feat(core): Implement V0 Episodic Context Manager

docs/cli/settings.md

@@ -165,6 +165,7 @@ they appear in the UI.
 | Model Steering                                       | `experimental.modelSteering`        | Enable model steering (user hints) to guide the model during tool execution.                                                                              | `false` |
 | Direct Web Fetch                                     | `experimental.directWebFetch`       | Enable web fetch behavior that bypasses LLM summarization.                                                                                                | `false` |
 | Memory Manager Agent                                 | `experimental.memoryManager`        | Replace the built-in save_memory tool with a memory manager subagent that supports adding, removing, de-duplicating, and organizing memories.             | `false` |
+| Use the power user profile for massive contexts.     | `experimental.powerUserProfile`     | Enables continuous minimal GC near the max tokens limit instead of a blocked backbuffer.                                                                  | `false` |
 | Use the generalist profile to manage agent contexts. | `experimental.generalistProfile`    | Suitable for general coding and software development tasks.                                                                                               | `false` |
 | Enable Context Management                            | `experimental.contextManagement`    | Enable logic for context management.                                                                                                                      | `false` |
 | Topic & Update Narration                             | `experimental.topicUpdateNarration` | Enable the experimental Topic & Update communication model for reduced chattiness and structured progress reporting.                                      | `false` |

docs/reference/configuration.md

@@ -1693,6 +1693,12 @@ their corresponding top-level category object in your `settings.json` file.
   - **Default:** `false`
   - **Requires restart:** Yes
 
+- **`experimental.powerUserProfile`** (boolean):
+  - **Description:** Enables continuous minimal GC near the max tokens limit
+    instead of a blocked backbuffer.
+  - **Default:** `false`
+  - **Requires restart:** Yes
+
 - **`experimental.generalistProfile`** (boolean):
   - **Description:** Suitable for general coding and software development tasks.
   - **Default:** `false`

packages/cli/src/config/config.test.ts

@@ -21,7 +21,7 @@ import {
   type MCPServerConfig,
   type GeminiCLIExtension,
   Storage,
-  generalistProfile,
+  GENERALIST_PROFILE,
   type ContextManagementConfig,
 } from '@google/gemini-cli-core';
 import { loadCliConfig, parseArguments, type CliArgs } from './config.js';
@@ -2211,7 +2211,7 @@ describe('loadCliConfig context management', () => {
     });
     const config = await loadCliConfig(settings, 'test-session', argv);
     expect(config.getContextManagementConfig()).toStrictEqual(
-      generalistProfile,
+      GENERALIST_PROFILE,
     );
     expect(config.isContextManagementEnabled()).toBe(true);
   });
@@ -2220,24 +2220,19 @@ describe('loadCliConfig context management', () => {
     process.argv = ['node', 'script.js'];
     const argv = await parseArguments(createTestMergedSettings());
     const contextManagementConfig: Partial<ContextManagementConfig> = {
-      historyWindow: {
+      budget: {
+        maxPressureStrategy: 'truncate',
         maxTokens: 100_000,
         retainedTokens: 50_000,
-      },
-      messageLimits: {
-        normalMaxTokens: 1000,
-        retainedMaxTokens: 10_000,
-        normalizationHeadRatio: 0.25,
-      },
-      tools: {
-        distillation: {
-          maxOutputTokens: 10_000,
-          summarizationThresholdTokens: 15_000,
-        },
-        outputMasking: {
-          protectionThresholdTokens: 30_000,
-          minPrunableThresholdTokens: 10_000,
-          protectLatestTurn: false,
+        protectedEpisodes: 1,
+        protectSystemEpisode: true,
+      },
+      strategies: {
+        historySquashing: { maxTokensPerNode: 12000 },
+        toolMasking: { stringLengthThresholdTokens: 10000 },
+        semanticCompression: {
+          nodeThresholdTokens: 5000,
+          compressionModel: 'chat-compression-2.5-flash-lite',
         },
       },
     };

packages/cli/src/config/config.ts

@@ -46,7 +46,9 @@ import {
   type HookEventName,
   type OutputFormat,
   detectIdeFromEnv,
-  generalistProfile,
+  GENERALIST_PROFILE,
+  POWER_USER_PROFILE,
+  STRESS_TEST_PROFILE,
 } from '@google/gemini-cli-core';
 import {
   type Settings,
@@ -886,12 +888,17 @@ export async function loadCliConfig(
 
   const useGeneralistProfile =
     settings.experimental?.generalistProfile ?? false;
+  const useStressTestProfile = settings.experimental?.stressTestProfile ?? false;
+  const usePowerUserProfile = settings.experimental?.powerUserProfile ?? false;
   const useContextManagement =
     settings.experimental?.contextManagement ?? false;
   const contextManagement = {
-    ...(useGeneralistProfile ? generalistProfile : {}),
+    ...(useGeneralistProfile ? GENERALIST_PROFILE : {}),
+    ...(usePowerUserProfile ? POWER_USER_PROFILE : {}),
+    ...(useStressTestProfile ? STRESS_TEST_PROFILE : {}),
     ...(useContextManagement ? settings?.contextManagement : {}),
-    enabled: useContextManagement || useGeneralistProfile,
+    enabled:
+      useContextManagement || useGeneralistProfile || usePowerUserProfile || useStressTestProfile,
   };
 
