diff --git a/src/cli/App.tsx b/src/cli/App.tsx
index 1da9267..70146ec 100644
--- a/src/cli/App.tsx
+++ b/src/cli/App.tsx
@@ -25,6 +25,7 @@ import type { ApprovalContext } from "../permissions/analyzer";
 import { permissionMode } from "../permissions/mode";
 import { updateProjectSettings } from "../settings";
 import { settingsManager } from "../settings-manager";
+import { telemetry } from "../telemetry";
 import type { ToolExecutionResult } from "../tools/manager";
 import {
   analyzeToolApproval,
@@ -314,6 +315,7 @@ export default function App({
   const agentIdRef = useRef(agentId);
   useEffect(() => {
     agentIdRef.current = agentId;
+    telemetry.setCurrentAgentId(agentId);
   }, [agentId]);
 
   const resumeKey = useSuspend();
@@ -473,6 +475,18 @@ export default function App({
   // Session stats tracking
   const sessionStatsRef = useRef(new SessionStats());
 
+  // Wire up session stats to telemetry for safety net handlers
+  useEffect(() => {
+    telemetry.setSessionStatsGetter(() =>
+      sessionStatsRef.current.getSnapshot(),
+    );
+
+    // Cleanup on unmount (defensive, prevents potential memory leak)
+    return () => {
+      telemetry.setSessionStatsGetter(undefined);
+    };
+  }, []);
+
   // Show exit stats on exit (double Ctrl+C)
   const [showExitStats, setShowExitStats] = useState(false);
 
@@ -1409,6 +1423,25 @@ export default function App({
           return;
         }
 
+        // Track error with enhanced context
+        const errorType =
+          e instanceof Error ? e.constructor.name : "UnknownError";
+        const errorMessage = e instanceof Error ? e.message : String(e);
+
+        // Extract HTTP status code if available (API errors often have this)
+        const httpStatus =
+          e &&
+          typeof e === "object" &&
+          "status" in e &&
+          typeof e.status === "number"
+            ? e.status
+            : undefined;
+
+        telemetry.trackError(errorType, errorMessage, "message_stream", {
+          httpStatus,
+          modelId: currentModelId || undefined,
+        });
+
         // Use comprehensive error formatting
         const errorDetails = formatErrorDetails(e, agentIdRef.current);
         appendError(errorDetails);
@@ -1418,11 +1451,25 @@ export default function App({
         abortControllerRef.current = null;
       }
     },
-    [appendError, refreshDerived, refreshDerivedThrottled, setStreaming],
+    [
+      appendError,
+      refreshDerived,
+      refreshDerivedThrottled,
+      setStreaming,
+      currentModelId,
+    ],
   );
 
-  const handleExit = useCallback(() => {
+  const handleExit = useCallback(async () => {
     saveLastAgentBeforeExit();
+
+    // Track session end explicitly (before exit) with stats
+    const stats = sessionStatsRef.current.getSnapshot();
+    telemetry.trackSessionEnd(stats, "exit_command");
+
+    // Flush telemetry before exit
+    await telemetry.flush();
+
     setShowExitStats(true);
     // Give React time to render the stats, then exit
     setTimeout(() => {
@@ -1678,6 +1725,9 @@ export default function App({
 
       if (!msg) return { submitted: false };
 
+      // Track user input (agent_id automatically added from telemetry.currentAgentId)
+      telemetry.trackUserInput(msg, "user", currentModelId || "unknown");
+
       // Block submission if waiting for explicit user action (approvals)
       // In this case, input is hidden anyway, so this shouldn't happen
       if (pendingApprovals.length > 0) {
@@ -2000,6 +2050,13 @@ export default function App({
 
             saveLastAgentBeforeExit();
 
