use observeOpenAI

- Add Langfuse client, otel, and tracing packages as dependencies
- Update OpenTelemetry packages to version 0.202.0 for compatibility
- Add Langfuse configuration options to .env.example
- Enable advanced LLM observability through Langfuse integration

Note: This  is a POC only and only basic llm call tracing info for each task is implemented. The span will not be available until the task is completed.

.env.example

@@ -91,6 +91,13 @@ POSTHOG_TELEMETRY_ENABLED=true          # Enable PostHog telemetry (default: tru
 #   OTEL_EXPORTER_OTLP_ENDPOINT=https://otel.example.com
 #   OTEL_EXPORTER_OTLP_HEADERS=authorization=Bearer your-token
 
+# ============================================================================
+# Langfuse Integration (Optional - for advanced LLM tracing)
+# ============================================================================
+# LANGFUSE_SECRET_KEY = "sk-lf-..."
+# LANGFUSE_PUBLIC_KEY = "pk-lf-..."
+# LANGFUSE_BASE_URL = "https://cloud.langfuse.com" # Default to "https://us.cloud.langfuse.com"
+
 # ============================================================================
 # OPTIONAL DEVELOPMENT SETTINGS
 # ============================================================================

package.json

@@ -462,26 +462,30 @@
 		"@google/genai": "^1.30.0",
 		"@grpc/grpc-js": "^1.9.15",
 		"@grpc/reflection": "^1.0.4",
+		"@langfuse/client": "^4.4.9",
+		"@langfuse/openai": "^4.4.9",
+		"@langfuse/otel": "^4.4.9",
+		"@langfuse/tracing": "^4.4.9",
 		"@mistralai/mistralai": "^1.5.0",
 		"@modelcontextprotocol/sdk": "^1.11.1",
 		"@opentelemetry/api": "^1.9.0",
 		"@opentelemetry/core": "^2.1.0",
-		"@opentelemetry/exporter-logs-otlp-grpc": "^0.56.0",
-		"@opentelemetry/exporter-logs-otlp-http": "^0.56.0",
-		"@opentelemetry/exporter-logs-otlp-proto": "^0.56.0",
-		"@opentelemetry/exporter-metrics-otlp-grpc": "^0.56.0",
-		"@opentelemetry/exporter-metrics-otlp-http": "^0.56.0",
-		"@opentelemetry/exporter-metrics-otlp-proto": "^0.56.0",
-		"@opentelemetry/exporter-prometheus": "^0.56.0",
-		"@opentelemetry/exporter-trace-otlp-http": "^0.56.0",
+		"@opentelemetry/exporter-logs-otlp-grpc": "^0.202.0",
+		"@opentelemetry/exporter-logs-otlp-http": "^0.202.0",
+		"@opentelemetry/exporter-logs-otlp-proto": "^0.202.0",
+		"@opentelemetry/exporter-metrics-otlp-grpc": "^0.202.0",
+		"@opentelemetry/exporter-metrics-otlp-http": "^0.202.0",
+		"@opentelemetry/exporter-metrics-otlp-proto": "^0.202.0",
+		"@opentelemetry/exporter-prometheus": "^0.202.0",
+		"@opentelemetry/exporter-trace-otlp-http": "^0.202.0",
 		"@opentelemetry/instrumentation": "^0.205.0",
 		"@opentelemetry/instrumentation-http": "^0.205.0",
-		"@opentelemetry/resources": "^1.30.1",
-		"@opentelemetry/sdk-logs": "^0.56.0",
-		"@opentelemetry/sdk-metrics": "^1.30.1",
-		"@opentelemetry/sdk-node": "^0.56.0",
+		"@opentelemetry/resources": "^2.1.0",
+		"@opentelemetry/sdk-logs": "^0.202.0",
+		"@opentelemetry/sdk-metrics": "^2.1.0",
+		"@opentelemetry/sdk-node": "^0.202.0",
 		"@opentelemetry/sdk-trace-base": "^2.1.0",
-		"@opentelemetry/sdk-trace-node": "^1.30.1",
+		"@opentelemetry/sdk-trace-node": "^2.1.0",
 		"@opentelemetry/semantic-conventions": "^1.37.0",
 		"@playwright/test": "^1.55.1",
 		"@sap-ai-sdk/ai-api": "^2.1.0",

