Support memory configurable parameters

cli/src/test/kotlin/io/askimo/core/config/AppConfigTest.kt

@@ -181,20 +181,79 @@ class AppConfigTest {
     fun `updateChatField should handle all fields`() {
         var config = ChatConfig()
 
-        // Test top-level fields
         config = updateChatFieldHelper(config, "maxTokens", 10000)
         assertEquals(10000, config.maxTokens)
+    }
+
+    @Test
+    fun `updateMemoryField should handle mode - swaps to full preset`() {
+        var config = MemoryConfig()
+
+        config = updateMemoryFieldHelper(config, "mode", MemoryMode.COMPACT)
+        assertEquals(MemoryMode.COMPACT, config.mode)
+        assertEquals(0.25, config.summarizationThreshold, 0.001)
+        assertEquals(3, config.protectedRecentTurns)
+        assertEquals(0.30, config.memoryBudgetFraction, 0.001)
+
+        config = updateMemoryFieldHelper(config, "mode", "DETAIL")
+        assertEquals(MemoryMode.DETAIL, config.mode)
+        assertEquals(0.60, config.summarizationThreshold, 0.001)
+        assertEquals(10, config.protectedRecentTurns)
+        assertEquals(0.50, config.memoryBudgetFraction, 0.001)
+    }
 
-        config = updateChatFieldHelper(config, "summarizationThreshold", 0.8)
-        assertEquals(0.8, config.summarizationThreshold, 0.001)
+    @Test
+    fun `updateMemoryField should handle all numeric fields`() {
+        var config = MemoryConfig()
 
-        config = updateChatFieldHelper(config, "enableAsyncSummarization", false)
-        assertFalse(config.enableAsyncSummarization)
+        config = updateMemoryFieldHelper(config, "summarizationThreshold", 0.5)
+        assertEquals(0.5, config.summarizationThreshold, 0.001)
 
-        // Verify all fields are correct after multiple updates
-        assertEquals(10000, config.maxTokens)
-        assertEquals(0.8, config.summarizationThreshold, 0.001)
-        assertFalse(config.enableAsyncSummarization)
+        config = updateMemoryFieldHelper(config, "protectedRecentTurns", 8)
+        assertEquals(8, config.protectedRecentTurns)
+
+        config = updateMemoryFieldHelper(config, "summarizationPruneFraction", 0.7)
+        assertEquals(0.7, config.summarizationPruneFraction, 0.001)
+
+        config = updateMemoryFieldHelper(config, "maxKeyFacts", 40)
+        assertEquals(40, config.maxKeyFacts)
+
+        config = updateMemoryFieldHelper(config, "maxMainTopics", 20)
+        assertEquals(20, config.maxMainTopics)
+
+        config = updateMemoryFieldHelper(config, "maxSummaryLength", 3000)
+        assertEquals(3000, config.maxSummaryLength)
+
+        config = updateMemoryFieldHelper(config, "memoryBudgetFraction", 0.45)
+        assertEquals(0.45, config.memoryBudgetFraction, 0.001)
+    }
+
+    @Test
+    fun `MemoryConfig preset values should be correct`() {
+        val compact = MemoryConfig.COMPACT
+        assertEquals(MemoryMode.COMPACT, compact.mode)
+        assertEquals(0.25, compact.summarizationThreshold, 0.001)
+        assertEquals(3, compact.protectedRecentTurns)
+        assertEquals(0.30, compact.memoryBudgetFraction, 0.001)
+
+        val balanced = MemoryConfig.BALANCED
+        assertEquals(MemoryMode.BALANCED, balanced.mode)
+        assertEquals(0.40, balanced.summarizationThreshold, 0.001)
+        assertEquals(6, balanced.protectedRecentTurns)
+        assertEquals(0.40, balanced.memoryBudgetFraction, 0.001)
+
+        val detail = MemoryConfig.DETAIL
+        assertEquals(MemoryMode.DETAIL, detail.mode)
+        assertEquals(0.60, detail.summarizationThreshold, 0.001)
+        assertEquals(10, detail.protectedRecentTurns)
+        assertEquals(0.50, detail.memoryBudgetFraction, 0.001)
+    }
+
+    @Test
+    fun `MemoryConfig preset factory should return correct preset`() {
+        assertEquals(MemoryConfig.preset(MemoryMode.COMPACT), MemoryConfig.COMPACT)
+        assertEquals(MemoryConfig.preset(MemoryMode.BALANCED), MemoryConfig.BALANCED)
+        assertEquals(MemoryConfig.preset(MemoryMode.DETAIL), MemoryConfig.DETAIL)
     }
 