+            // Track session end explicitly (before exit) with stats
+            const stats = sessionStatsRef.current.getSnapshot();
+            telemetry.trackSessionEnd(stats, "logout");
+
+            // Flush telemetry before exit
+            await telemetry.flush();
+
             // Exit after a brief delay to show the message
             setTimeout(() => process.exit(0), 500);
           } catch (error) {
diff --git a/src/index.ts b/src/index.ts
index 11f1b5a..9fcf695 100755
--- a/src/index.ts
+++ b/src/index.ts
@@ -8,6 +8,7 @@ import type { AgentProvenance } from "./agent/create";
 import { LETTA_CLOUD_API_URL } from "./auth/oauth";
 import { permissionMode } from "./permissions/mode";
 import { settingsManager } from "./settings-manager";
+import { telemetry } from "./telemetry";
 import {
   forceUpsertTools,
   isToolsNotFoundError,
@@ -214,6 +215,9 @@ async function main(): Promise<void> {
   await settingsManager.initialize();
   const settings = settingsManager.getSettings();
 
+  // Initialize telemetry (enabled by default, opt-out via LETTA_CODE_TELEM=0)
+  telemetry.init();
+
   // Check for updates on startup (non-blocking)
   const { checkAndAutoUpdate } = await import("./updater/auto-update");
   checkAndAutoUpdate().catch(() => {
diff --git a/src/telemetry/index.ts b/src/telemetry/index.ts
new file mode 100644
index 0000000..f449307
--- /dev/null
+++ b/src/telemetry/index.ts
@@ -0,0 +1,433 @@
+export interface TelemetryEvent {
+  type: "session_start" | "session_end" | "tool_usage" | "error" | "user_input";
+  timestamp: string;
+  data: Record<string, unknown>;
+}
+
+export interface SessionStartData {
+  startup_command: string;
+  version: string;
+  platform: string;
+  node_version: string;
+}
+
+export interface SessionEndData {
+  duration: number; // in seconds
+  message_count: number;
+  tool_call_count: number;
+  exit_reason?: string; // e.g., "exit_command", "logout", "sigint", "process_exit"
+  total_api_ms?: number;
+  total_wall_ms?: number;
+  prompt_tokens?: number;
+  completion_tokens?: number;
+  total_tokens?: number;
+  cached_tokens?: number;
+  reasoning_tokens?: number;
+  step_count?: number;
+}
+
+export interface ToolUsageData {
+  tool_name: string;
+  success: boolean;
+  duration: number;
+  response_length?: number;
+  error_type?: string;
+  stderr?: string;
+}
+
+export interface ErrorData {
+  error_type: string;
+  error_message: string;
+  context?: string;
+  http_status?: number;
+  model_id?: string;
+}
+
+export interface UserInputData {
+  input_length: number;
+  is_command: boolean;
+  command_name?: string;
+  message_type: string;
+  model_id: string;
+}
+
+class TelemetryManager {
+  private events: TelemetryEvent[] = [];
+  private sessionId: string;
+  private currentAgentId: string | null = null;
+  private sessionStartTime: number;
+  private messageCount = 0;
+  private toolCallCount = 0;
+  private sessionEndTracked = false;
+  private flushInterval: NodeJS.Timeout | null = null;
+  private readonly FLUSH_INTERVAL_MS = 5 * 60 * 1000; // 5 minutes
+  private readonly MAX_BATCH_SIZE = 100;
+  private sessionStatsGetter?: () => {
+    totalWallMs: number;
+    totalApiMs: number;
+    usage: {
+      promptTokens: number;
+      completionTokens: number;
+      totalTokens: number;
+      cachedTokens: number;
+      reasoningTokens: number;
+      stepCount: number;
+    };
+  };
+
+  constructor() {
+    this.sessionId = this.generateSessionId();
+    this.sessionStartTime = Date.now();
+  }
+
+  private generateSessionId(): string {
+    return `${Date.now()}-${Math.random().toString(36).substring(2, 15)}`;
+  }
+
+  /**
+   * Check if telemetry is enabled based on LETTA_CODE_TELEM env var
+   * Enabled by default unless explicitly disabled or using self-hosted server
+   */
+  private isTelemetryEnabled(): boolean {
+    // Check environment variable - must be explicitly set to "0" or "false" to disable
+    const envValue = process.env.LETTA_CODE_TELEM;
+    if (envValue === "0" || envValue === "false") {
+      return false;
+    }
+
+    // Disable telemetry if using self-hosted server (not api.letta.com)
+    const baseURL = process.env.LETTA_BASE_URL;
+    if (baseURL && !baseURL.includes("api.letta.com")) {
+      return false;
+    }
+
+    // Disable telemetry if no API key is set
+    const apiKey = process.env.LETTA_API_KEY;
+    if (!apiKey) {
+      return false;
+    }
+
+    return true;
+  }
+
+  /**
+   * Initialize telemetry and start periodic flushing
+   */
+  init() {
+    if (!this.isTelemetryEnabled()) {
+      return;
+    }
+
