@bike4mind/cli 0.2.70 → 0.2.71-feat-7629-unified-streaming-model-fallback.22033

package/dist/{ConfigStore-DH64GYfC.mjs → ConfigStore-DCZ0ojCS.mjs}

@@ -8091,7 +8091,8 @@ const CliConfigSchema = z.object({
 	features: z.object({ tavern: z.boolean().optional() }).optional().prefault({}),
 	trustedTools: z.array(z.string()).optional().prefault([]),
 	sandbox: SandboxConfigSchema.optional(),
-	additionalDirectories: z.array(z.string()).optional().prefault([])
+	additionalDirectories: z.array(z.string()).optional().prefault([]),
+	fallbackModels: z.array(z.string()).optional()
 });
 /**
 * Zod schema for ProjectConfig validation

package/dist/commands/doctorCommand.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { i as version, n as fetchLatestVersion, r as forceCheckForUpdate } from "../updateChecker-BEb2EBef.mjs";
+import { i as version, n as fetchLatestVersion, r as forceCheckForUpdate } from "../updateChecker-CRjs4C6H.mjs";
 import { execSync } from "child_process";
 import { constants, existsSync, promises } from "fs";
 import { homedir } from "os";

package/dist/commands/headlessCommand.mjs

@@ -1,6 +1,6 @@
 #!/usr/bin/env node
-import { I as isReadOnlyTool, L as ReActAgent, M as setWebSocketToolExecutor, P as buildCoreSystemPrompt, R as CustomCommandStore, S as getApiUrl, T as generateCliTools, V as SessionStore, _ as McpManager, a as createBackgroundAgentTools, c as AgentStore, f as ApiClient, g as ServerLlmBackend, h as WebSocketLlmBackend, i as createWriteTodosTool, l as SubagentOrchestrator, m as WebSocketConnectionManager, n as createFindDefinitionTool, o as BackgroundAgentManager, p as WebSocketToolExecutor, r as createTodoStore, s as createAgentDelegateTool, t as createGetFileStructureTool, u as createSkillTool, w as PermissionManager, x as loadContextFiles, z as CheckpointStore } from "../tools-CJquAyBq.mjs";
-import { n as logger, t as ConfigStore } from "../ConfigStore-DH64GYfC.mjs";
+import { B as CheckpointStore, C as getApiUrl, E as generateCliTools, F as buildCoreSystemPrompt, H as SessionStore, L as isReadOnlyTool, N as setWebSocketToolExecutor, R as ReActAgent, S as loadContextFiles, T as PermissionManager, _ as ServerLlmBackend, a as createBackgroundAgentTools, c as AgentStore, f as ApiClient, g as WebSocketLlmBackend, h as FallbackLlmBackend, i as createWriteTodosTool, l as SubagentOrchestrator, m as WebSocketConnectionManager, n as createFindDefinitionTool, o as BackgroundAgentManager, p as WebSocketToolExecutor, r as createTodoStore, s as createAgentDelegateTool, t as createGetFileStructureTool, u as createSkillTool, v as McpManager, z as CustomCommandStore } from "../tools-8CXe7kfN.mjs";
+import { n as logger, t as ConfigStore } from "../ConfigStore-DCZ0ojCS.mjs";
 import { t as DEFAULT_SANDBOX_CONFIG } from "../types-DBEjF9YS.mjs";
 import { t as createSandboxRuntime } from "../SandboxRuntimeAdapter-C1B4t20N.mjs";
 import { t as SandboxOrchestrator } from "../SandboxOrchestrator-BEW3rqYi.mjs";
@@ -97,6 +97,9 @@ async function handleHeadlessCommand(options) {
 		if (models.length === 0) throw new Error("No models available from server.");
 		const modelInfo = models.find((m) => m.id === config.defaultModel) ?? models[0];
 		llm.currentModel = modelInfo.id;
+		const effectiveLlm = config.fallbackModels && config.fallbackModels.length > 0 ? new FallbackLlmBackend(llm, config.fallbackModels, (fromModel, toModel, error) => {
+			process.stderr.write(`⚠️  Model "${fromModel}" failed (${error.message}). Falling back to "${toModel}"...\n`);
+		}) : llm;
 		const session = {
 			id: v4(),
 			name: `Headless ${(/* @__PURE__ */ new Date()).toISOString()}`,
@@ -195,7 +198,7 @@ async function handleHeadlessCommand(options) {
 		const agent = new ReActAgent({
 			userId: config.userId,
 			logger: silentLogger,
-			llm,
+			llm: effectiveLlm,
 			model: modelInfo.id,
 			tools: allTools,
 			maxIterations,

package/dist/commands/mcpCommand.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { t as ConfigStore } from "../ConfigStore-DH64GYfC.mjs";
+import { t as ConfigStore } from "../ConfigStore-DCZ0ojCS.mjs";
 //#region src/commands/mcpCommand.ts
 /**
 * External MCP commands (b4m mcp list, b4m mcp add, etc.)

