[Condense] Add a slider for the context condense threshold

.changeset/whole-swans-cheer.md

@@ -0,0 +1,5 @@
+---
+"roo-cline": patch
+---
+
+Adds a slider to configure threshold to trigger intelligent context condensing

src/core/sliding-window/__tests__/sliding-window.test.ts

@@ -248,6 +248,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 
@@ -277,6 +279,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 
@@ -304,6 +308,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo1.contextWindow,
 			maxTokens: modelInfo1.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 
@@ -313,6 +319,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo2.contextWindow,
 			maxTokens: modelInfo2.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 
@@ -329,6 +337,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo1.contextWindow,
 			maxTokens: modelInfo1.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 
@@ -338,6 +348,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo2.contextWindow,
 			maxTokens: modelInfo2.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 
@@ -369,6 +381,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(resultWithSmall).toEqual({
@@ -399,6 +413,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(resultWithLarge.messages).not.toEqual(messagesWithLargeContent) // Should truncate
@@ -422,6 +438,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(resultWithVeryLarge.messages).not.toEqual(messagesWithVeryLargeContent) // Should truncate
@@ -448,6 +466,8 @@ describe("truncateConversationIfNeeded", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(result).toEqual({
@@ -488,6 +508,7 @@ describe("truncateConversationIfNeeded", () => {
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
 			autoCondenseContext: true,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 
@@ -534,6 +555,7 @@ describe("truncateConversationIfNeeded", () => {
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
 			autoCondenseContext: true,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 
@@ -570,6 +592,7 @@ describe("truncateConversationIfNeeded", () => {
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
 			autoCondenseContext: false,
+			autoCondenseContextPercent: 50, // This shouldn't matter since autoCondenseContext is false
 			systemPrompt: "System prompt",
 		})
 
@@ -587,6 +610,94 @@ describe("truncateConversationIfNeeded", () => {
 		// Clean up
 		summarizeSpy.mockRestore()
 	})
+
+	it("should use summarizeConversation when autoCondenseContext is true and context percent exceeds threshold", async () => {
+		// Mock the summarizeConversation function
+		const mockSummary = "This is a summary of the conversation"
+		const mockCost = 0.05
+		const mockSummarizeResponse: condenseModule.SummarizeResponse = {
+			messages: [
+				{ role: "user", content: "First message" },
+				{ role: "assistant", content: mockSummary, isSummary: true },
+				{ role: "user", content: "Last message" },
+			],
+			summary: mockSummary,
+			cost: mockCost,
+			newContextTokens: 100,
+		}
+
+		const summarizeSpy = jest
+			.spyOn(condenseModule, "summarizeConversation")
+			.mockResolvedValue(mockSummarizeResponse)
+
+		const modelInfo = createModelInfo(100000, 30000)
+		// Set tokens to be below the allowedTokens threshold but above the percentage threshold
+		const contextWindow = modelInfo.contextWindow
+		const totalTokens = 60000 // Below allowedTokens but 60% of context window
+		const messagesWithSmallContent = [...messages.slice(0, -1), { ...messages[messages.length - 1], content: "" }]
+
+		const result = await truncateConversationIfNeeded({
+			messages: messagesWithSmallContent,
+			totalTokens,
+			contextWindow,
+			maxTokens: modelInfo.maxTokens,
+			apiHandler: mockApiHandler,
+			autoCondenseContext: true,
+			autoCondenseContextPercent: 50, // Set threshold to 50% - our tokens are at 60%
+			systemPrompt: "System prompt",
+		})
+
+		// Verify summarizeConversation was called with the right parameters
+		expect(summarizeSpy).toHaveBeenCalledWith(messagesWithSmallContent, mockApiHandler, "System prompt")
+
+		// Verify the result contains the summary information
+		expect(result).toMatchObject({
+			messages: mockSummarizeResponse.messages,
+			summary: mockSummary,
+			cost: mockCost,
+			prevContextTokens: totalTokens,
+		})
+
+		// Clean up
+		summarizeSpy.mockRestore()
+	})
+
+	it("should not use summarizeConversation when autoCondenseContext is true but context percent is below threshold", async () => {
+		// Reset any previous mock calls
+		jest.clearAllMocks()
+		const summarizeSpy = jest.spyOn(condenseModule, "summarizeConversation")
+
+		const modelInfo = createModelInfo(100000, 30000)
+		// Set tokens to be below both the allowedTokens threshold and the percentage threshold
+		const contextWindow = modelInfo.contextWindow
+		const totalTokens = 40000 // 40% of context window
+		const messagesWithSmallContent = [...messages.slice(0, -1), { ...messages[messages.length - 1], content: "" }]
+
+		const result = await truncateConversationIfNeeded({
+			messages: messagesWithSmallContent,
+			totalTokens,
+			contextWindow,
+			maxTokens: modelInfo.maxTokens,
+			apiHandler: mockApiHandler,
+			autoCondenseContext: true,
+			autoCondenseContextPercent: 50, // Set threshold to 50% - our tokens are at 40%
+			systemPrompt: "System prompt",
+		})
+
+		// Verify summarizeConversation was not called
+		expect(summarizeSpy).not.toHaveBeenCalled()
+
+		// Verify no truncation or summarization occurred
+		expect(result).toEqual({
+			messages: messagesWithSmallContent,
+			summary: "",
+			cost: 0,
+			prevContextTokens: totalTokens,
+		})
+
+		// Clean up
+		summarizeSpy.mockRestore()
+	})
 })
 
