78b43a824129e3af1cec98431028fba7fa954f", size = 57964512, upload-time = "2026-05-13T09:43:52.968Z" },
+]
+
+[[package]]
+name = "pytest"
+version = "8.4.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+    { name = "iniconfig" },
+    { name = "packaging" },
+    { name = "pluggy" },
+    { name = "pygments" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/a3/5c/00a0e072241553e1a7496d638deababa67c5058571567b92a7eaa258397c/pytest-8.4.2.tar.gz", hash = "sha256:86c0d0b93306b961d58d62a4db4879f27fe25513d4b969df351abdddb3c30e01", size = 1519618, upload-time = "2025-09-04T14:34:22.711Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/a8/a4/20da314d277121d6534b3a980b29035dcd51e6744bd79075a6ce8fa4eb8d/pytest-8.4.2-py3-none-any.whl", hash = "sha256:872f880de3fc3a5bdc88a11b39c9710c3497a547cfa9320bc3c5e62fbf272e79", size = 365750, upload-time = "2025-09-04T14:34:20.226Z" },
+]
+
 [[package]]
 name = "pywin32-ctypes"
 version = "0.2.3"
@@ -684,6 +799,18 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/b7/46/f5af3402b579fd5e11573ce652019a67074317e18c1935cc0b4ba9b35552/secretstorage-3.5.0-py3-none-any.whl", hash = "sha256:0ce65888c0725fcb2c5bc0fdb8e5438eece02c523557ea40ce0703c266248137", size = 15554, upload-time = "2025-11-23T19:02:51.545Z" },
 ]
 
+[[package]]
+name = "shiboken6"
+version = "6.11.1"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/17/f3/f2b63df0251e7cd3172ea28e32ede52739de9566bcefcd0178681538ac81/shiboken6-6.11.1-cp310-abi3-macosx_13_0_universal2.whl", hash = "sha256:1a16867f103ef1c662a5f09dfed03273a9f81688b174555162c58e83650a3f02", size = 476874, upload-time = "2026-05-13T09:47:01.091Z" },
+    { url = "https://files.pythonhosted.org/packages/c7/9b/e0355d8897b5c150770f1d95718aad17d432fcc9c035c04f3f58427d4693/shiboken6-6.11.1-cp310-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:9a8bccfafc8805254cabcfa1edfaf55cd52889f4998c91ad0d9a4433fb1bcdbe", size = 272222, upload-time = "2026-05-13T09:47:02.653Z" },
+    { url = "https://files.pythonhosted.org/packages/57/d5/dd4f1defed400be03340f2ede34b61f846776650b4e7ed9ebaf4c71979a2/shiboken6-6.11.1-cp310-abi3-manylinux_2_39_aarch64.whl", hash = "sha256:1bd2f4314414df2d122d9f646e03b731bc6d6b5f77a5f53f99a4fe4e97d84e6f", size = 270350, upload-time = "2026-05-13T09:47:04.02Z" },
+    { url = "https://files.pythonhosted.org/packages/52/b5/3f6fb2ee65b534193fb4ef713dd619dc31dadff5d12c16979a7699ad58be/shiboken6-6.11.1-cp310-abi3-win_amd64.whl", hash = "sha256:c2c6863aa80ec18c0f82cea3417837b279cdc60024ac17123461dc9042577df7", size = 1223647, upload-time = "2026-05-13T09:47:05.924Z" },
+    { url = "https://files.pythonhosted.org/packages/98/d1/f15ca0e1666faae02c945f48e745ea35f8fcd8243b176109b4e2c4251f47/shiboken6-6.11.1-cp310-abi3-win_arm64.whl", hash = "sha256:7c8d9af17db4495d4fa5b1c393f218311c4855546b9dfa6a0bd21bcd66b55e9d", size = 1784170, upload-time = "2026-05-13T09:47:07.617Z" },
+]
+
 [[package]]
 name = "sniffio"
 version = "1.3.1"