package/dist/commands/updateCommand.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { i as version, r as forceCheckForUpdate } from "../updateChecker-BEb2EBef.mjs";
+import { i as version, r as forceCheckForUpdate } from "../updateChecker-CRjs4C6H.mjs";
 import { execSync } from "child_process";
 //#region src/commands/updateCommand.ts
 /**

package/dist/index.mjs

@@ -1,8 +1,8 @@
 #!/usr/bin/env node
 import { n as useCliStore, t as selectActiveBackgroundAgents } from "./store-Dw1nZX2Y.mjs";
-import { A as clearFeatureModuleTools, B as CommandHistoryStore, C as getEnvironmentName, D as DEFAULT_AGENT_MODEL, E as ALWAYS_DENIED_FOR_AGENTS, F as buildSkillsPromptSection, G as searchCommands, H as OAuthClient, I as isReadOnlyTool, J as searchFiles, K as mergeCommands, L as ReActAgent, M as setWebSocketToolExecutor, N as OllamaBackend, O as DEFAULT_MAX_ITERATIONS, P as buildCoreSystemPrompt, R as CustomCommandStore, S as getApiUrl, T as generateCliTools, U as hasFileReferences, V as SessionStore, W as processFileReferences, Y as warmFileCache, _ as McpManager, a as createBackgroundAgentTools, b as extractCompactInstructions, c as AgentStore, d as parseAgentConfig, f as ApiClient, g as ServerLlmBackend, h as WebSocketLlmBackend, i as createWriteTodosTool, j as registerFeatureModuleTools, k as DEFAULT_THOROUGHNESS, l as SubagentOrchestrator, m as WebSocketConnectionManager, n as createFindDefinitionTool, o as BackgroundAgentManager, p as WebSocketToolExecutor, q as formatFileSize, r as createTodoStore, s as createAgentDelegateTool, t as createGetFileStructureTool, u as createSkillTool, v as substituteArguments, w as PermissionManager, x as loadContextFiles, y as formatStep, z as CheckpointStore } from "./tools-CJquAyBq.mjs";
-import { Dt as validateNotebookPath$1, Et as validateJupyterKernelName, g as ChatModels, m as CREDIT_DEDUCT_TRANSACTION_TYPES, n as logger, t as ConfigStore } from "./ConfigStore-DH64GYfC.mjs";
-import { i as version, t as checkForUpdate } from "./updateChecker-BEb2EBef.mjs";
+import { A as DEFAULT_THOROUGHNESS, B as CheckpointStore, C as getApiUrl, D as ALWAYS_DENIED_FOR_AGENTS, E as generateCliTools, F as buildCoreSystemPrompt, G as processFileReferences, H as SessionStore, I as buildSkillsPromptSection, J as formatFileSize, K as searchCommands, L as isReadOnlyTool, M as registerFeatureModuleTools, N as setWebSocketToolExecutor, O as DEFAULT_AGENT_MODEL, P as OllamaBackend, R as ReActAgent, S as loadContextFiles, T as PermissionManager, U as OAuthClient, V as CommandHistoryStore, W as hasFileReferences, X as warmFileCache, Y as searchFiles, _ as ServerLlmBackend, a as createBackgroundAgentTools, b as formatStep, c as AgentStore, d as parseAgentConfig, f as ApiClient, g as WebSocketLlmBackend, h as FallbackLlmBackend, i as createWriteTodosTool, j as clearFeatureModuleTools, k as DEFAULT_MAX_ITERATIONS, l as SubagentOrchestrator, m as WebSocketConnectionManager, n as createFindDefinitionTool, o as BackgroundAgentManager, p as WebSocketToolExecutor, q as mergeCommands, r as createTodoStore, s as createAgentDelegateTool, t as createGetFileStructureTool, u as createSkillTool, v as McpManager, w as getEnvironmentName, x as extractCompactInstructions, y as substituteArguments, z as CustomCommandStore } from "./tools-8CXe7kfN.mjs";
+import { Dt as validateNotebookPath$1, Et as validateJupyterKernelName, g as ChatModels, m as CREDIT_DEDUCT_TRANSACTION_TYPES, n as logger, t as ConfigStore } from "./ConfigStore-DCZ0ojCS.mjs";
+import { i as version, t as checkForUpdate } from "./updateChecker-CRjs4C6H.mjs";
 import React, { useCallback, useEffect, useMemo, useReducer, useRef, useState } from "react";
 import { Box, Static, Text, render, useApp, useInput } from "ink";
 import { execSync } from "child_process";
@@ -4606,7 +4606,9 @@ function CliApp() {
 			const agentDelegateTool = createAgentDelegateTool(orchestrator, agentStore, newSession.id, backgroundManager);
 			const dynamicAgentTool = config.preferences.enableDynamicAgentCreation === true ? createDynamicAgentTool(orchestrator, newSession.id, backgroundManager) : null;
 			const backgroundTools = createBackgroundAgentTools(backgroundManager);
-			const notifyingLlm = new NotifyingLlmBackend(llm, backgroundManager);
+			const notifyingLlm = new NotifyingLlmBackend(config.fallbackModels && config.fallbackModels.length > 0 ? new FallbackLlmBackend(llm, config.fallbackModels, (fromModel, toModel) => {
+				logger.warn(`⚠️  Model "${fromModel}" failed — falling back to "${toModel}"`);
+			}) : llm, backgroundManager);
 			const writeTodosTool = createWriteTodosTool(createTodoStore());
 			const enableSkillTool = config.preferences.enableSkillTool !== false;
 			const skillTool = enableSkillTool ? createSkillTool({

