@bike4mind/cli 0.2.29-feat-session-permission.18841 → 0.2.29-feat-cli-websocket-streaming.18841

package/dist/{artifactExtractor-EE753J4D.js → artifactExtractor-SNWW3EVH.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   CurationArtifactType
-} from "./chunk-5GSDPBTM.js";
+} from "./chunk-KPOK6V6E.js";
 
 // ../../b4m-core/packages/services/dist/src/notebookCurationService/artifactExtractor.js
 var ARTIFACT_TAG_REGEX = /<artifact\s+(.*?)>([\s\S]*?)<\/artifact>/gi;

package/dist/{chunk-PKKJ44C5.js → chunk-BXCARBUQ.js}

@@ -2,7 +2,7 @@
 import {
   BadRequestError,
   secureParameters
-} from "./chunk-S7BPPS33.js";
+} from "./chunk-TY6Z2UV2.js";
 import {
   CompletionApiUsageTransaction,
   GenericCreditDeductTransaction,
@@ -12,7 +12,7 @@ import {
   TextGenerationUsageTransaction,
   TransferCreditTransaction,
   VideoGenerationUsageTransaction
-} from "./chunk-5GSDPBTM.js";
+} from "./chunk-KPOK6V6E.js";
 
 // ../../b4m-core/packages/services/dist/src/creditService/subtractCredits.js
 import { z } from "zod";

package/dist/{chunk-ELTTBWAQ.js → chunk-EC2VEDD2.js}

@@ -7,11 +7,11 @@ import {
   getSettingsMap,
   getSettingsValue,
   secureParameters
-} from "./chunk-S7BPPS33.js";
+} from "./chunk-TY6Z2UV2.js";
 import {
   KnowledgeType,
   SupportedFabFileMimeTypes
-} from "./chunk-5GSDPBTM.js";
+} from "./chunk-KPOK6V6E.js";
 
 // ../../b4m-core/packages/services/dist/src/fabFileService/create.js
 import { z } from "zod";

package/dist/{chunk-5GSDPBTM.js → chunk-KPOK6V6E.js}

@@ -1012,6 +1012,29 @@ var VoiceSessionSendTranscriptAction = z10.object({
   conversationItemId: z10.string(),
   timestamp: z10.coerce.date().optional()
 });
+var CliCompletionRequestAction = z10.object({
+  action: z10.literal("cli_completion_request"),
+  accessToken: z10.string(),
+  requestId: z10.string().uuid(),
+  model: z10.string(),
+  messages: z10.array(z10.object({
+    role: z10.enum(["user", "assistant", "system"]),
+    content: z10.union([z10.string(), z10.array(z10.unknown())])
+  })),
+  options: z10.object({
+    temperature: z10.number().optional(),
+    maxTokens: z10.number().optional(),
+    stream: z10.boolean().optional(),
+    tools: z10.array(z10.unknown()).optional()
+  }).optional()
+});
+var CliToolRequestAction = z10.object({
+  action: z10.literal("cli_tool_request"),
+  accessToken: z10.string(),
+  requestId: z10.string().uuid(),
+  toolName: z10.string(),
+  input: z10.record(z10.unknown())
+});
 var VoiceSessionEndedAction = z10.object({
   action: z10.literal("voice_session_ended"),
   userId: z10.string(),
@@ -1258,6 +1281,36 @@ var ResearchModeStreamAction = z10.object({
     completionInfo: z10.any().optional()
   }).optional()
 });
+var CliCompletionChunkAction = z10.object({
+  action: z10.literal("cli_completion_chunk"),
+  requestId: z10.string(),
+  chunk: z10.object({
+    type: z10.enum(["content", "tool_use"]),
+    text: z10.string(),
+    tools: z10.array(z10.unknown()).optional(),
+    usage: z10.object({
+      inputTokens: z10.number().optional(),
+      outputTokens: z10.number().optional()
+    }).optional(),
+    thinking: z10.array(z10.unknown()).optional()
+  })
+});
+var CliCompletionDoneAction = z10.object({
+  action: z10.literal("cli_completion_done"),
+  requestId: z10.string()
+});
+var CliCompletionErrorAction = z10.object({
+  action: z10.literal("cli_completion_error"),
+  requestId: z10.string(),
+  error: z10.string()
+});
+var CliToolResponseAction = z10.object({
+  action: z10.literal("cli_tool_response"),
+  requestId: z10.string(),
+  success: z10.boolean(),
+  content: z10.unknown().optional(),
+  error: z10.string().optional()
+});
 var SessionCreatedAction = shareableDocumentSchema.extend({
   action: z10.literal("session.created"),
   id: z10.string(),
@@ -1292,7 +1345,9 @@ var MessageDataToServer = z10.discriminatedUnion("action", [
   DataUnsubscribeRequestAction,
   HeartbeatAction,
   VoiceSessionSendTranscriptAction,
-  VoiceSessionEndedAction
+  VoiceSessionEndedAction,
+  CliCompletionRequestAction,
+  CliToolRequestAction
 ]);
 var MessageDataToClient = z10.discriminatedUnion("action", [
   DataSubscriptionUpdateAction,
@@ -1314,7 +1369,11 @@ var MessageDataToClient = z10.discriminatedUnion("action", [
   SpiderProgressUpdateAction,
   SpiderCompleteAction,
   SpiderErrorAction,
-  SessionCreatedAction
+  SessionCreatedAction,
+  CliCompletionChunkAction,
+  CliCompletionDoneAction,
+  CliCompletionErrorAction,
+  CliToolResponseAction
 ]);
 