+    this.trackSessionStart();
+
+    // Set up periodic flushing
+    this.flushInterval = setInterval(() => {
+      this.flush().catch((err) => {
+        // Silently fail - we don't want telemetry to interfere with user experience
+        if (process.env.LETTA_DEBUG) {
+          console.error("Telemetry flush error:", err);
+        }
+      });
+    }, this.FLUSH_INTERVAL_MS);
+
+    // Don't let the interval prevent process from exiting
+    this.flushInterval.unref();
+
+    // Safety net: Handle Ctrl+C interruption
+    // Note: Normal exits via handleExit flush explicitly
+    process.on("SIGINT", () => {
+      try {
+        this.trackSessionEnd(undefined, "sigint");
+        // Fire and forget - try to flush but don't wait (might not complete)
+        this.flush().catch(() => {
+          // Silently ignore
+        });
+      } catch {
+        // Silently ignore - don't prevent process from exiting
+      }
+      // Exit immediately - don't wait for flush
+      process.exit(0);
+    });
+
+    // TODO: Add telemetry for crashes and abnormal exits
+    // Current limitation: We can't reliably flush telemetry on process.on("exit")
+    // because the event loop is shut down and async operations don't work.
+    // Potential solution: Write unsent events to ~/.letta/telemetry-queue.json
+    // and send them on next startup. This would capture crash telemetry without
+    // risking hangs on exit.
+  }
+
+  /**
+   * Track a telemetry event
+   */
+  private track(
+    type: TelemetryEvent["type"],
+    data:
+      | Record<string, unknown>
+      | SessionStartData
+      | SessionEndData
+      | ToolUsageData
+      | ErrorData
+      | UserInputData,
+  ) {
+    if (!this.isTelemetryEnabled()) {
+      return;
+    }
+
+    const event: TelemetryEvent = {
+      type,
+      timestamp: new Date().toISOString(),
+      data: {
+        ...data,
+        session_id: this.sessionId,
+        agent_id: this.currentAgentId || undefined,
+      },
+    };
+
+    this.events.push(event);
+
+    // Flush if batch size is reached
+    if (this.events.length >= this.MAX_BATCH_SIZE) {
+      this.flush().catch((err) => {
+        if (process.env.LETTA_DEBUG) {
+          console.error("Telemetry flush error:", err);
+        }
+      });
+    }
+  }
+
+  /**
+   * Set the current agent ID (called from App.tsx when agent changes)
+   * This is automatically added to all telemetry events
+   */
+  setCurrentAgentId(agentId: string | null) {
+    this.currentAgentId = agentId;
+  }
+
+  /**
+   * Set a getter function for session stats (called from App.tsx)
+   * This allows safety net handlers to access stats even if not explicitly passed
+   * Pass undefined to clear the getter (for cleanup)
+   */
+  setSessionStatsGetter(
+    getter?: () => {
+      totalWallMs: number;
+      totalApiMs: number;
+      usage: {
+        promptTokens: number;
+        completionTokens: number;
+        totalTokens: number;
+        cachedTokens: number;
+        reasoningTokens: number;
+        stepCount: number;
+      };
+    },
+  ) {
+    this.sessionStatsGetter = getter;
+  }
+
+  /**
+   * Track session start
+   */
+  trackSessionStart() {
+    // Extract agent ID from startup args if --agent or -a is provided
+    const args = process.argv.slice(2);
+    const agentFlagIndex = args.findIndex(
+      (arg) => arg === "--agent" || arg === "-a",
+    );
+    if (agentFlagIndex !== -1 && agentFlagIndex + 1 < args.length) {
+      const agentId = args[agentFlagIndex + 1];
+      if (agentId) {
+        this.currentAgentId = agentId;
+      }
+    }
+
+    const data: SessionStartData = {
+      startup_command: args.join(" "),
+      version: process.env.npm_package_version || "unknown",
+      platform: process.platform,
+      node_version: process.version,
+    };
+    this.track("session_start", data);
+  }
+
+  /**
+   * Track session end
+   * @param stats Optional session stats (from sessionStatsRef.current.getSnapshot() in App.tsx)
+   * @param exitReason Optional reason for exit (e.g., "exit_command", "logout", "sigint", "process_exit")
+   */
+  trackSessionEnd(
+    stats?: {
+      totalWallMs: number;
+      totalApiMs: number;
+      usage: {
+        promptTokens: number;
+        completionTokens: number;
+        totalTokens: number;
+        cachedTokens: number;
+        reasoningTokens: number;
+        stepCount: number;
+      };
+    },
+    exitReason?: string,
+  ) {
+    // Prevent double-tracking (can be called from both handleExit and process.on("exit"))
+    if (this.sessionEndTracked) {
+      return;
+    }
