Add a context percentage threshold setting for auto compression (#5721)

packages/core/src/config/config.ts

@@ -69,6 +69,10 @@ export interface BugCommandSettings {
   urlTemplate: string;
 }
 
+export interface ChatCompressionSettings {
+  contextPercentageThreshold?: number;
+}
+
 export interface SummarizeToolOutputSettings {
   tokenBudget?: number;
 }
@@ -191,6 +195,7 @@ export interface ConfigParameters {
   folderTrustFeature?: boolean;
   ideMode?: boolean;
   loadMemoryFromIncludeDirectories?: boolean;
+  chatCompression?: ChatCompressionSettings;
 }
 
 export class Config {
@@ -252,6 +257,7 @@ export class Config {
     | undefined;
   private readonly experimentalAcp: boolean = false;
   private readonly loadMemoryFromIncludeDirectories: boolean = false;
+  private readonly chatCompression: ChatCompressionSettings | undefined;
 
   constructor(params: ConfigParameters) {
     this.sessionId = params.sessionId;
@@ -316,6 +322,7 @@ export class Config {
     }
     this.loadMemoryFromIncludeDirectories =
       params.loadMemoryFromIncludeDirectories ?? false;
+    this.chatCompression = params.chatCompression;
 
     if (params.contextFileName) {
       setGeminiMdFilename(params.contextFileName);
@@ -667,6 +674,10 @@ export class Config {
     return this.ideClient;
   }
 
+  getChatCompression(): ChatCompressionSettings | undefined {
+    return this.chatCompression;
+  }
+
   async getGitService(): Promise<GitService> {
     if (!this.gitService) {
       this.gitService = new GitService(this.targetDir);

packages/core/src/core/client.test.ts

@@ -206,6 +206,7 @@ describe('Gemini Client (client.ts)', () => {
       }),
       getGeminiClient: vi.fn(),
       setFallbackMode: vi.fn(),
+      getChatCompression: vi.fn().mockReturnValue(undefined),
     };
     const MockedConfig = vi.mocked(Config, true);
     MockedConfig.mockImplementation(
@@ -531,14 +532,19 @@ describe('Gemini Client (client.ts)', () => {
       expect(newChat).toBe(initialChat);
     });
 
-    it('should trigger summarization if token count is at threshold', async () => {
+    it('should trigger summarization if token count is at threshold with contextPercentageThreshold setting', async () => {
       const MOCKED_TOKEN_LIMIT = 1000;
+      const MOCKED_CONTEXT_PERCENTAGE_THRESHOLD = 0.5;
       vi.mocked(tokenLimit).mockReturnValue(MOCKED_TOKEN_LIMIT);
+      vi.spyOn(client['config'], 'getChatCompression').mockReturnValue({
+        contextPercentageThreshold: MOCKED_CONTEXT_PERCENTAGE_THRESHOLD,
+      });
       mockGetHistory.mockReturnValue([
         { role: 'user', parts: [{ text: '...history...' }] },
       ]);
 
-      const originalTokenCount = 1000 * 0.7;
+      const originalTokenCount =
+        MOCKED_TOKEN_LIMIT * MOCKED_CONTEXT_PERCENTAGE_THRESHOLD;
       const newTokenCount = 100;
 
       mockCountTokens

packages/core/src/core/client.ts

@@ -596,12 +596,16 @@ export class GeminiClient {
       return null;
     }
 
+    const contextPercentageThreshold =
+      this.config.getChatCompression()?.contextPercentageThreshold;
+
     // Don't compress if not forced and we are under the limit.
-    if (
-      !force &&
-      originalTokenCount < this.COMPRESSION_TOKEN_THRESHOLD * tokenLimit(model)
-    ) {
-      return null;
+    if (!force) {
+      const threshold =
+        contextPercentageThreshold ?? this.COMPRESSION_TOKEN_THRESHOLD;
+      if (originalTokenCount < threshold * tokenLimit(model)) {
+        return null;
+      }
     }
 
     let compressBeforeIndex = findIndexAfterFraction(