     @Test
@@ -252,6 +311,17 @@ class AppConfigTest {
         return method.invoke(AppConfig, config, field, value) as ChatConfig
     }
 
+    private fun updateMemoryFieldHelper(config: MemoryConfig, field: String, value: Any): MemoryConfig {
+        val method = AppConfig::class.java.getDeclaredMethod(
+            "updateMemoryField",
+            MemoryConfig::class.java,
+            String::class.java,
+            Any::class.java,
+        )
+        method.isAccessible = true
+        return method.invoke(AppConfig, config, field, value) as MemoryConfig
+    }
+
     private fun updateDeveloperFieldHelper(config: DeveloperConfig, field: String, value: Any): DeveloperConfig {
         val method = AppConfig::class.java.getDeclaredMethod(
             "updateDeveloperField",

cli/src/test/kotlin/io/askimo/core/memory/TokenAwareSummarizingMemoryTest.kt

@@ -10,6 +10,7 @@ import dev.langchain4j.data.message.SystemMessage
 import dev.langchain4j.data.message.UserMessage
 import io.askimo.core.chat.domain.SessionMemory
 import io.askimo.core.chat.repository.SessionMemoryRepository
+import io.askimo.core.config.AppConfig
 import io.askimo.core.context.AppContext
 import io.askimo.core.providers.ChatClient
 import io.askimo.test.extensions.AskimoTestHome
@@ -542,17 +543,20 @@ class TokenAwareSummarizingMemoryTest {
     // ── Helper ───────────────────────────────────────────────────────────────
 
     private fun createMemory(
-        summarizationThreshold: Double = 0.4, // matches updated default
+        summarizationThreshold: Double = 0.4,
         asyncSummarization: Boolean = true,
         tokenEstimator: (ChatMessage) -> Int = TokenAwareSummarizingMemory.defaultTokenEstimator(),
         summarizationTimeoutSeconds: Long = 30,
-    ): TokenAwareSummarizingMemory = TokenAwareSummarizingMemory(
-        appContext = mockAppContext,
-        sessionId = sessionId,
-        sessionMemoryRepository = mockRepository,
-        tokenEstimator = tokenEstimator,
-        summarizationThreshold = summarizationThreshold,
-        asyncSummarization = asyncSummarization,
-        summarizationTimeoutSeconds = summarizationTimeoutSeconds,
-    )
+    ): TokenAwareSummarizingMemory {
+        // Drive threshold through AppConfig — @AskimoTestHome resets config between tests
+        AppConfig.updateField("memory.summarizationThreshold", summarizationThreshold)
+        return TokenAwareSummarizingMemory(
+            appContext = mockAppContext,
+            sessionId = sessionId,
+            sessionMemoryRepository = mockRepository,
+            tokenEstimator = tokenEstimator,
+            asyncSummarization = asyncSummarization,
+            summarizationTimeoutSeconds = summarizationTimeoutSeconds,
+        )
+    }
 }

desktop-shared/src/main/resources/i18n/messages.properties

@@ -1147,6 +1147,20 @@ settings.models.description=Configure global AI model behavior shared across all
 settings.models.max.tool.calling.round.trips=Max Tool-Calling Round Trips
 settings.models.max.tool.calling.round.trips.hint=Maximum number of tool calls the AI can make in a single response (default: 10). Higher values allow more complex agent tasks but increase latency and cost.
 