src/core/api/index.ts

@@ -1,3 +1,4 @@
+import { LangfuseSpan } from "@langfuse/tracing"
 import { ApiConfiguration, ModelInfo, QwenApiRegions } from "@shared/api"
 import { Mode } from "@shared/storage/types"
 import { ClineStorageMessage } from "@/shared/messages/content"
@@ -48,7 +49,7 @@ export type CommonApiHandlerOptions = {
 	onRetryAttempt?: ApiConfiguration["onRetryAttempt"]
 }
 export interface ApiHandler {
-	createMessage(systemPrompt: string, messages: ClineStorageMessage[], tools?: ClineTool[], useResponseApi?: boolean): ApiStream
+	createMessage(systemPrompt: string, messages: ClineStorageMessage[], tools?: ClineTool[], span?: LangfuseSpan): ApiStream
 	getModel(): ApiHandlerModel
 	getApiStreamUsage?(): Promise<ApiStreamUsageChunk | undefined>
 	abort?(): void

src/core/api/providers/cline.ts

@@ -1,3 +1,5 @@
+import { observeOpenAI } from "@langfuse/openai"
+import { LangfuseSpan, updateActiveTrace } from "@langfuse/tracing"
 import { ModelInfo, openRouterDefaultModelId, openRouterDefaultModelInfo } from "@shared/api"
 import { shouldSkipReasoningForModel } from "@utils/model-utils"
 import axios from "axios"
@@ -98,20 +100,38 @@ export class ClineHandler implements ApiHandler {
 	}
 
 	@withRetry()
-	async *createMessage(systemPrompt: string, messages: ClineStorageMessage[], tools?: OpenAITool[]): ApiStream {
+	async *createMessage(
+		systemPrompt: string,
+		messages: ClineStorageMessage[],
+		tools?: OpenAITool[],
+		span?: LangfuseSpan,
+	): ApiStream {
 		try {
-			const client = await this.ensureClient()
-
 			this.lastGenerationId = undefined
 			this.lastRequestId = undefined
 
 			let didOutputUsage: boolean = false
+			const openaiClient = await this.ensureClient()
+			const client = observeOpenAI(openaiClient, {
+				generationName: "cline",
+				sessionId: this.options.ulid,
+			})
+			const model = this.getModel()
+			const toolCallProcessor = new ToolCallProcessor()
+
+			const lastUserContent = messages.at(-1)?.content
+			const input =
+				typeof lastUserContent === "string"
+					? lastUserContent
+					: lastUserContent?.map((m) => (m.type === "text" ? m.text : "")).join("\n\n")
+
+			span?.update({ input, metadata: { model: model.id } })
 
 			const stream = await createOpenRouterStream(
 				client,
 				systemPrompt,
 				messages,
-				this.getModel(),
+				model,
 				this.options.reasoningEffort,
 				this.options.thinkingBudgetTokens,
 				this.options.openRouterProviderSorting,
@@ -119,8 +139,6 @@ export class ClineHandler implements ApiHandler {
 				this.options.geminiThinkingLevel,
 			)
 
-			const toolCallProcessor = new ToolCallProcessor()
-
 			for await (const chunk of stream) {
 				Logger.debug("ClineHandler chunk:" + JSON.stringify(chunk))
 				// openrouter returns an error object instead of the openai sdk throwing an error
@@ -203,12 +221,15 @@ export class ClineHandler implements ApiHandler {
 						totalCost = 0
 					}
 
+					const inputTokens = (chunk.usage.prompt_tokens || 0) - (chunk.usage.prompt_tokens_details?.cached_tokens || 0)
+					const outputTokens = chunk.usage.completion_tokens || 0
+
 					yield {
 						type: "usage",
 						cacheWriteTokens: 0,
 						cacheReadTokens: chunk.usage.prompt_tokens_details?.cached_tokens || 0,
-						inputTokens: (chunk.usage.prompt_tokens || 0) - (chunk.usage.prompt_tokens_details?.cached_tokens || 0),
-						outputTokens: chunk.usage.completion_tokens || 0,
+						inputTokens,
+						outputTokens,
 						totalCost,
 					}
 					didOutputUsage = true
@@ -226,6 +247,11 @@ export class ClineHandler implements ApiHandler {
 		} catch (error) {
 			console.error("Cline API Error:", error)
 			throw error
+		} finally {
+			updateActiveTrace({
+				userId: "test",
+				sessionId: this.options.ulid,
+			})
 		}
 	}
 