 // ../../b4m-core/packages/common/dist/src/schemas/cliCompletions.js
@@ -7926,6 +7985,8 @@ export {
   DataUnsubscribeRequestAction,
   HeartbeatAction,
   VoiceSessionSendTranscriptAction,
+  CliCompletionRequestAction,
+  CliToolRequestAction,
   VoiceSessionEndedAction,
   DataSubscriptionUpdateAction,
   LLMStatusUpdateAction,
@@ -7946,6 +8007,10 @@ export {
   UpdateResearchTaskStatusAction,
   ImportHistoryJobProgressUpdateAction,
   ResearchModeStreamAction,
+  CliCompletionChunkAction,
+  CliCompletionDoneAction,
+  CliCompletionErrorAction,
+  CliToolResponseAction,
   SessionCreatedAction,
   MessageDataToServer,
   MessageDataToClient,

package/dist/{chunk-BCOGDPUB.js → chunk-QQL3OBKL.js}

@@ -6,12 +6,12 @@ import {
   getSettingsByNames,
   obfuscateApiKey,
   secureParameters
-} from "./chunk-S7BPPS33.js";
+} from "./chunk-TY6Z2UV2.js";
 import {
   ApiKeyType,
   MementoTier,
   isSupportedEmbeddingModel
-} from "./chunk-5GSDPBTM.js";
+} from "./chunk-KPOK6V6E.js";
 
 // ../../b4m-core/packages/services/dist/src/apiKeyService/get.js
 import { z } from "zod";

package/dist/{chunk-S7BPPS33.js → chunk-TY6Z2UV2.js}

@@ -15,7 +15,7 @@ import {
   dayjsConfig_default,
   extractSnippetMeta,
   settingsMap
-} from "./chunk-5GSDPBTM.js";
+} from "./chunk-KPOK6V6E.js";
 import {
   Logger
 } from "./chunk-OCYRD7D6.js";

package/dist/{create-O3PVSKFE.js → create-QOJPAIOR.js}

@@ -2,9 +2,9 @@
 import {
   createFabFile,
   createFabFileSchema
-} from "./chunk-ELTTBWAQ.js";
-import "./chunk-S7BPPS33.js";
-import "./chunk-5GSDPBTM.js";
+} from "./chunk-EC2VEDD2.js";
+import "./chunk-TY6Z2UV2.js";
+import "./chunk-KPOK6V6E.js";
 import "./chunk-OCYRD7D6.js";
 export {
   createFabFile,

package/dist/index.js

@@ -5,7 +5,7 @@ import {
   getEffectiveApiKey,
   getOpenWeatherKey,
   getSerperKey
-} from "./chunk-BCOGDPUB.js";
+} from "./chunk-QQL3OBKL.js";
 import {
   ConfigStore,
   logger
@@ -14,8 +14,8 @@ import {
   selectActiveBackgroundAgents,
   useCliStore
 } from "./chunk-TVW4ZESU.js";
-import "./chunk-PKKJ44C5.js";
-import "./chunk-ELTTBWAQ.js";
+import "./chunk-BXCARBUQ.js";
+import "./chunk-EC2VEDD2.js";
 import {
   BFLImageService,
   BaseStorage,
@@ -27,7 +27,7 @@ import {
   OpenAIBackend,
   OpenAIImageService,
   XAIImageService
-} from "./chunk-S7BPPS33.js";
+} from "./chunk-TY6Z2UV2.js";
 import {
   AiEvents,
   ApiKeyEvents,
@@ -83,7 +83,7 @@ import {
   XAI_IMAGE_MODELS,
   b4mLLMTools,
   getMcpProviderMetadata
-} from "./chunk-5GSDPBTM.js";
+} from "./chunk-KPOK6V6E.js";
 import {
   Logger
 } from "./chunk-OCYRD7D6.js";
@@ -93,7 +93,7 @@ import React21, { useState as useState9, useEffect as useEffect6, useCallback as
 import { render, Box as Box20, Text as Text20, useApp, useInput as useInput9 } from "ink";
 import { execSync } from "child_process";
 import { randomBytes as randomBytes5 } from "crypto";
-import { v4 as uuidv411 } from "uuid";
+import { v4 as uuidv413 } from "uuid";
 
 // src/components/App.tsx
 import React15, { useState as useState5, useEffect as useEffect4 } from "react";
@@ -1260,12 +1260,10 @@ function PermissionPrompt({
 }) {
   const items = canBeTrusted ? [
     { label: "\u2713 Allow once", value: "allow-once" },
-    { label: "\u2713 Allow for this session", value: "allow-session" },
     { label: "\u2713 Always allow (trust this tool)", value: "allow-always" },
     { label: "\u2717 Deny", value: "deny" }
   ] : [
     { label: "\u2713 Allow once", value: "allow-once" },
-    { label: "\u2713 Allow for this session", value: "allow-session" },
     { label: "\u2717 Deny", value: "deny" }
   ];
   const [selectedIndex, setSelectedIndex] = useState3(0);
@@ -11440,6 +11438,10 @@ var ServerToolExecutor = class {
 };
 