+# Memory Settings
+settings.memory.title=Memory Settings
+settings.memory.description=Controls how the AI manages conversation history to stay within the model's context window.
+settings.memory.mode=Memory Mode
+settings.memory.mode.compact=Compact
+settings.memory.mode.compact.hint=Lower token cost, less detail
+settings.memory.mode.compact.description=Aggressive summarization. Lower token cost, summarizes early and often. Best for very long sessions or cost-sensitive models.
+settings.memory.mode.balanced=Balanced
+settings.memory.mode.balanced.hint=Good quality/cost trade-off
+settings.memory.mode.balanced.description=Good quality/cost trade-off. Fires summarization at 40% of the memory budget. Suitable for most sessions.
+settings.memory.mode.detail=Detail
+settings.memory.mode.detail.hint=High fidelity, more tokens
+settings.memory.mode.detail.description=Minimal summarization. Keeps more verbatim turns and richer summaries. Best for short sessions or when precise recall matters.
+
 ## Settings - Utility Models
 settings.utility.models.title=Utility Models
 settings.utility.models.uses.selected=(uses selected model)

desktop-shared/src/main/resources/i18n/messages_de.properties

@@ -1137,6 +1137,20 @@ settings.models.description=Konfigurieren Sie das globale KI-Modellverhalten, da
 settings.models.max.tool.calling.round.trips=Maximale Tool-Aufruf-Runden
 settings.models.max.tool.calling.round.trips.hint=Maximale Anzahl an Tool-Aufrufen, die die KI in einer einzelnen Antwort durchführen kann (Standard: 10). Höhere Werte ermöglichen komplexere Agenten-Aufgaben, erhöhen jedoch Latenz und Kosten.
 
+# Memory Settings
+settings.memory.title=Speichereinstellungen
+settings.memory.description=Steuert, wie die KI den Gesprächsverlauf verwaltet, um im Kontextfenster des Modells zu bleiben.
+settings.memory.mode=Speichermodus
+settings.memory.mode.compact=Kompakt
+settings.memory.mode.compact.hint=Geringere Token-Kosten, weniger Details
+settings.memory.mode.compact.description=Aggressives Zusammenfassen. Geringe Token-Kosten, häufiges Zusammenfassen. Ideal für sehr lange Sitzungen oder kostensensitive Modelle.
+settings.memory.mode.balanced=Ausgewogen
+settings.memory.mode.balanced.hint=Gutes Qualitäts-/Kostenverhältnis
+settings.memory.mode.balanced.description=Gutes Qualitäts-/Kostenverhältnis. Löst die Zusammenfassung bei 40 % des Speicherbudgets aus. Geeignet für die meisten Sitzungen.
+settings.memory.mode.detail=Detailliert
+settings.memory.mode.detail.hint=Hohe Wiedergabetreue, mehr Tokens
+settings.memory.mode.detail.description=Minimales Zusammenfassen. Behält mehr wörtliche Gesprächsrunden und reichhaltigere Zusammenfassungen bei. Ideal für kurze Sitzungen oder wenn präziser Abruf wichtig ist.
+
 # Settings - Utility Models
 settings.utility.models.title=Hilfsmodelle
 settings.utility.models.uses.selected=(verwendet das ausgewählte Modell)

desktop-shared/src/main/resources/i18n/messages_es.properties

@@ -1133,6 +1133,20 @@ settings.models.description=Configura el comportamiento global de los modelos de
 settings.models.max.tool.calling.round.trips=Máximo de rondas de llamadas a herramientas
 settings.models.max.tool.calling.round.trips.hint=Número máximo de llamadas a herramientas que la IA puede realizar en una sola respuesta (predeterminado: 10). Valores más altos permiten tareas de agente más complejas pero aumentan la latencia y el costo.
 