   return new Config({

packages/cli/src/config/settingsSchema.ts

@@ -2149,6 +2149,26 @@ const SETTINGS_SCHEMA = {
           'Replace the built-in save_memory tool with a memory manager subagent that supports adding, removing, de-duplicating, and organizing memories.',
         showInDialog: true,
       },
+      powerUserProfile: {
+        type: 'boolean',
+        label: 'Use the power user profile for massive contexts.',
+        category: 'Experimental',
+        requiresRestart: true,
+        default: false,
+        description:
+          'Enables continuous minimal GC near the max tokens limit instead of a blocked backbuffer.',
+        showInDialog: true,
+      },
+      stressTestProfile: {
+        type: 'boolean',
+        label: 'Stress Test Profile (Context GC)',
+        category: 'Experimental',
+        requiresRestart: true,
+        default: false,
+        description:
+          'Aggressively limits the token budget (6k retained, 12k max) to force rapid background snapshotting and foreground truncations for local E2E testing of the context system.',
+        showInDialog: true,
+      },
       generalistProfile: {
         type: 'boolean',
         label: 'Use the generalist profile to manage agent contexts.',
@@ -2462,6 +2482,35 @@ const SETTINGS_SCHEMA = {
         default: {},
         showInDialog: false,
         properties: {
+          maxPressureStrategy: {
+            type: 'string',
+            label: 'Max Pressure Strategy',
+            category: 'Context Management',
+            requiresRestart: true,
+            default: 'truncate',
+            description: 'Action to take when hitting the synchronous token ceiling.',
+            showInDialog: false,
+            enum: ['truncate', 'compress', 'rollingSummarizer'],
+          },
+          gcTarget: {
+            type: 'string',
+            label: 'GC Target',
+            category: 'Context Management',
+            requiresRestart: true,
+            default: 'incremental',
+            description: 'The target floor for synchronous context shedding.',
+            showInDialog: false,
+            enum: ['incremental', 'freeNTokens', 'max'],
+          },
+          freeTokensTarget: {
+            type: 'number',
+            label: 'Free Tokens Target',
+            category: 'Context Management',
+            requiresRestart: true,
+            default: undefined,
+            description: 'Amount of tokens to free when gcTarget is freeNTokens.',
+            showInDialog: false,
+          },
           maxTokens: {
             type: 'number',
             label: 'Max Tokens',

packages/core/src/config/config.ts

@@ -11,11 +11,7 @@ import { inspect } from 'node:util';
 import process from 'node:process';
 import { z } from 'zod';
 import type { ConversationRecord } from '../services/chatRecordingService.js';
-import type {
-  AgentHistoryProviderConfig,
-  ContextManagementConfig,
-  ToolOutputMaskingConfig,
-} from '../context/types.js';
+import type { ContextManagementConfig } from '../context/types.js';
 export type { ConversationRecord };
 import {
   AuthType,
@@ -436,12 +432,6 @@ import {
   DEFAULT_MEMORY_FILE_FILTERING_OPTIONS,
   type FileFilteringOptions,
 } from './constants.js';
-import {
-  DEFAULT_TOOL_PROTECTION_THRESHOLD,
-  DEFAULT_MIN_PRUNABLE_TOKENS_THRESHOLD,
-  DEFAULT_PROTECT_LATEST_TURN,
-} from '../context/toolOutputMaskingService.js';
-
 import {
   type ExtensionLoader,
   SimpleExtensionLoader,
@@ -1139,40 +1129,30 @@ export class Config implements McpContext, AgentLoopContext {
     this.memoryBoundaryMarkers = params.memoryBoundaryMarkers ?? ['.git'];
     this.contextManagement = {
       enabled: params.contextManagement?.enabled ?? false,
-      historyWindow: {
-        maxTokens: params.contextManagement?.historyWindow?.maxTokens ?? 150000,
+      budget: {
+        maxTokens: params.contextManagement?.budget?.maxTokens ?? 150000,
         retainedTokens:
-          params.contextManagement?.historyWindow?.retainedTokens ?? 40000,
+          params.contextManagement?.budget?.retainedTokens ?? 40000,
+        gcTarget: params.contextManagement?.budget?.gcTarget ?? 'incremental',
+        freeTokensTarget: params.contextManagement?.budget?.freeTokensTarget ?? undefined,
+        maxPressureStrategy: params.contextManagement?.budget?.maxPressureStrategy ?? 'truncate',
       },
-      messageLimits: {
-        normalMaxTokens:
-          params.contextManagement?.messageLimits?.normalMaxTokens ?? 2500,
-        retainedMaxTokens:
-          params.contextManagement?.messageLimits?.retainedMaxTokens ?? 12000,
-        normalizationHeadRatio:
-          params.contextManagement?.messageLimits?.normalizationHeadRatio ??
-          0.25,
-      },
-      tools: {
-        distillation: {
-          maxOutputTokens:
-            params.contextManagement?.tools?.distillation?.maxOutputTokens ??
-            10000,
-          summarizationThresholdTokens:
-            params.contextManagement?.tools?.distillation
-              ?.summarizationThresholdTokens ?? 20000,
+      strategies: {
+        historySquashing: {
+          maxTokensPerNode:
+            params.contextManagement?.strategies?.historySquashing
+              ?.maxTokensPerNode ?? 3000,
+        },
+        toolMasking: {
+          stringLengthThresholdTokens:
+            params.contextManagement?.strategies?.toolMasking
+              ?.stringLengthThresholdTokens ?? 10000,
         },
-        outputMasking: {
-          protectionThresholdTokens:
-            params.contextManagement?.tools?.outputMasking
-              ?.protectionThresholdTokens ?? DEFAULT_TOOL_PROTECTION_THRESHOLD,
-          minPrunableThresholdTokens:
-            params.contextManagement?.tools?.outputMasking
-              ?.minPrunableThresholdTokens ??
-            DEFAULT_MIN_PRUNABLE_TOKENS_THRESHOLD,
-          protectLatestTurn:
-            params.contextManagement?.tools?.outputMasking?.protectLatestTurn ??
-            DEFAULT_PROTECT_LATEST_TURN,
+        semanticCompression: {
+          nodeThresholdTokens:
+            params.contextManagement?.strategies?.semanticCompression
+              ?.nodeThresholdTokens ?? 5000,
+
         },
       },
     };
@@ -2391,18 +2371,6 @@ export class Config implements McpContext, AgentLoopContext {
     return this.contextManagement;
   }
 