package/dist/{tools-CJquAyBq.mjs → tools-8CXe7kfN.mjs}

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { $ as RegInviteEvents, A as ImageGenerationUsageTransaction, B as OpenAIEmbeddingModel, C as FileEvents, Ct as isGPTImageModel, D as GenericCreditAddTransaction, E as GenerateImageToolCallSchema, F as KnowledgeType, G as ProjectEvents, H as Permission, I as LLMEvents, J as QuestMasterParamsSchema, K as PromptMetaZodSchema, L as MiscEvents, M as InboxEvents, N as InviteEvents, O as GenericCreditDeductTransaction, Ot as CollectionType, P as InviteType, Q as RechartsChartTypeList, R as ModalEvents, S as FeedbackEvents, St as getViewById, T as GEMINI_IMAGE_MODELS, Tt as sanitizeTelemetryError, U as PermissionDeniedError, V as OpenAIImageGenerationInput, W as ProfileEvents, X as RealtimeVoiceUsageTransaction, Y as REASONING_SUPPORTED_MODELS, Z as ReceivedCreditTransaction, _ as CompletionApiUsageTransaction, _t as VoyageAIEmbeddingModel, a as ApiKeyEvents, at as SpeechToTextModels, b as FIXED_TEMPERATURE_MODELS, bt as getDataLakeTags, c as AppFileEvents, ct as TagType, d as BFL_IMAGE_MODELS, dt as ToolUsageTransaction, et as ResearchModeParamsSchema, f as BFL_SAFETY_TOLERANCE, ft as TransferCreditTransaction, g as ChatModels, gt as VideoModels, h as ChatCompletionCreateInputSchema, ht as VideoGenerationUsageTransaction, i as AiEvents, it as SessionEvents, j as ImageModels, k as ImageEditUsageTransaction, l as ArtifactTypeSchema, lt as TaskScheduleHandler, mt as VIDEO_SIZE_CONSTRAINTS, n as logger, nt as ResearchTaskPeriodicFrequencyType, o as ApiKeyScope, ot as SubscriptionCreditTransaction, p as BedrockEmbeddingModel, pt as UiNavigationEvents, q as PurchaseTransaction, r as ALERT_THRESHOLDS, rt as ResearchTaskType, s as ApiKeyType, st as SupportedFabFileMimeTypes, t as ConfigStore, tt as ResearchTaskExecutionType, u as AuthEvents, ut as TextGenerationUsageTransaction, v as DashboardParamsSchema, vt as XAI_IMAGE_MODELS, w as FriendshipEvents, wt as resolveNavigationIntents, x as FavoriteDocumentType, xt as getMcpProviderMetadata, y as ElabsEvents, yt as b4mLLMTools, z as ModelBackend } from "./ConfigStore-DH64GYfC.mjs";
+import { $ as RegInviteEvents, A as ImageGenerationUsageTransaction, B as OpenAIEmbeddingModel, C as FileEvents, Ct as isGPTImageModel, D as GenericCreditAddTransaction, E as GenerateImageToolCallSchema, F as KnowledgeType, G as ProjectEvents, H as Permission, I as LLMEvents, J as QuestMasterParamsSchema, K as PromptMetaZodSchema, L as MiscEvents, M as InboxEvents, N as InviteEvents, O as GenericCreditDeductTransaction, Ot as CollectionType, P as InviteType, Q as RechartsChartTypeList, R as ModalEvents, S as FeedbackEvents, St as getViewById, T as GEMINI_IMAGE_MODELS, Tt as sanitizeTelemetryError, U as PermissionDeniedError, V as OpenAIImageGenerationInput, W as ProfileEvents, X as RealtimeVoiceUsageTransaction, Y as REASONING_SUPPORTED_MODELS, Z as ReceivedCreditTransaction, _ as CompletionApiUsageTransaction, _t as VoyageAIEmbeddingModel, a as ApiKeyEvents, at as SpeechToTextModels, b as FIXED_TEMPERATURE_MODELS, bt as getDataLakeTags, c as