 // src/llm/ToolRouter.ts
+var wsToolExecutor = null;
+function setWebSocketToolExecutor(executor) {
+  wsToolExecutor = executor;
+}
 var SERVER_TOOLS = ["weather_info", "web_search", "web_fetch"];
 var LOCAL_TOOLS = [
   "file_read",
@@ -11462,7 +11464,15 @@ function isLocalTool(toolName) {
 }
 async function executeTool(toolName, input, apiClient, localToolFn) {
   if (isServerTool(toolName)) {
-    logger.debug(`[ToolRouter] Routing ${toolName} to server`);
+    if (wsToolExecutor) {
+      logger.debug(`[ToolRouter] Routing ${toolName} to server via WebSocket`);
+      const result = await wsToolExecutor.execute(toolName, input);
+      if (!result.success) {
+        return `Error executing ${toolName}: ${result.error || "Tool execution failed"}`;
+      }
+      return typeof result.content === "string" ? result.content : JSON.stringify(result.content ?? "");
+    }
+    logger.debug(`[ToolRouter] Routing ${toolName} to server via HTTP`);
     const executor = new ServerToolExecutor(apiClient);
     return await executor.executeTool(toolName, input);
   } else if (isLocalTool(toolName)) {
@@ -11766,9 +11776,6 @@ function wrapToolWithPermission(tool, permissionManager, showPermissionPrompt, a
       if (response.action === "deny") {
         throw new PermissionDeniedError(toolName, args);
       }
-      if (response.action === "allow-session") {
-        permissionManager.trustToolForSession(toolName);
-      }
       if (response.action === "allow-always") {
         const canTrust = permissionManager.trustTool(toolName);
         if (canTrust) {
@@ -11973,7 +11980,6 @@ function generateCliTools(userId, llm, model, permissionManager, showPermissionP
 var PermissionManager = class {
   constructor(trustedTools = [], customCategories, deniedTools) {
     this.trustedTools = /* @__PURE__ */ new Set();
-    this.sessionTrustedTools = /* @__PURE__ */ new Set();
     this.deniedTools = /* @__PURE__ */ new Set();
     this.trustedTools = new Set(trustedTools);
     this.customCategories = new Map(Object.entries(customCategories || {}));
@@ -11999,9 +12005,6 @@ var PermissionManager = class {
     if (category === "auto_approve") {
       return false;
     }
-    if (this.sessionTrustedTools.has(toolName)) {
-      return false;
-    }
     if (category === "prompt_always") {
       return true;
     }
@@ -12069,29 +12072,6 @@ var PermissionManager = class {
   getDeniedTools() {
     return Array.from(this.deniedTools).sort();
   }
-  /**
-   * Trust a tool for the current session only (in-memory, no persistence)
-   * Works for all tool categories including prompt_always, but NOT project-denied tools
-   */
-  trustToolForSession(toolName) {
-    if (this.deniedTools.has(toolName)) {
-      return false;
-    }
-    this.sessionTrustedTools.add(toolName);
-    return true;
-  }
-  /**
-   * Check if a tool is trusted for the current session
-   */
-  isSessionTrusted(toolName) {
-    return this.sessionTrustedTools.has(toolName);
-  }
-  /**
-   * Clear all session-scoped trust (called on exit or session reset)
-   */
-  clearSessionTrust() {
-    this.sessionTrustedTools.clear();
-  }
   /**
    * Clear all trusted tools
    */
@@ -13422,6 +13402,186 @@ var ServerLlmBackend = class {
   }
 };
 