@@ -249,14 +275,20 @@ export class ClineHandler implements ApiHandler {
 				})
 
 				const generation = response.data
+
+				const totalCost = generation?.total_cost || 0
+				const inputTokens =
+					(generation.usage.prompt_tokens || 0) - (generation.usage.prompt_tokens_details?.cached_tokens || 0)
+				const outputTokens = generation.usage.completion_tokens || 0
+
 				return {
 					type: "usage",
 					cacheWriteTokens: 0,
 					cacheReadTokens: generation?.native_tokens_cached || 0,
 					// openrouter generation endpoint fails often
-					inputTokens: (generation?.native_tokens_prompt || 0) - (generation?.native_tokens_cached || 0),
-					outputTokens: generation?.native_tokens_completion || 0,
-					totalCost: generation?.total_cost || 0,
+					inputTokens,
+					outputTokens,
+					totalCost,
 				}
 			} catch (error) {
 				// ignore if fails

src/core/api/providers/openai-native.ts

@@ -1,3 +1,4 @@
+import { observeOpenAI } from "@langfuse/openai"
 import {
 	ModelInfo,
 	OpenAiCompatibleModelInfo,
@@ -86,7 +87,10 @@ export class OpenAiNativeHandler implements ApiHandler {
 		messages: ClineStorageMessage[],
 		tools?: ChatCompletionTool[],
 	): ApiStream {
-		const client = this.ensureClient()
+		const openaiClient = this.ensureClient()
+		const client = observeOpenAI(openaiClient, {
+			generationName: "openai-native",
+		})
 		const model = this.getModel()
 		const toolCallProcessor = new ToolCallProcessor()
 

src/core/api/providers/vercel-ai-gateway.ts

@@ -1,3 +1,4 @@
+import { observeOpenAI } from "@langfuse/openai"
 import { ModelInfo, openRouterDefaultModelId, openRouterDefaultModelInfo } from "@shared/api"
 import OpenAI from "openai"
 import type { ChatCompletionTool as OpenAITool } from "openai/resources/chat/completions"
@@ -48,7 +49,10 @@ export class VercelAIGatewayHandler implements ApiHandler {
 
 	@withRetry()
 	async *createMessage(systemPrompt: string, messages: ClineStorageMessage[], tools?: OpenAITool[]): ApiStream {
-		const client = this.ensureClient()
+		const openaiClient = this.ensureClient()
+		const client = observeOpenAI(openaiClient, {
+			generationName: "vercel-ai-gateway",
+		})
 		const modelId = this.getModel().id
 		const modelInfo = this.getModel().info
 

src/core/task/index.ts

@@ -44,6 +44,7 @@ import { formatContentBlockToMarkdown } from "@integrations/misc/export-markdown
 import { processFilesIntoText } from "@integrations/misc/extract-text"
 import { showSystemNotification } from "@integrations/notifications"
 import { ITerminalManager } from "@integrations/terminal/types"
+import { startObservation } from "@langfuse/tracing"
 import { BrowserSession } from "@services/browser/BrowserSession"
 import { UrlContentFetcher } from "@services/browser/UrlContentFetcher"
 import { featureFlagsService } from "@services/feature-flags"
@@ -135,6 +136,7 @@ export class Task {
 	private taskIsFavorited?: boolean
 	private cwd: string
 	private taskInitializationStartTime: number
+	private span
 
 	taskState: TaskState
 
@@ -332,6 +334,8 @@ export class Task {
 			updateTaskHistory: this.updateTaskHistory,
 		})
 
+		this.span = startObservation("task")
+
 		// Initialize context trackers
 		this.fileContextTracker = new FileContextTracker(controller, this.taskId)
 		this.modelContextTracker = new ModelContextTracker(this.taskId)
@@ -1324,6 +1328,7 @@ export class Task {
 		let includeFileDetails = true
 		while (!this.taskState.abort) {
 			const didEndLoop = await this.recursivelyMakeClineRequests(nextUserContent, includeFileDetails)
+
 			includeFileDetails = false // we only need file details the first time
 
 			//  The way this agentic loop works is that cline will be given a task that he then calls tools to complete. unless there's an attempt_completion call, we keep responding back to him with his tool's responses until he either attempt_completion or does not use anymore tools. If he does not use anymore tools, we ask him to consider if he's completed the task and then call attempt_completion, otherwise proceed with completing the task.
@@ -1814,7 +1819,12 @@ export class Task {
 		}
 