+# Memory Settings
+settings.memory.title=Configuración de memoria
+settings.memory.description=Controla cómo la IA gestiona el historial de conversación para mantenerse dentro del contexto del modelo.
+settings.memory.mode=Modo de memoria
+settings.memory.mode.compact=Compacto
+settings.memory.mode.compact.hint=Menor costo de tokens, menos detalle
+settings.memory.mode.compact.description=Resumen agresivo. Menor costo de tokens, resume pronto y con frecuencia. Ideal para sesiones muy largas o modelos sensibles al costo.
+settings.memory.mode.balanced=Equilibrado
+settings.memory.mode.balanced.hint=Buen equilibrio calidad/costo
+settings.memory.mode.balanced.description=Buen equilibrio calidad/costo. Activa el resumen al 40% del presupuesto de memoria. Adecuado para la mayoría de sesiones.
+settings.memory.mode.detail=Detallado
+settings.memory.mode.detail.hint=Alta fidelidad, más tokens
+settings.memory.mode.detail.description=Resumen mínimo. Conserva más turnos literales y resúmenes más ricos. Ideal para sesiones cortas o cuando el recuerdo preciso es importante.
+
 # Settings - Utility Models
 settings.utility.models.title=Modelos de utilidad
 settings.utility.models.uses.selected=(usa el modelo seleccionado)

desktop-shared/src/main/resources/i18n/messages_fr.properties

@@ -1134,6 +1134,20 @@ settings.models.description=Configurez le comportement global du modèle IA part
 settings.models.max.tool.calling.round.trips=Nombre maximal d'allers-retours d'appel d'outils
 settings.models.max.tool.calling.round.trips.hint=Nombre maximum d'appels d'outils que l'IA peut effectuer dans une seule réponse (défaut : 10). Des valeurs plus élevées permettent des tâches d'agent plus complexes mais augmentent la latence et le coût.
 
+# Memory Settings
+settings.memory.title=Paramètres de mémoire
+settings.memory.description=Contrôle la façon dont l'IA gère l'historique des conversations pour rester dans la fenêtre de contexte du modèle.
+settings.memory.mode=Mode mémoire
+settings.memory.mode.compact=Compact
+settings.memory.mode.compact.hint=Coût en tokens réduit, moins de détails
+settings.memory.mode.compact.description=Résumé agressif. Faible coût en tokens, résumé fréquent. Idéal pour les très longues sessions ou les modèles sensibles aux coûts.
+settings.memory.mode.balanced=Équilibré
+settings.memory.mode.balanced.hint=Bon équilibre qualité/coût
+settings.memory.mode.balanced.description=Bon équilibre qualité/coût. Déclenche le résumé à 40 % du budget mémoire. Convient à la plupart des sessions.
+settings.memory.mode.detail=Détaillé
+settings.memory.mode.detail.hint=Haute fidélité, plus de tokens
+settings.memory.mode.detail.description=Résumé minimal. Conserve plus de tours verbatim et des résumés plus riches. Idéal pour les sessions courtes ou lorsque le rappel précis est important.
+
 # Settings - Utility Models
 settings.utility.models.title=Modèles utilitaires
 settings.utility.models.uses.selected=(utilise le modèle sélectionné)

desktop-shared/src/main/resources/i18n/messages_ja_JP.properties

@@ -1135,6 +1135,20 @@ settings.models.description=すべてのプロバイダーで共有されるグ
 settings.models.max.tool.calling.round.trips=最大ツール呼び出しラウンドトリップ数
 settings.models.max.tool.calling.round.trips.hint=AI が 1 回の応答で実行できるツール呼び出しの最大回数です（デフォルト: 10）。値を大きくするほど複雑なエージェントタスクが可能になりますが、レイテンシとコストが増加します。
 
+# Memory Settings
+settings.memory.title=メモリ設定
+settings.memory.description=AIがモデルのコンテキストウィンドウ内に収まるよう会話履歴を管理する方法を制御します。
+settings.memory.mode=メモリモード
+settings.memory.mode.compact=コンパクト
+settings.memory.mode.compact.hint=トークンコストが低く、詳細が少ない
+settings.memory.mode.compact.description=積極的な要約。トークンコストが低く、早期かつ頻繁に要約します。非常に長いセッションやコストに敏感なモデルに最適です。
+settings.memory.mode.balanced=バランス
+settings.memory.mode.balanced.hint=品質とコストの良好なバランス
+settings.memory.mode.balanced.description=品質とコストの良好なバランス。メモリ予算の40%で要約が実行されます。ほとんどのセッションに適しています。
+settings.memory.mode.detail=詳細
+settings.memory.mode.detail.hint=高精度、より多くのトークン
+settings.memory.mode.detail.description=最小限の要約。より多くの原文の会話と豊富な要約を保持します。短いセッションや正確な記憶が重要な場合に最適です。
+
 # Settings - Utility Models
 settings.utility.models.title=ユーティリティモデル
 settings.utility.models.uses.selected=（選択されたモデルを使用）