+// src/llm/WebSocketLlmBackend.ts
+import { v4 as uuidv411 } from "uuid";
+function stripThinkingBlocks2(text) {
+  return text.replace(/<think>[\s\S]*?<\/think>/g, "").trim();
+}
+var WebSocketLlmBackend = class {
+  constructor(options) {
+    this.wsManager = options.wsManager;
+    this.apiClient = options.apiClient;
+    this.currentModel = options.model;
+    this.tokenGetter = options.tokenGetter;
+  }
+  /**
+   * Make LLM completion request via WebSocket.
+   * Mirrors ServerLlmBackend accumulation logic — collects all chunks,
+   * calls callback once at completion with full accumulated content.
+   */
+  async complete(model, messages, options, callback) {
+    logger.debug(`[WebSocketLlmBackend] Starting complete() with model: ${model}`);
+    if (options.abortSignal?.aborted) {
+      logger.debug("[WebSocketLlmBackend] Request aborted before start");
+      return;
+    }
+    if (!this.wsManager.isConnected) {
+      throw new Error("WebSocket is not connected");
+    }
+    const requestId = uuidv411();
+    const token = await this.tokenGetter();
+    if (!token) {
+      throw new Error("No access token available");
+    }
+    return new Promise((resolve3, reject) => {
+      const isVerbose = process.env.B4M_VERBOSE === "1";
+      const isUltraVerbose = process.env.B4M_DEBUG_STREAM === "1";
+      const streamLogger = new StreamLogger(logger, "WebSocketLlmBackend", isVerbose, isUltraVerbose);
+      streamLogger.streamStart();
+      let eventCount = 0;
+      let accumulatedText = "";
+      let lastUsageInfo = {};
+      let toolsUsed = [];
+      let thinkingBlocks = [];
+      let settled = false;
+      const cleanup = () => {
+        this.wsManager.offRequest(requestId);
+      };
+      const settleResolve = () => {
+        if (settled) return;
+        settled = true;
+        cleanup();
+        resolve3();
+      };
+      const settleReject = (err) => {
+        if (settled) return;
+        settled = true;
+        cleanup();
+        reject(err);
+      };
+      if (options.abortSignal) {
+        const abortHandler = () => {
+          logger.debug("[WebSocketLlmBackend] Abort signal received");
+          settleResolve();
+        };
+        if (options.abortSignal.aborted) {
+          settleResolve();
+          return;
+        }
+        options.abortSignal.addEventListener("abort", abortHandler, { once: true });
+      }
+      this.wsManager.onRequest(requestId, (message) => {
+        if (options.abortSignal?.aborted) return;
+        const action = message.action;
+        if (action === "cli_completion_chunk") {
+          eventCount++;
+          const chunk = message.chunk;
+          streamLogger.onEvent(eventCount, JSON.stringify(chunk));
+          if (chunk.type === "content") {
+            const textChunk = chunk.text || "";
+            accumulatedText += textChunk;
+            streamLogger.onContent(eventCount, textChunk, accumulatedText);
+            if (chunk.usage) {
+              lastUsageInfo = {
+                inputTokens: chunk.usage.inputTokens,
+                outputTokens: chunk.usage.outputTokens
+              };
+            }
+          } else if (chunk.type === "tool_use") {
+            streamLogger.onCriticalEvent(eventCount, "TOOL_USE", `tools: ${chunk.tools?.length}`);
+            const textChunk = chunk.text || "";
+            if (textChunk) accumulatedText += textChunk;
+            if (chunk.tools && chunk.tools.length > 0) {
+              toolsUsed = chunk.tools;
+            }
+            if (chunk.thinking && chunk.thinking.length > 0) {
+              thinkingBlocks = chunk.thinking;
+            }
+            if (chunk.usage) {
+              lastUsageInfo = {
+                inputTokens: chunk.usage.inputTokens,
+                outputTokens: chunk.usage.outputTokens
+              };
+            }
+          }
+        } else if (action === "cli_completion_done") {
+          streamLogger.streamComplete(accumulatedText);
+          const cleanedText = stripThinkingBlocks2(accumulatedText);
+          if (toolsUsed.length > 0) {
+            const info = {
+              toolsUsed,
+              thinking: thinkingBlocks.length > 0 ? thinkingBlocks : void 0,
+              ...lastUsageInfo
+            };
+            callback([cleanedText], info).then(() => settleResolve()).catch((err) => settleReject(err));
+          } else if (cleanedText) {
+            callback([cleanedText], lastUsageInfo).then(() => settleResolve()).catch((err) => settleReject(err));
+          } else {
+            settleResolve();
+          }
+        } else if (action === "cli_completion_error") {
+          const errorMsg = message.error || "Server error";
+          streamLogger.onCriticalEvent(eventCount, "ERROR", errorMsg);
+          settleReject(new Error(errorMsg));
+        }
+      });
+      try {
+        this.wsManager.send({
+          action: "cli_completion_request",
+          accessToken: token,
+          requestId,
+          model,
+          messages,
+          options: {
+            temperature: options.temperature,
+            maxTokens: options.maxTokens,
+            stream: true,
+            tools: options.tools || []
+          }
+        });
+      } catch (err) {
+        settleReject(err instanceof Error ? err : new Error(String(err)));
+      }
+    });
+  }
+  /**
+   * Get available models from server (REST call, not streaming).
+   * Delegates to ApiClient — same as ServerLlmBackend.
+   */
+  async getModelInfo() {
+    try {
+      logger.debug("[WebSocketLlmBackend] Fetching models from /api/models");
+      const response = await this.apiClient.get("/api/models");
+      if (!response || typeof response !== "object" || !Array.isArray(response.models)) {
+        logger.warn("[WebSocketLlmBackend] Invalid API response format, using fallback models");
+        return this.getFallbackModels();
+      }
+      const filteredModels = response.models.filter(
+        (model) => model.type === "text" && model.supportsTools === true
+      );
+      if (filteredModels.length === 0) {
+        logger.warn("[WebSocketLlmBackend] No CLI-compatible models found, using fallback");
+        return this.getFallbackModels();
+      }
+      logger.debug(`[WebSocketLlmBackend] Loaded ${filteredModels.length} models`);
+      return filteredModels;
+    } catch (error) {
+      logger.warn(
+        `[WebSocketLlmBackend] Failed to fetch models: ${error instanceof Error ? error.message : String(error)}`
+      );
+      return this.getFallbackModels();
+    }
+  }
+  getFallbackModels() {
+    return [
+      { id: "claude-sonnet-4-5-20250929", name: "Claude 4.5 Sonnet" },
+      { id: "claude-3-5-haiku-20241022", name: "Claude 3.5 Haiku" },
+      { id: "gpt-4o", name: "GPT-4o" },
+      { id: "gpt-4o-mini", name: "GPT-4o Mini" }
+    ];
+  }
+};
+
 // src/llm/NotifyingLlmBackend.ts
 var NotifyingLlmBackend = class {
   constructor(inner, backgroundManager) {
@@ -13456,6 +13616,223 @@ Please acknowledge these background agent results and incorporate them into your
   }
 };
 