 		// Response API requires native tool calls to be enabled
-		const stream = this.api.createMessage(systemPrompt, contextManagementMetadata.truncatedConversationHistory, tools)
+		const stream = this.api.createMessage(
+			systemPrompt,
+			contextManagementMetadata.truncatedConversationHistory,
+			tools,
+			this.span,
+		)
 
 		const iterator = stream[Symbol.asyncIterator]()
 
@@ -1967,6 +1977,7 @@ export class Task {
 				// Reset the automatic retry flag so the request can proceed
 				this.taskState.didAutomaticallyRetryFailedApiRequest = false
 			}
+
 			// delegate generator output from the recursive call
 			yield* this.attemptApiRequest(previousApiReqIndex)
 			return
@@ -2521,6 +2532,7 @@ export class Task {
 							taskMetrics.cacheWriteTokens += chunk.cacheWriteTokens ?? 0
 							taskMetrics.cacheReadTokens += chunk.cacheReadTokens ?? 0
 							taskMetrics.totalCost = chunk.totalCost ?? taskMetrics.totalCost
+							this.span.update({ output: assistantTextOnly })
 							break
 						case "reasoning": {
 							// Process the reasoning delta through the handler
@@ -2676,6 +2688,7 @@ export class Task {
 					await this.reinitExistingTaskFromId(this.taskId)
 				}
 			} finally {
+				this.span.end()
 				this.taskState.isStreaming = false
 			}
 

src/extension.ts

@@ -44,6 +44,7 @@ import { ExtensionRegistryInfo } from "./registry"
 import { AuthService } from "./services/auth/AuthService"
 import { LogoutReason } from "./services/auth/types"
 import { telemetryService } from "./services/telemetry"
+import { langfuse } from "./services/trace/langfuse"
 import { SharedUriHandler } from "./services/uri/SharedUriHandler"
 import { ShowMessageType } from "./shared/proto/host/window"
 import { fileExistsAtPath } from "./utils/fs"
@@ -431,6 +432,8 @@ export async function activate(context: vscode.ExtensionContext) {
 function setupHostProvider(context: ExtensionContext) {
 	console.log("Setting up vscode host providers...")
 
+	langfuse.start()
+
 	const createWebview = () => new VscodeWebviewProvider(context)
 	const createDiffView = () => new VscodeDiffViewProvider()
 	const createCommentReview = () => getVscodeCommentReviewController()

src/services/telemetry/providers/opentelemetry/OpenTelemetryClientProvider.ts

@@ -1,6 +1,6 @@
 import { metrics } from "@opentelemetry/api"
 import { logs } from "@opentelemetry/api-logs"
-import { Resource } from "@opentelemetry/resources"
+import { defaultResource, type Resource, resourceFromAttributes } from "@opentelemetry/resources"
 import { BatchLogRecordProcessor, LoggerProvider } from "@opentelemetry/sdk-logs"
 import { MeterProvider } from "@opentelemetry/sdk-metrics"
 import { ATTR_SERVICE_NAME, ATTR_SERVICE_VERSION } from "@opentelemetry/semantic-conventions"
@@ -80,10 +80,12 @@ export class OpenTelemetryClientProvider {
 		}
 
 		// Create resource with service information
-		const resource = new Resource({
-			[ATTR_SERVICE_NAME]: "cline",
-			[ATTR_SERVICE_VERSION]: ExtensionRegistryInfo.version,
-		})
+		const resource = defaultResource().merge(
+			resourceFromAttributes({
+				[ATTR_SERVICE_NAME]: "cline",
+				[ATTR_SERVICE_VERSION]: ExtensionRegistryInfo.version,
+			}),
+		)
 
 		// Initialize metrics if configured
 		if (this.config.metricsExporter) {

src/services/trace/langfuse.ts

@@ -0,0 +1,13 @@
+import { LangfuseClient } from "@langfuse/client"
+import { LangfuseSpanProcessor } from "@langfuse/otel"
+import { NodeSDK } from "@opentelemetry/sdk-node"
+
+// TODO: Convert this into a singleton pattern that can be configured with different providers apart from Langfuse
+
+new LangfuseClient({
+	publicKey: process.env.LANGFUSE_PUBLIC_KEY,
+	secretKey: process.env.LANGFUSE_SECRET_KEY,
+	baseUrl: process.env.LANGFUSE_BASE_URL || "https://us.cloud.langfuse.com",
+})
+
+export const langfuse = new NodeSDK({ spanProcessors: [new LangfuseSpanProcessor()] })