 /**
@@ -624,6 +735,8 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(result1).toEqual({
@@ -640,6 +753,8 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(result2.messages).not.toEqual(messagesWithSmallContent)
@@ -664,6 +779,8 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(result1).toEqual({
@@ -680,6 +797,8 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(result2.messages).not.toEqual(messagesWithSmallContent)
@@ -703,6 +822,8 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(result1.messages).toEqual(messagesWithSmallContent)
@@ -714,6 +835,8 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(result2).not.toEqual(messagesWithSmallContent)
@@ -735,6 +858,8 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(result1.messages).toEqual(messagesWithSmallContent)
@@ -746,6 +871,8 @@ describe("getMaxTokens", () => {
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 			apiHandler: mockApiHandler,
+			autoCondenseContext: false,
+			autoCondenseContextPercent: 100,
 			systemPrompt: "System prompt",
 		})
 		expect(result2).not.toEqual(messagesWithSmallContent)

src/core/sliding-window/index.ts

@@ -63,7 +63,8 @@ type TruncateOptions = {
 	contextWindow: number
 	maxTokens?: number | null
 	apiHandler: ApiHandler
-	autoCondenseContext?: boolean
+	autoCondenseContext: boolean
+	autoCondenseContextPercent: number
 	systemPrompt: string
 }
 
@@ -83,6 +84,7 @@ export async function truncateConversationIfNeeded({
 	maxTokens,
 	apiHandler,
 	autoCondenseContext,
+	autoCondenseContextPercent,
 	systemPrompt,
 }: TruncateOptions): Promise<TruncateResponse> {
 	// Calculate the maximum tokens reserved for response
@@ -96,21 +98,28 @@ export async function truncateConversationIfNeeded({
 		: await estimateTokenCount([{ type: "text", text: lastMessageContent as string }], apiHandler)
 
 	// Calculate total effective tokens (totalTokens never includes the last message)
-	const effectiveTokens = totalTokens + lastMessageTokens
+	const prevContextTokens = totalTokens + lastMessageTokens
 
 	// Calculate available tokens for conversation history
 	// Truncate if we're within TOKEN_BUFFER_PERCENTAGE of the context window
 	const allowedTokens = contextWindow * (1 - TOKEN_BUFFER_PERCENTAGE) - reservedTokens
 
