[Condense] Track telemetry for condense and truncate operations

Author: canrobins13

src/core/condense/index.ts

@@ -2,6 +2,7 @@ import Anthropic from "@anthropic-ai/sdk"
 import { ApiHandler } from "../../api"
 import { ApiMessage } from "../task-persistence/apiMessages"
 import { maybeRemoveImageBlocks } from "../../api/transform/image-cleaning"
+import { telemetryService } from "../../services/telemetry/TelemetryService"
 
 export const N_MESSAGES_TO_KEEP = 3
 
@@ -58,13 +59,16 @@ export type SummarizeResponse = {
  * @param {ApiMessage[]} messages - The conversation messages
  * @param {ApiHandler} apiHandler - The API handler to use for token counting.
  * @param {string} systemPrompt - The system prompt for API requests, which should be considered in the context token count
+ * @param {string} taskId - The task ID for the conversation, used for telemetry
  * @returns {SummarizeResponse} - The result of the summarization operation (see above)
  */
 export async function summarizeConversation(
 	messages: ApiMessage[],
 	apiHandler: ApiHandler,
 	systemPrompt: string,
+	taskId: string,
 ): Promise<SummarizeResponse> {
+	telemetryService.captureContextCondensed(taskId)
 	const response: SummarizeResponse = { messages, cost: 0, summary: "" }
 	const messagesToSummarize = getMessagesSinceLastSummary(messages.slice(0, -N_MESSAGES_TO_KEEP))
 	if (messagesToSummarize.length <= 1) {

src/core/sliding-window/index.ts

@@ -2,6 +2,7 @@ import { Anthropic } from "@anthropic-ai/sdk"
 import { ApiHandler } from "../../api"
 import { summarizeConversation, SummarizeResponse } from "../condense"
 import { ApiMessage } from "../task-persistence/apiMessages"
+import { telemetryService } from "../../services/telemetry/TelemetryService"
 
 /**
  * Default percentage of the context window to use as a buffer when deciding when to truncate
@@ -31,9 +32,11 @@ export async function estimateTokenCount(
  *
  * @param {ApiMessage[]} messages - The conversation messages.
  * @param {number} fracToRemove - The fraction (between 0 and 1) of messages (excluding the first) to remove.
+ * @param {string} taskId - The task ID for the conversation, used for telemetry
  * @returns {ApiMessage[]} The truncated conversation messages.
  */
-export function truncateConversation(messages: ApiMessage[], fracToRemove: number): ApiMessage[] {
+export function truncateConversation(messages: ApiMessage[], fracToRemove: number, taskId: string): ApiMessage[] {
+	telemetryService.captureSlidingWindowTruncation(taskId)
 	const truncatedMessages = [messages[0]]
 	const rawMessagesToRemove = Math.floor((messages.length - 1) * fracToRemove)
 	const messagesToRemove = rawMessagesToRemove - (rawMessagesToRemove % 2)
@@ -66,6 +69,7 @@ type TruncateOptions = {
 	autoCondenseContext: boolean
 	autoCondenseContextPercent: number
 	systemPrompt: string
+	taskId: string
 }
 
 type TruncateResponse = SummarizeResponse & { prevContextTokens: number }
@@ -86,6 +90,7 @@ export async function truncateConversationIfNeeded({
 	autoCondenseContext,
 	autoCondenseContextPercent,
 	systemPrompt,
+	taskId,
 }: TruncateOptions): Promise<TruncateResponse> {
 	// Calculate the maximum tokens reserved for response
 	const reservedTokens = maxTokens || contextWindow * 0.2
@@ -108,7 +113,7 @@ export async function truncateConversationIfNeeded({
 		const contextPercent = (100 * prevContextTokens) / contextWindow
 		if (contextPercent >= autoCondenseContextPercent || prevContextTokens > allowedTokens) {
 			// Attempt to intelligently condense the context
-			const result = await summarizeConversation(messages, apiHandler, systemPrompt)
+			const result = await summarizeConversation(messages, apiHandler, systemPrompt, taskId)
 			if (result.summary) {
 				return { ...result, prevContextTokens }
 			}
@@ -117,7 +122,7 @@ export async function truncateConversationIfNeeded({
 
 	// Fall back to sliding window truncation if needed
 	if (prevContextTokens > allowedTokens) {
-		const truncatedMessages = truncateConversation(messages, 0.5)
+		const truncatedMessages = truncateConversation(messages, 0.5, taskId)
 		return { messages: truncatedMessages, prevContextTokens, summary: "", cost: 0 }
 	}
 	// No truncation or condensation needed

src/core/task/Task.ts

@@ -488,7 +488,7 @@ export class Task extends EventEmitter<ClineEvents> {
 			summary,
 			cost,
 			newContextTokens = 0,
-		} = await summarizeConversation(this.apiConversationHistory, this.api, systemPrompt)
+		} = await summarizeConversation(this.apiConversationHistory, this.api, systemPrompt, this.taskId)
 		if (!summary) {
 			return
 		}
@@ -1518,6 +1518,7 @@ export class Task extends EventEmitter<ClineEvents> {
 				autoCondenseContext,
 				autoCondenseContextPercent,
 				systemPrompt,
+				taskId: this.taskId,
 			})
 			if (truncateResult.messages !== this.apiConversationHistory) {
 				await this.overwriteApiConversationHistory(truncateResult.messages)

src/services/telemetry/PostHogClient.ts

@@ -28,6 +28,8 @@ export class PostHogClient {
 			CHECKPOINT_DIFFED: "Checkpoint Diffed",
 			CODE_ACTION_USED: "Code Action Used",
 			PROMPT_ENHANCED: "Prompt Enhanced",
+			CONTEXT_CONDENSED: "Context Condensed",
+			SLIDING_WINDOW_TRUNCATION: "Sliding Window Truncation",
 		},
 		ERRORS: {
 			SCHEMA_VALIDATION_ERROR: "Schema Validation Error",

src/services/telemetry/TelemetryService.ts

@@ -120,6 +120,14 @@ class TelemetryService {
 		this.captureEvent(PostHogClient.EVENTS.TASK.CHECKPOINT_RESTORED, { taskId })
 	}
 
+	public captureContextCondensed(taskId: string): void {
+		this.captureEvent(PostHogClient.EVENTS.TASK.CONTEXT_CONDENSED, { taskId })
+	}
+
+	public captureSlidingWindowTruncation(taskId: string): void {
+		this.captureEvent(PostHogClient.EVENTS.TASK.SLIDING_WINDOW_TRUNCATION, { taskId })
+	}
+
 	public captureCodeActionUsed(actionType: string): void {
 		this.captureEvent(PostHogClient.EVENTS.TASK.CODE_ACTION_USED, { actionType })
 	}