feat(memory): Implement token based summarization

.env.example

@@ -321,16 +321,14 @@ SAGEMAKER_ACCEPT=application/json            # Response content type
 # =============================================================================
 # CONVERSATION MEMORY CONFIGURATION (Optional)
 # =============================================================================
-# Enable conversation memory feature (disabled by default)
-NEUROLINK_MEMORY_ENABLED=false
+# Enable conversation memory feature (ENABLED BY DEFAULT)
+NEUROLINK_MEMORY_ENABLED=true                 # Enable conversation memory (default: true)
 
 # Memory storage type (memory or redis)
 STORAGE_TYPE=memory                            # Options: memory, redis
 
 # Memory limits
 NEUROLINK_MEMORY_MAX_SESSIONS=50              # Maximum number of sessions to keep in memory
-NEUROLINK_MEMORY_MAX_TURNS_PER_SESSION=50     # Maximum conversation turns per session
-
 # Redis Storage Configuration (used when STORAGE_TYPE=redis)
 REDIS_HOST=localhost                           # Redis server hostname
 REDIS_PORT=6379                                # Redis server port
@@ -403,12 +401,16 @@ NEUROLINK_MAX_TOOLS_PER_PROVIDER=5
 # Config File Location
 NEUROLINK_CONFIG_FILE=./neurolink.config.json
 
-# Summarization (Conversation Memory)
-NEUROLINK_SUMMARIZATION_ENABLED=false
-NEUROLINK_SUMMARIZATION_THRESHOLD_TURNS=20
-NEUROLINK_SUMMARIZATION_TARGET_TURNS=10
-NEUROLINK_SUMMARIZATION_PROVIDER=google-ai
-NEUROLINK_SUMMARIZATION_MODEL=gemini-2.5-flash
+# Summarization (Conversation Memory) - TOKEN-BASED MEMORY ENABLED BY DEFAULT
+# Token-based memory is enabled by default and uses 80% of each model's context window
+NEUROLINK_SUMMARIZATION_ENABLED=true             # Enable summarization (default: true)
+NEUROLINK_TOKEN_THRESHOLD=50000                  # Optional: Override token threshold (default: 80% of model context)
+NEUROLINK_SUMMARIZATION_PROVIDER=vertex          # Provider for summarization (default: vertex)
+NEUROLINK_SUMMARIZATION_MODEL=gemini-2.5-flash   # Model for summarization (default: gemini-2.5-flash)
+
+# Deprecated: Turn-based memory settings (use TOKEN_THRESHOLD instead)
+# NEUROLINK_SUMMARIZATION_THRESHOLD_TURNS=20     # Deprecated: Use token threshold
+# NEUROLINK_SUMMARIZATION_TARGET_TURNS=10        # Deprecated: Use token threshold
 
 # Default Generation Parameters
 NEUROLINK_DEFAULT_MAX_TOKENS=4096

src/cli/loop/optionsSchema.ts

@@ -87,4 +87,9 @@ export const textGenerationOptionsSchema: Record<
     type: "string",
     description: "Context about tools/MCPs used in the interaction.",
   },
+  enableSummarization: {
+    type: "boolean",
+    description:
+      "Enable or disable automatic conversation summarization for this request.",
+  },
 };

src/lib/config/conversationMemory.ts

@@ -33,6 +33,24 @@ IMPORTANT: You are continuing an ongoing conversation. The previous messages in
 
 Always reference and build upon this conversation history when relevant. If the user asks about information mentioned earlier in the conversation, refer to those previous messages to provide accurate, contextual responses.`;
 
+/**
+ * Percentage of model context window to use for conversation memory threshold
+ * Default: 80% of model's context window
+ */
+export const MEMORY_THRESHOLD_PERCENTAGE = 0.8;
+
+/**
+ * Fallback token threshold if model context unknown
+ */
+export const DEFAULT_FALLBACK_THRESHOLD = 50000;
+
+/**
+ * Ratio of threshold to keep as recent unsummarized messages
+ * When summarization triggers, this percentage of tokens from the end
+ * are preserved as detailed messages, while older content gets summarized.
+ */
+export const RECENT_MESSAGES_RATIO = 0.3;
+
 /**
  * Structured output instructions for JSON/structured output mode
  * Used to ensure AI providers output only valid JSON without conversational filler
@@ -57,17 +75,23 @@ export function getConversationMemoryDefaults(): ConversationMemoryConfig {
     enabled: process.env.NEUROLINK_MEMORY_ENABLED === "true",
     maxSessions:
       Number(process.env.NEUROLINK_MEMORY_MAX_SESSIONS) || DEFAULT_MAX_SESSIONS,
+    enableSummarization:
+      process.env.NEUROLINK_SUMMARIZATION_ENABLED !== "false",
+    tokenThreshold: process.env.NEUROLINK_TOKEN_THRESHOLD
+      ? Number(process.env.NEUROLINK_TOKEN_THRESHOLD)
+      : undefined,
+    summarizationProvider:
+      process.env.NEUROLINK_SUMMARIZATION_PROVIDER || "vertex",
+    summarizationModel:
+      process.env.NEUROLINK_SUMMARIZATION_MODEL || "gemini-2.5-flash",
+
+    // Deprecated (for backward compatibility)
     maxTurnsPerSession:
       Number(process.env.NEUROLINK_MEMORY_MAX_TURNS_PER_SESSION) ||
       DEFAULT_MAX_TURNS_PER_SESSION,
-    enableSummarization: process.env.NEUROLINK_SUMMARIZATION_ENABLED === "true",
     summarizationThresholdTurns:
       Number(process.env.NEUROLINK_SUMMARIZATION_THRESHOLD_TURNS) || 20,
     summarizationTargetTurns:
       Number(process.env.NEUROLINK_SUMMARIZATION_TARGET_TURNS) || 10,
-    summarizationProvider:
-      process.env.NEUROLINK_SUMMARIZATION_PROVIDER || "vertex",
-    summarizationModel:
-      process.env.NEUROLINK_SUMMARIZATION_MODEL || "gemini-2.5-flash",
   };
 }

src/lib/core/conversationMemoryFactory.ts

@@ -27,10 +27,7 @@ export function createConversationMemoryManager(
       config: {
         enabled: config.enabled,
         maxSessions: config.maxSessions,
-        maxTurnsPerSession: config.maxTurnsPerSession,
         enableSummarization: config.enableSummarization,
-        summarizationThresholdTurns: config.summarizationThresholdTurns,
-        summarizationTargetTurns: config.summarizationTargetTurns,
         summarizationProvider: config.summarizationProvider,
         summarizationModel: config.summarizationModel,
       },

src/lib/core/conversationMemoryInitializer.ts

@@ -113,15 +113,6 @@ export async function initializeConversationMemory(config?: {
 
       logger.info(
         "[conversationMemoryInitializer] Redis conversation memory manager created successfully",
-        {
-          configSource,
-          host: redisConfig.host || "localhost",
-          port: redisConfig.port || 6379,
-          keyPrefix: redisConfig.keyPrefix || "neurolink:conversation:",
-          maxSessions: memoryConfig.maxSessions,
-          maxTurnsPerSession: memoryConfig.maxTurnsPerSession,
-          managerType: redisMemoryManager?.constructor?.name,
-        },
       );
 
       // Perform basic validation