desktop-shared/src/main/resources/i18n/messages_ko_KR.properties

@@ -1134,6 +1134,20 @@ settings.models.description=모든 공급자에서 공유되는 전역 AI 모델
 settings.models.max.tool.calling.round.trips=최대 도구 호출 라운드 트립
 settings.models.max.tool.calling.round.trips.hint=AI가 단일 응답에서 수행할 수 있는 최대 도구 호출 횟수입니다 (기본값: 10). 값이 높을수록 복잡한 에이전트 작업이 가능하지만 지연 시간과 비용이 증가합니다.
 
+# Memory Settings
+settings.memory.title=메모리 설정
+settings.memory.description=AI가 모델의 컨텍스트 창 안에서 대화 기록을 관리하는 방식을 제어합니다.
+settings.memory.mode=메모리 모드
+settings.memory.mode.compact=간결
+settings.memory.mode.compact.hint=낮은 토큰 비용, 적은 상세 정보
+settings.memory.mode.compact.description=적극적 요약. 토큰 비용이 낮고 자주 요약합니다. 매우 긴 세션이나 비용에 민감한 모델에 적합합니다.
+settings.memory.mode.balanced=균형
+settings.memory.mode.balanced.hint=품질과 비용의 적절한 균형
+settings.memory.mode.balanced.description=품질과 비용의 적절한 균형. 메모리 예산의 40%에서 요약이 시작됩니다. 대부분의 세션에 적합합니다.
+settings.memory.mode.detail=상세
+settings.memory.mode.detail.hint=높은 정확도, 더 많은 토큰
+settings.memory.mode.detail.description=최소 요약. 더 많은 원문 대화와 풍부한 요약을 유지합니다. 짧은 세션이나 정확한 기억이 중요할 때 적합합니다.
+
 # Settings - Utility Models
 settings.utility.models.title=유틸리티 모델
 settings.utility.models.uses.selected=(선택된 모델 사용)

desktop-shared/src/main/resources/i18n/messages_pt_BR.properties

@@ -1137,6 +1137,20 @@ settings.models.description=Configure o comportamento global do modelo de IA com
 settings.models.max.tool.calling.round.trips=Máximo de rodadas de chamadas de ferramentas
 settings.models.max.tool.calling.round.trips.hint=Número máximo de chamadas de ferramentas que a IA pode fazer em uma única resposta (padrão: 10). Valores mais altos permitem tarefas de agente mais complexas, mas aumentam a latência e o custo.
 
+# Memory Settings
+settings.memory.title=Configurações de memória
+settings.memory.description=Controla como a IA gerencia o histórico de conversas para se manter dentro da janela de contexto do modelo.
+settings.memory.mode=Modo de memória
+settings.memory.mode.compact=Compacto
+settings.memory.mode.compact.hint=Menor custo de tokens, menos detalhes
+settings.memory.mode.compact.description=Resumo agressivo. Menor custo de tokens, resume cedo e com frequência. Ideal para sessões muito longas ou modelos sensíveis ao custo.
+settings.memory.mode.balanced=Equilibrado
+settings.memory.mode.balanced.hint=Bom equilíbrio qualidade/custo
+settings.memory.mode.balanced.description=Bom equilíbrio qualidade/custo. Aciona o resumo a 40% do orçamento de memória. Adequado para a maioria das sessões.
+settings.memory.mode.detail=Detalhado
+settings.memory.mode.detail.hint=Alta fidelidade, mais tokens
+settings.memory.mode.detail.description=Resumo mínimo. Mantém mais turnos verbatim e resumos mais ricos. Ideal para sessões curtas ou quando o recall preciso é importante.
+
 # Settings - Utility Models
 settings.utility.models.title=Modelos utilitários
 settings.utility.models.uses.selected=(usa o modelo selecionado)