+// src/ws/WebSocketConnectionManager.ts
+var WebSocketConnectionManager = class {
+  constructor(wsUrl, getToken) {
+    this.ws = null;
+    this.heartbeatInterval = null;
+    this.reconnectAttempts = 0;
+    this.maxReconnectDelay = 3e4;
+    this.handlers = /* @__PURE__ */ new Map();
+    this.connected = false;
+    this.connecting = false;
+    this.closed = false;
+    this.wsUrl = wsUrl;
+    this.getToken = getToken;
+  }
+  /**
+   * Connect to the WebSocket server.
+   * Resolves when connection is established, rejects on failure.
+   */
+  async connect() {
+    if (this.connected || this.connecting) return;
+    this.connecting = true;
+    const token = await this.getToken();
+    if (!token) {
+      this.connecting = false;
+      throw new Error("No access token available for WebSocket connection");
+    }
+    return new Promise((resolve3, reject) => {
+      const url = `${this.wsUrl}?token=${token}`;
+      logger.debug(`[WS] Connecting to ${this.wsUrl}...`);
+      this.ws = new WebSocket(url);
+      this.ws.onopen = () => {
+        logger.debug("[WS] Connected");
+        this.connected = true;
+        this.connecting = false;
+        this.reconnectAttempts = 0;
+        this.startHeartbeat();
+        resolve3();
+      };
+      this.ws.onmessage = (event) => {
+        try {
+          const data = typeof event.data === "string" ? event.data : event.data.toString();
+          const message = JSON.parse(data);
+          const requestId = message.requestId;
+          if (requestId && this.handlers.has(requestId)) {
+            this.handlers.get(requestId)(message);
+          } else {
+            logger.debug(`[WS] Unhandled message: ${message.action || "unknown"}`);
+          }
+        } catch (err) {
+          logger.debug(`[WS] Failed to parse message: ${err}`);
+        }
+      };
+      this.ws.onclose = () => {
+        logger.debug("[WS] Connection closed");
+        this.cleanup();
+        if (!this.closed) {
+          this.scheduleReconnect();
+        }
+      };
+      this.ws.onerror = (err) => {
+        logger.debug(`[WS] Error: ${err}`);
+        if (this.connecting) {
+          this.connecting = false;
+          this.connected = false;
+          reject(new Error("WebSocket connection failed"));
+        }
+      };
+    });
+  }
+  /** Whether the connection is currently established */
+  get isConnected() {
+    return this.connected;
+  }
+  /**
+   * Send a JSON message over the WebSocket connection.
+   */
+  send(data) {
+    if (!this.ws || this.ws.readyState !== WebSocket.OPEN) {
+      throw new Error("WebSocket is not connected");
+    }
+    this.ws.send(JSON.stringify(data));
+  }
+  /**
+   * Register a handler for messages matching a specific requestId.
+   */
+  onRequest(requestId, handler) {
+    this.handlers.set(requestId, handler);
+  }
+  /**
+   * Remove a handler for a specific requestId.
+   */
+  offRequest(requestId) {
+    this.handlers.delete(requestId);
+  }
+  /**
+   * Close the connection and stop all heartbeat/reconnect logic.
+   */
+  disconnect() {
+    this.closed = true;
+    this.cleanup();
+    if (this.ws) {
+      this.ws.close();
+      this.ws = null;
+    }
+    this.handlers.clear();
+  }
+  startHeartbeat() {
+    this.stopHeartbeat();
+    this.heartbeatInterval = setInterval(
+      () => {
+        if (this.ws && this.ws.readyState === WebSocket.OPEN) {
+          this.ws.send(JSON.stringify({ action: "heartbeat" }));
+          logger.debug("[WS] Heartbeat sent");
+        }
+      },
+      5 * 60 * 1e3
+    );
+  }
+  stopHeartbeat() {
+    if (this.heartbeatInterval) {
+      clearInterval(this.heartbeatInterval);
+      this.heartbeatInterval = null;
+    }
+  }
+  cleanup() {
+    this.connected = false;
+    this.connecting = false;
+    this.stopHeartbeat();
+  }
+  scheduleReconnect() {
+    if (this.closed) return;
+    this.reconnectAttempts++;
+    const delay = Math.min(1e3 * Math.pow(2, this.reconnectAttempts - 1), this.maxReconnectDelay);