+    this.sessionEndTracked = true;
+
+    // Try to get stats from getter if not provided (for safety net handlers)
+    let sessionStats = stats;
+    if (!sessionStats && this.sessionStatsGetter) {
+      try {
+        sessionStats = this.sessionStatsGetter();
+      } catch {
+        // Ignore errors - stats will be undefined
+      }
+    }
+
+    const duration = Math.floor((Date.now() - this.sessionStartTime) / 1000);
+    const data: SessionEndData = {
+      duration,
+      message_count: this.messageCount,
+      tool_call_count: this.toolCallCount,
+      exit_reason: exitReason,
+      // Include optional stats if available
+      total_api_ms: sessionStats?.totalApiMs,
+      total_wall_ms: sessionStats?.totalWallMs,
+      prompt_tokens: sessionStats?.usage.promptTokens,
+      completion_tokens: sessionStats?.usage.completionTokens,
+      total_tokens: sessionStats?.usage.totalTokens,
+      cached_tokens: sessionStats?.usage.cachedTokens,
+      reasoning_tokens: sessionStats?.usage.reasoningTokens,
+      step_count: sessionStats?.usage.stepCount,
+    };
+    this.track("session_end", data);
+  }
+
+  /**
+   * Track tool usage
+   */
+  trackToolUsage(
+    toolName: string,
+    success: boolean,
+    duration: number,
+    responseLength?: number,
+    errorType?: string,
+    stderr?: string,
+  ) {
+    this.toolCallCount++;
+    const data: ToolUsageData = {
+      tool_name: toolName,
+      success,
+      duration,
+      response_length: responseLength,
+      error_type: errorType,
+      stderr,
+    };
+    this.track("tool_usage", data);
+  }
+
+  /**
+   * Track errors
+   */
+  trackError(
+    errorType: string,
+    errorMessage: string,
+    context?: string,
+    options?: {
+      httpStatus?: number;
+      modelId?: string;
+    },
+  ) {
+    const data: ErrorData = {
+      error_type: errorType,
+      error_message: errorMessage,
+      context,
+      http_status: options?.httpStatus,
+      model_id: options?.modelId,
+    };
+    this.track("error", data);
+  }
+
+  /**
+   * Track user input
+   * Note: agent_id is automatically added from currentAgentId
+   */
+  trackUserInput(input: string, messageType: string, modelId: string) {
+    this.messageCount++;
+
+    const isCommand = input.trim().startsWith("/");
+    const commandName = isCommand ? input.trim().split(/\s+/)[0] : undefined;
+
+    const data: UserInputData = {
+      input_length: input.length,
+      is_command: isCommand,
+      command_name: commandName,
+      message_type: messageType,
+      model_id: modelId,
+    };
+    this.track("user_input", data);
+  }
+
+  /**
+   * Flush events to the server
+   */
+  async flush(): Promise<void> {
+    if (this.events.length === 0 || !this.isTelemetryEnabled()) {
+      return;
+    }
+
+    const eventsToSend = [...this.events];
+    this.events = [];
+
+    const baseURL = process.env.LETTA_BASE_URL || "https://api.letta.com";
+    const apiKey = process.env.LETTA_API_KEY;
+
+    try {
+      // Add 5 second timeout to prevent telemetry from blocking shutdown
+      const timeoutPromise = new Promise((_, reject) =>
+        setTimeout(() => reject(new Error("Telemetry request timeout")), 5000),
+      );
+
+      const fetchPromise = fetch(`${baseURL}/v1/metadata/telemetry`, {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          Authorization: `Bearer ${apiKey}`,
+          "X-Letta-Source": "letta-code",
+        },
+        body: JSON.stringify({
+          service: "letta-code",
+          events: eventsToSend,
+        }),
+      });
+
+      const response = (await Promise.race([
+        fetchPromise,
+        timeoutPromise,
+      ])) as Response;
+
+      if (!response.ok) {
+        throw new Error(`Telemetry flush failed: ${response.status}`);
+      }
+    } catch {
+      // If flush fails, put events back in queue, but don't throw error
+      this.events.unshift(...eventsToSend);
+    }
+  }
+
+  /**
+   * Clean up resources
+   */
+  cleanup() {
+    if (this.flushInterval) {
+      clearInterval(this.flushInterval);
+      this.flushInterval = null;
+    }
+  }
+}
+
+// Export singleton instance
+export const telemetry = new TelemetryManager();
diff --git a/src/tools/manager.ts b/src/tools/manager.ts
index f211fb0..4a838bc 100644
--- a/src/tools/manager.ts
+++ b/src/tools/manager.ts
@@ -6,6 +6,7 @@ import {
 } from "@letta-ai/letta-client";
 import { getModelInfo } from "../agent/model";
 import { getAllSubagentConfigs } from "../agent/subagents";
+import { telemetry } from "../telemetry";
 import { TOOL_DEFINITIONS, type ToolName } from "./toolDefinitions";
 