-	// Determine if truncation is needed and apply if necessary
-	if (effectiveTokens <= allowedTokens) {
-		return { messages, summary: "", cost: 0, prevContextTokens: effectiveTokens }
-	} else if (autoCondenseContext) {
-		const result = await summarizeConversation(messages, apiHandler, systemPrompt)
-		if (result.summary) {
-			return { ...result, prevContextTokens: effectiveTokens }
+	if (autoCondenseContext) {
+		const contextPercent = (100 * prevContextTokens) / contextWindow
+		if (contextPercent >= autoCondenseContextPercent || prevContextTokens > allowedTokens) {
+			// Attempt to intelligently condense the context
+			const result = await summarizeConversation(messages, apiHandler, systemPrompt)
+			if (result.summary) {
+				return { ...result, prevContextTokens }
+			}
 		}
 	}
-	const truncatedMessages = truncateConversation(messages, 0.5)
-	return { messages: truncatedMessages, prevContextTokens: effectiveTokens, summary: "", cost: 0 }
+
+	// Fall back to sliding window truncation if needed
+	if (prevContextTokens > allowedTokens) {
+		const truncatedMessages = truncateConversation(messages, 0.5)
+		return { messages: truncatedMessages, prevContextTokens, summary: "", cost: 0 }
+	}
+	// No truncation or condensation needed
+	return { messages, summary: "", cost: 0, prevContextTokens }
 }

src/core/task/Task.ts

@@ -1460,8 +1460,14 @@ export class Task extends EventEmitter<ClineEvents> {
 	}
 
 	public async *attemptApiRequest(retryAttempt: number = 0): ApiStream {
-		const { apiConfiguration, autoApprovalEnabled, alwaysApproveResubmit, requestDelaySeconds, experiments } =
-			(await this.providerRef.deref()?.getState()) ?? {}
+		const {
+			apiConfiguration,
+			autoApprovalEnabled,
+			alwaysApproveResubmit,
+			requestDelaySeconds,
+			experiments,
+			autoCondenseContextPercent = 100,
+		} = (await this.providerRef.deref()?.getState()) ?? {}
 
 		let rateLimitDelay = 0
 
@@ -1510,6 +1516,7 @@ export class Task extends EventEmitter<ClineEvents> {
 				contextWindow,
 				apiHandler: this.api,
 				autoCondenseContext,
+				autoCondenseContextPercent,
 				systemPrompt,
 			})
 			if (truncateResult.messages !== this.apiConversationHistory) {

src/core/webview/ClineProvider.ts

@@ -1222,6 +1222,7 @@ export class ClineProvider extends EventEmitter<ClineProviderEvents> implements
 			alwaysAllowModeSwitch,
 			alwaysAllowSubtasks,
 			allowedMaxRequests,
+			autoCondenseContextPercent,
 			soundEnabled,
 			ttsEnabled,
 			ttsSpeed,
@@ -1293,6 +1294,7 @@ export class ClineProvider extends EventEmitter<ClineProviderEvents> implements
 			alwaysAllowModeSwitch: alwaysAllowModeSwitch ?? false,
 			alwaysAllowSubtasks: alwaysAllowSubtasks ?? false,
 			allowedMaxRequests: allowedMaxRequests ?? Infinity,
+			autoCondenseContextPercent: autoCondenseContextPercent ?? 100,
 			uriScheme: vscode.env.uriScheme,
 			currentTaskItem: this.getCurrentCline()?.taskId
 				? (taskHistory || []).find((item: HistoryItem) => item.id === this.getCurrentCline()?.taskId)
@@ -1396,6 +1398,7 @@ export class ClineProvider extends EventEmitter<ClineProviderEvents> implements
 			alwaysAllowModeSwitch: stateValues.alwaysAllowModeSwitch ?? false,
 			alwaysAllowSubtasks: stateValues.alwaysAllowSubtasks ?? false,
 			allowedMaxRequests: stateValues.allowedMaxRequests ?? Infinity,
+			autoCondenseContextPercent: stateValues.autoCondenseContextPercent ?? 100,
 			taskHistory: stateValues.taskHistory,
 			allowedCommands: stateValues.allowedCommands,
 			soundEnabled: stateValues.soundEnabled ?? false,