+    logger.debug(`[WS] Reconnecting in ${delay}ms (attempt ${this.reconnectAttempts})`);
+    setTimeout(async () => {
+      if (this.closed) return;
+      try {
+        await this.connect();
+      } catch {
+        logger.debug("[WS] Reconnection failed");
+      }
+    }, delay);
+  }
+};
+
+// src/ws/WebSocketToolExecutor.ts
+import { v4 as uuidv412 } from "uuid";
+var WebSocketToolExecutor = class {
+  constructor(wsManager, tokenGetter) {
+    this.wsManager = wsManager;
+    this.tokenGetter = tokenGetter;
+  }
+  /**
+   * Execute a server-side tool via WebSocket.
+   * Returns the tool result or throws on error.
+   */
+  async execute(toolName, input, abortSignal) {
+    if (!this.wsManager.isConnected) {
+      throw new Error("WebSocket is not connected");
+    }
+    const token = await this.tokenGetter();
+    if (!token) {
+      throw new Error("No access token available");
+    }
+    const requestId = uuidv412();
+    return new Promise((resolve3, reject) => {
+      let settled = false;
+      const cleanup = () => {
+        this.wsManager.offRequest(requestId);
+      };
+      const settleResolve = (result) => {
+        if (settled) return;
+        settled = true;
+        cleanup();
+        resolve3(result);
+      };
+      const settleReject = (err) => {
+        if (settled) return;
+        settled = true;
+        cleanup();
+        reject(err);
+      };
+      if (abortSignal) {
+        const abortHandler = () => {
+          settleReject(new Error("Tool execution aborted"));
+        };
+        if (abortSignal.aborted) {
+          settleReject(new Error("Tool execution aborted"));
+          return;
+        }
+        abortSignal.addEventListener("abort", abortHandler, { once: true });
+      }
+      this.wsManager.onRequest(requestId, (message) => {
+        const action = message.action;
+        if (action === "cli_tool_response") {
+          settleResolve({
+            success: message.success,
+            content: message.content,
+            error: message.error
+          });
+        }
+      });
+      try {
+        this.wsManager.send({
+          action: "cli_tool_request",
+          accessToken: token,
+          requestId,
+          toolName,
+          input
+        });
+      } catch (err) {
+        settleReject(err instanceof Error ? err : new Error(String(err)));
+      }
+    });
+  }
+};
+
 // src/auth/ApiClient.ts
 import axios11 from "axios";
 var ApiClient = class {
@@ -13593,7 +13970,7 @@ import { isAxiosError as isAxiosError2 } from "axios";
 // package.json
 var package_default = {
   name: "@bike4mind/cli",
-  version: "0.2.29-feat-session-permission.18841+6d9c4deb5",
+  version: "0.2.29-feat-cli-websocket-streaming.18841+5c940d031",
   type: "module",
   description: "Interactive CLI tool for Bike4Mind with ReAct agents",
   license: "UNLICENSED",
@@ -13703,10 +14080,10 @@ var package_default = {
   },
   devDependencies: {
     "@bike4mind/agents": "0.1.0",
-    "@bike4mind/common": "2.50.1-feat-session-permission.18841+6d9c4deb5",
-    "@bike4mind/mcp": "1.29.1-feat-session-permission.18841+6d9c4deb5",
-    "@bike4mind/services": "2.48.1-feat-session-permission.18841+6d9c4deb5",
-    "@bike4mind/utils": "2.5.1-feat-session-permission.18841+6d9c4deb5",
+    "@bike4mind/common": "2.50.1-feat-cli-websocket-streaming.18841+5c940d031",
+    "@bike4mind/mcp": "1.29.1-feat-cli-websocket-streaming.18841+5c940d031",
+    "@bike4mind/services": "2.48.1-feat-cli-websocket-streaming.18841+5c940d031",
+    "@bike4mind/utils": "2.5.1-feat-cli-websocket-streaming.18841+5c940d031",
     "@types/better-sqlite3": "^7.6.13",
     "@types/diff": "^5.0.9",
     "@types/jsonwebtoken": "^9.0.4",
@@ -13723,7 +14100,7 @@ var package_default = {
   optionalDependencies: {
     "@vscode/ripgrep": "^1.17.0"
   },
-  gitHead: "6d9c4deb5a83cfc1087043d22f407d3444a91cca"
+  gitHead: "5c940d031610617c76d3c12cc611b71db8bb4caf"
 };
 