AppFileEvents, ct as TagType, d as BFL_IMAGE_MODELS, dt as ToolUsageTransaction, et as ResearchModeParamsSchema, f as BFL_SAFETY_TOLERANCE, ft as TransferCreditTransaction, g as ChatModels, gt as VideoModels, h as ChatCompletionCreateInputSchema, ht as VideoGenerationUsageTransaction, i as AiEvents, it as SessionEvents, j as ImageModels, k as ImageEditUsageTransaction, l as ArtifactTypeSchema, lt as TaskScheduleHandler, mt as VIDEO_SIZE_CONSTRAINTS, n as logger, nt as ResearchTaskPeriodicFrequencyType, o as ApiKeyScope, ot as SubscriptionCreditTransaction, p as BedrockEmbeddingModel, pt as UiNavigationEvents, q as PurchaseTransaction, r as ALERT_THRESHOLDS, rt as ResearchTaskType, s as ApiKeyType, st as SupportedFabFileMimeTypes, t as ConfigStore, tt as ResearchTaskExecutionType, u as AuthEvents, ut as TextGenerationUsageTransaction, v as DashboardParamsSchema, vt as XAI_IMAGE_MODELS, w as FriendshipEvents, wt as resolveNavigationIntents, x as FavoriteDocumentType, xt as getMcpProviderMetadata, y as ElabsEvents, yt as b4mLLMTools, z as ModelBackend } from "./ConfigStore-DCZ0ojCS.mjs";
 import { n as isPathAllowed, t as assertPathAllowed } from "./pathValidation-CIytuhr3-Dt5dntLx.mjs";
 import { execFile, execFileSync, spawn } from "child_process";
 import { createHash, randomBytes } from "crypto";
@@ -18188,16 +18188,16 @@ var StreamLogger = class StreamLogger {
 	}
 };
 //#endregion
-//#region src/llm/ServerLlmBackend.ts
+//#region src/llm/streamAccumulator.ts
 /**
-* Strip <think>...</think> blocks from text
-* Claude's extended thinking should not be shown in final output
+* Strip <think>...</think> blocks from text.
+* Claude's extended thinking should not be shown in final output.
 */
-function stripThinkingBlocks$1(text) {
+function stripThinkingBlocks(text) {
 	return text.replace(/<think>[\s\S]*?<\/think>/g, "").trim();
 }
 /**
-* Extract usage and credit information from SSE event
+* Extract usage and credit information into CompletionInfo shape.
 */
 function extractUsageInfo(parsed) {
 	return {
@@ -18208,6 +18208,64 @@ function extractUsageInfo(parsed) {
 	};
 }
 /**
+* Accumulates streaming LLM response chunks (text, tool calls, thinking blocks, usage)
+* and fires the completion callback once at the end.
+*
+* Shared between ServerLlmBackend (SSE) and WebSocketLlmBackend (WebSocket frames)
+* so accumulation logic lives in exactly one place.
+*/
+var StreamAccumulator = class {
+	constructor() {
+		this.accumulatedText = "";
+		this.toolsUsed = [];
+		this.thinkingBlocks = [];
+		this.lastUsageInfo = {};
+	}
+	onContent(text, usage, credits) {
+		this.accumulatedText += text;
+		if (usage || credits) this.lastUsageInfo = extractUsageInfo({
+			usage,
+			credits
+		});
+	}
+	onToolUse(text, tools, thinking, usage, credits) {
+		if (text) this.accumulatedText += text;
+		if (tools && tools.length > 0) this.toolsUsed = tools;
+		if (thinking && thinking.length > 0) this.thinkingBlocks = thinking;
+		if (usage || credits) this.lastUsageInfo = extractUsageInfo({
+			usage,
+			credits
+		});
+	}
+	/** True when neither text nor tools have been accumulated (stream produced nothing useful). */
