condense if we reach the threshold

Author: canrobins13

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) {