 // src/config/constants.ts
@@ -15531,7 +15908,8 @@ function CliApp() {
     agentStore: null,
     abortController: null,
     contextContent: "",
-    backgroundManager: null
+    backgroundManager: null,
+    wsManager: null
   });
   const [isInitialized, setIsInitialized] = useState9(false);
   const [initError, setInitError] = useState9(null);
@@ -15559,6 +15937,10 @@ function CliApp() {
         })
       );
     }
+    if (state.wsManager) {
+      state.wsManager.disconnect();
+      setWebSocketToolExecutor(null);
+    }
     if (state.agent) {
       state.agent.removeAllListeners();
     }
@@ -15577,7 +15959,7 @@ function CliApp() {
     setTimeout(() => {
       process.exit(0);
     }, 100);
-  }, [state.session, state.sessionStore, state.mcpManager, state.agent, state.imageStore]);
+  }, [state.session, state.sessionStore, state.mcpManager, state.agent, state.imageStore, state.wsManager]);
   useInput9((input, key) => {
     if (key.escape) {
       if (state.abortController) {
@@ -15647,7 +16029,7 @@ function CliApp() {
       if (!isAuthenticated) {
         console.log("\u2139\uFE0F  AI features disabled. Available commands: /login, /help, /config\n");
         const minimalSession = {
-          id: uuidv411(),
+          id: uuidv413(),
           name: `Session ${(/* @__PURE__ */ new Date()).toLocaleString()}`,
           createdAt: (/* @__PURE__ */ new Date()).toISOString(),
           updatedAt: (/* @__PURE__ */ new Date()).toISOString(),
@@ -15675,10 +16057,41 @@ function CliApp() {
         console.log(`\u{1F30D} API Environment: ${envName} (${apiBaseURL})`);
       }
       const apiClient = new ApiClient(apiBaseURL, state.configStore);
-      const llm = new ServerLlmBackend({
-        apiClient,
-        model: config.defaultModel
-      });
+      const tokenGetter = async () => {
+        const tokens = await state.configStore.getAuthTokens();
+        return tokens?.accessToken ?? null;
+      };
+      let wsManager = null;
+      let llm;
+      try {
+        const serverConfig = await apiClient.get("/api/settings/serverConfig");
+        const wsUrl = serverConfig?.websocketUrl;
+        if (wsUrl) {
+          wsManager = new WebSocketConnectionManager(wsUrl, tokenGetter);
+          await wsManager.connect();
+          const wsToolExecutor2 = new WebSocketToolExecutor(wsManager, tokenGetter);
+          setWebSocketToolExecutor(wsToolExecutor2);
+          llm = new WebSocketLlmBackend({
+            wsManager,
+            apiClient,
+            model: config.defaultModel,
+            tokenGetter
+          });
+          logger.debug("\u{1F50C} Using WebSocket transport (bypasses CloudFront timeout)");
+        } else {
+          throw new Error("No websocketUrl in server config");
+        }
+      } catch (wsError) {
+        logger.debug(
+          `[WS] WebSocket unavailable, using SSE fallback: ${wsError instanceof Error ? wsError.message : String(wsError)}`
+        );
+        wsManager = null;
+        setWebSocketToolExecutor(null);
+        llm = new ServerLlmBackend({
+          apiClient,
+          model: config.defaultModel
+        });
+      }
       const models = await llm.getModelInfo();
       if (models.length === 0) {
         throw new Error("No models available from server.");
@@ -15691,7 +16104,7 @@ function CliApp() {
       }
       llm.currentModel = modelInfo.id;
       const newSession = {
-        id: uuidv411(),
+        id: uuidv413(),
         name: `Session ${(/* @__PURE__ */ new Date()).toLocaleString()}`,
         createdAt: (/* @__PURE__ */ new Date()).toISOString(),
         updatedAt: (/* @__PURE__ */ new Date()).toISOString(),
@@ -15911,8 +16324,10 @@ ${contextResult.mergedContent}` : "";
         // Store agent store for agent management commands
         contextContent: contextResult.mergedContent,
         // Store raw context for compact instructions
-        backgroundManager
+        backgroundManager,
         // Store for grouped notification turn tracking
+        wsManager
+        // WebSocket connection manager (null if using SSE fallback)
       }));
       setStoreSession(newSession);
       const bannerLines = [
@@ -16007,13 +16422,13 @@ ${contextResult.mergedContent}` : "";
         messageContent = multimodalMessage.content;
       }
       const userMessage = {
-        id: uuidv411(),
+        id: uuidv413(),
         role: "user",
         content: userMessageContent,
         timestamp: (/* @__PURE__ */ new Date()).toISOString()
       };
       const pendingAssistantMessage = {
-        id: uuidv411(),
+        id: uuidv413(),
         role: "assistant",
         content: "...",
         timestamp: (/* @__PURE__ */ new Date()).toISOString(),
@@ -16219,13 +16634,13 @@ ${contextResult.mergedContent}` : "";
         userMessageContent = message;
       }
       const userMessage = {
-        id: uuidv411(),
+        id: uuidv413(),
         role: "user",
         content: userMessageContent,
         timestamp: (/* @__PURE__ */ new Date()).toISOString()
       };
       const pendingAssistantMessage = {
-        id: uuidv411(),
+        id: uuidv413(),
         role: "assistant",
         content: "...",
         timestamp: (/* @__PURE__ */ new Date()).toISOString(),
@@ -16303,7 +16718,7 @@ ${contextResult.mergedContent}` : "";
         const currentSession = useCliStore.getState().session;
         if (currentSession) {
           const cancelMessage = {
-            id: uuidv411(),
+            id: uuidv413(),
             role: "assistant",
             content: "\u26A0\uFE0F Operation cancelled by user",
             timestamp: (/* @__PURE__ */ new Date()).toISOString(),
@@ -16348,7 +16763,7 @@ ${contextResult.mergedContent}` : "";
     setState((prev) => ({ ...prev, abortController }));
     try {
       const pendingAssistantMessage = {
-        id: uuidv411(),
+        id: uuidv413(),
         role: "assistant",
         content: "...",
         timestamp: (/* @__PURE__ */ new Date()).toISOString(),
@@ -16376,7 +16791,7 @@ ${contextResult.mergedContent}` : "";
       const currentSession = useCliStore.getState().session;
       if (!currentSession) return;
       const continuationMessage = {
-        id: uuidv411(),
+        id: uuidv413(),
         role: "assistant",
         content: "---\n\n**Background Agent Results:**\n\n" + result.finalAnswer,
         timestamp: (/* @__PURE__ */ new Date()).toISOString(),
@@ -16435,13 +16850,13 @@ ${contextResult.mergedContent}` : "";
         isError = true;
       }
       const userMessage = {
-        id: uuidv411(),
+        id: uuidv413(),
         role: "user",
         content: `$ ${command}`,
         timestamp: (/* @__PURE__ */ new Date()).toISOString()
       };
       const assistantMessage = {
-        id: uuidv411(),
+        id: uuidv413(),
         role: "assistant",
         content: isError ? `\u274C Error:
 ${output}` : output.trim() || "(no output)",
@@ -16910,7 +17325,7 @@ Keyboard Shortcuts:
         console.clear();
         const model = state.session?.model || state.config?.defaultModel || "claude-sonnet";
         const newSession = {
-          id: uuidv411(),
+          id: uuidv413(),
           name: `Session ${(/* @__PURE__ */ new Date()).toLocaleString()}`,
           createdAt: (/* @__PURE__ */ new Date()).toISOString(),
           updatedAt: (/* @__PURE__ */ new Date()).toISOString(),
@@ -17405,9 +17820,9 @@ No usage data available for the last ${USAGE_DAYS} days.`);
       return { ...prev, config: updatedConfig };
     });
     if (modelChanged && state.agent) {
-      const llm = state.agent.context.llm;
-      if (llm) {
-        llm.currentModel = updatedConfig.defaultModel;
+      const backend = state.agent.context.llm;
+      if (backend) {
+        backend.currentModel = updatedConfig.defaultModel;
       }
     }
   };