 export const TOOL_NAMES = Object.keys(TOOL_DEFINITIONS) as ToolName[];
@@ -899,6 +900,8 @@ export async function executeTool(
     };
   }
 
+  const startTime = Date.now();
+
   try {
     // Inject options for tools that support them without altering schemas
     let enhancedArgs = args;
@@ -914,6 +917,7 @@ export async function executeTool(
     }
 
     const result = await tool.fn(enhancedArgs);
+    const duration = Date.now() - startTime;
 
     // Extract stdout/stderr if present (for bash tools)
     const recordResult = isRecord(result) ? result : undefined;
@@ -924,6 +928,16 @@ export async function executeTool(
     // Flatten the response to plain text
     const flattenedResponse = flattenToolResponse(result);
 
+    // Track tool usage (success path - we're in the try block)
+    telemetry.trackToolUsage(
+      internalName,
+      true, // Hardcoded to true since tool execution succeeded
+      duration,
+      flattenedResponse.length,
+      undefined, // no error_type on success
+      stderr ? stderr.join("\n") : undefined,
+    );
+
     // Return the full response (truncation happens in UI layer only)
     return {
       toolReturn: flattenedResponse,
@@ -932,24 +946,38 @@ export async function executeTool(
       ...(stderr && { stderr }),
     };
   } catch (error) {
+    const duration = Date.now() - startTime;
     const isAbort =
       error instanceof Error &&
       (error.name === "AbortError" ||
         error.message === "The operation was aborted" ||
         // node:child_process AbortError may include code/message variants
         ("code" in error && error.code === "ABORT_ERR"));
+    const errorType = isAbort
+      ? "abort"
+      : error instanceof Error
+        ? error.name
+        : "unknown";
+    const errorMessage = isAbort
+      ? "User interrupted tool execution"
+      : error instanceof Error
+        ? error.message
+        : String(error);
 
-    if (isAbort) {
-      return {
-        toolReturn: "User interrupted tool execution",
-        status: "error",
-      };
-    }
+    // Track tool usage error
+    telemetry.trackToolUsage(
+      internalName,
+      false,
+      duration,
+      errorMessage.length,
+      errorType,
+      errorMessage,
+    );
 
     // Don't console.error here - it pollutes the TUI
     // The error message is already returned in toolReturn
     return {
-      toolReturn: error instanceof Error ? error.message : String(error),
+      toolReturn: errorMessage,
       status: "error",
     };
   }