-  get agentHistoryProviderConfig(): AgentHistoryProviderConfig {
-    return {
-      maxTokens: this.contextManagement.historyWindow.maxTokens,
-      retainedTokens: this.contextManagement.historyWindow.retainedTokens,
-      normalMessageTokens: this.contextManagement.messageLimits.normalMaxTokens,
-      maximumMessageTokens:
-        this.contextManagement.messageLimits.retainedMaxTokens,
-      normalizationHeadRatio:
-        this.contextManagement.messageLimits.normalizationHeadRatio,
-    };
-  }
-
   isTopicUpdateNarrationEnabled(): boolean {
     return this.topicUpdateNarration;
   }
@@ -2411,43 +2379,6 @@ export class Config implements McpContext, AgentLoopContext {
     return this.modelSteering;
   }
 
-  async getToolOutputMaskingConfig(): Promise<ToolOutputMaskingConfig> {
-    await this.ensureExperimentsLoaded();
-
-    const remoteProtection =
-      this.experiments?.flags[ExperimentFlags.MASKING_PROTECTION_THRESHOLD]
-        ?.intValue;
-    const remotePrunable =
-      this.experiments?.flags[ExperimentFlags.MASKING_PRUNABLE_THRESHOLD]
-        ?.intValue;
-    const remoteProtectLatest =
-      this.experiments?.flags[ExperimentFlags.MASKING_PROTECT_LATEST_TURN]
-        ?.boolValue;
-
-    const parsedProtection = remoteProtection
-      ? parseInt(remoteProtection, 10)
-      : undefined;
-    const parsedPrunable = remotePrunable
-      ? parseInt(remotePrunable, 10)
-      : undefined;
-
-    return {
-      protectionThresholdTokens:
-        parsedProtection !== undefined && !isNaN(parsedProtection)
-          ? parsedProtection
-          : this.contextManagement.tools.outputMasking
-              .protectionThresholdTokens,
-      minPrunableThresholdTokens:
-        parsedPrunable !== undefined && !isNaN(parsedPrunable)
-          ? parsedPrunable
-          : this.contextManagement.tools.outputMasking
-              .minPrunableThresholdTokens,
-      protectLatestTurn:
-        remoteProtectLatest ??
-        this.contextManagement.tools.outputMasking.protectLatestTurn,
-    };
-  }
-
   getGeminiMdFileCount(): number {
     if (this.experimentalJitContext && this.memoryContextManager) {
       return this.memoryContextManager.getLoadedPaths().size;
@@ -3298,15 +3229,6 @@ export class Config implements McpContext, AgentLoopContext {
     );
   }
 
-  getToolMaxOutputTokens(): number {
-    return this.contextManagement.tools.distillation.maxOutputTokens;
-  }
-
-  getToolSummarizationThresholdTokens(): number {
-    return this.contextManagement.tools.distillation
-      .summarizationThresholdTokens;
-  }
-
   getNextCompressionTruncationId(): number {
     return ++this.compressionTruncationCounter;
   }