package/dist/{llmMarkdownGenerator-IJKJYJ3U.js → llmMarkdownGenerator-JPDNDGE4.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   CurationArtifactType
-} from "./chunk-5GSDPBTM.js";
+} from "./chunk-KPOK6V6E.js";
 
 // ../../b4m-core/packages/services/dist/src/notebookCurationService/llmMarkdownGenerator.js
 var DEFAULT_OPTIONS = {

package/dist/{markdownGenerator-SJT7AXLT.js → markdownGenerator-NHN4MNEA.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   CurationArtifactType
-} from "./chunk-5GSDPBTM.js";
+} from "./chunk-KPOK6V6E.js";
 
 // ../../b4m-core/packages/services/dist/src/notebookCurationService/markdownGenerator.js
 var DEFAULT_OPTIONS = {

package/dist/{mementoService-PKCNMN4V.js → mementoService-EN7FZK7B.js}

@@ -2,9 +2,9 @@
 import {
   findMostSimilarMemento,
   getRelevantMementos
-} from "./chunk-BCOGDPUB.js";
-import "./chunk-S7BPPS33.js";
-import "./chunk-5GSDPBTM.js";
+} from "./chunk-QQL3OBKL.js";
+import "./chunk-TY6Z2UV2.js";
+import "./chunk-KPOK6V6E.js";
 import "./chunk-OCYRD7D6.js";
 export {
   findMostSimilarMemento,

package/dist/{src-GRO5NIC3.js → src-36AL3BJR.js}

@@ -41,6 +41,12 @@ import {
   ChatModels,
   CitableSourceSchema,
   ClaudeArtifactMimeTypes,
+  CliCompletionChunkAction,
+  CliCompletionDoneAction,
+  CliCompletionErrorAction,
+  CliCompletionRequestAction,
+  CliToolRequestAction,
+  CliToolResponseAction,
   CollectionType,
   CompletionApiUsageTransaction,
   CompletionRequestSchema,
@@ -340,7 +346,7 @@ import {
   validateReactArtifactV2,
   validateSvgArtifactV2,
   wikiMarkupToAdf
-} from "./chunk-5GSDPBTM.js";
+} from "./chunk-KPOK6V6E.js";
 export {
   ALL_IMAGE_MODELS,
   ALL_IMAGE_SIZES,
@@ -384,6 +390,12 @@ export {
   ChatModels,
   CitableSourceSchema,
   ClaudeArtifactMimeTypes,
+  CliCompletionChunkAction,
+  CliCompletionDoneAction,
+  CliCompletionErrorAction,
+  CliCompletionRequestAction,
+  CliToolRequestAction,
+  CliToolResponseAction,
   CollectionType,
   CompletionApiUsageTransaction,
   CompletionRequestSchema,

package/dist/{src-C5CZFXVM.js → src-JZN5O6F3.js}

@@ -134,8 +134,8 @@ import {
   validateMermaidSyntax,
   warmUpSettingsCache,
   withRetry
-} from "./chunk-S7BPPS33.js";
-import "./chunk-5GSDPBTM.js";
+} from "./chunk-TY6Z2UV2.js";
+import "./chunk-KPOK6V6E.js";
 import {
   Logger,
   NotificationDeduplicator,

package/dist/{subtractCredits-WZZAP57N.js → subtractCredits-FCBMFPIP.js}

@@ -2,9 +2,9 @@
 import {
   SubtractCreditsSchema,
   subtractCredits
-} from "./chunk-PKKJ44C5.js";
-import "./chunk-S7BPPS33.js";
-import "./chunk-5GSDPBTM.js";
+} from "./chunk-BXCARBUQ.js";
+import "./chunk-TY6Z2UV2.js";
+import "./chunk-KPOK6V6E.js";
 import "./chunk-OCYRD7D6.js";
 export {
   SubtractCreditsSchema,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bike4mind/cli",
-  "version": "0.2.29-feat-session-permission.18841+6d9c4deb5",
+  "version": "0.2.29-feat-cli-websocket-streaming.18841+5c940d031",
   "type": "module",
   "description": "Interactive CLI tool for Bike4Mind with ReAct agents",
   "license": "UNLICENSED",
@@ -110,10 +110,10 @@
   },
   "devDependencies": {
     "@bike4mind/agents": "0.1.0",
-    "@bike4mind/common": "2.50.1-feat-session-permission.18841+6d9c4deb5",
-    "@bike4mind/mcp": "1.29.1-feat-session-permission.18841+6d9c4deb5",
-    "@bike4mind/services": "2.48.1-feat-session-permission.18841+6d9c4deb5",
-    "@bike4mind/utils": "2.5.1-feat-session-permission.18841+6d9c4deb5",
+    "@bike4mind/common": "2.50.1-feat-cli-websocket-streaming.18841+5c940d031",
+    "@bike4mind/mcp": "1.29.1-feat-cli-websocket-streaming.18841+5c940d031",
+    "@bike4mind/services": "2.48.1-feat-cli-websocket-streaming.18841+5c940d031",
+    "@bike4mind/utils": "2.5.1-feat-cli-websocket-streaming.18841+5c940d031",
     "@types/better-sqlite3": "^7.6.13",
     "@types/diff": "^5.0.9",
     "@types/jsonwebtoken": "^9.0.4",
@@ -130,5 +130,5 @@
   "optionalDependencies": {
     "@vscode/ripgrep": "^1.17.0"
   },
-  "gitHead": "6d9c4deb5a83cfc1087043d22f407d3444a91cca"
+  "gitHead": "5c940d031610617c76d3c12cc611b71db8bb4caf"
 }