+	isEmpty() {
+		return this.accumulatedText.trim().length === 0 && this.toolsUsed.length === 0;
+	}
+	get accumulatedLength() {
+		return this.accumulatedText.length;
+	}
+	get toolCount() {
+		return this.toolsUsed.length;
+	}
+	/**
+	* Calls the completion callback with all accumulated content.
+	* Strips thinking blocks from text before delivering.
+	*/
+	async finalize(callback) {
+		const cleanedText = stripThinkingBlocks(this.accumulatedText);
+		if (this.toolsUsed.length > 0) {
+			const info = {
+				toolsUsed: this.toolsUsed,
+				thinking: this.thinkingBlocks.length > 0 ? this.thinkingBlocks : void 0,
+				...this.lastUsageInfo
+			};
+			await callback([cleanedText], info);
+		} else if (cleanedText) await callback([cleanedText], this.lastUsageInfo);
+	}
+};
+//#endregion
+//#region src/llm/ServerLlmBackend.ts
+/**
 * Server-side LLM backend that proxies requests through Bike4Mind API
 * Uses Server-Sent Events (SSE) for streaming responses
 * API keys remain secure on server - never exposed to CLI
@@ -18310,10 +18368,7 @@ var ServerLlmBackend = class ServerLlmBackend {
 			const streamLogger = new StreamLogger(logger, "ServerLlmBackend", process.env.B4M_VERBOSE === "1", process.env.B4M_DEBUG_STREAM === "1");
 			streamLogger.streamStart();
 			let eventCount = 0;
-			let accumulatedText = "";
-			let lastUsageInfo = {};
-			let toolsUsed = [];
-			let thinkingBlocks = [];
+			const accumulator = new StreamAccumulator();
 			let receivedDone = false;
 			const parser = createParser({ onEvent: (event) => {
 				eventCount++;
@@ -18321,28 +18376,15 @@ var ServerLlmBackend = class ServerLlmBackend {
 				const data = event.data;
 				if (data === "[DONE]") {
 					receivedDone = true;
-					streamLogger.onCriticalEvent(eventCount, "[DONE]", `accumulated text length: ${accumulatedText.length}`);
-					const cleanedText = stripThinkingBlocks$1(accumulatedText);
-					streamLogger.streamComplete(accumulatedText);
-					if (toolsUsed.length > 0) {
-						const info = {
-							toolsUsed,
-							thinking: thinkingBlocks.length > 0 ? thinkingBlocks : void 0,
-							...lastUsageInfo
-						};
-						logger.debug(`[ServerLlmBackend] Calling callback with tools, thinking blocks: ${thinkingBlocks.length}`);
-						callback([cleanedText], info).catch((err) => {
-							logger.error("[ServerLlmBackend] Callback error:", err);
-							reject(err);
-						}).then(() => {
-							logger.debug("[ServerLlmBackend] Callback completed, resolving");
-							resolve();
-						});
-					} else if (cleanedText) callback([cleanedText], lastUsageInfo).catch((err) => {
+					streamLogger.onCriticalEvent(eventCount, "[DONE]", `accumulated text length: ${accumulator.accumulatedLength}`);
+					streamLogger.streamComplete("");
+					accumulator.finalize(callback).catch((err) => {
 						logger.error("[ServerLlmBackend] Callback error:", err);
 						reject(err);
-					}).then(() => resolve());
-					else resolve();
+					}).then(() => {
+						logger.debug("[ServerLlmBackend] Callback completed, resolving");
+						resolve();
+					});
 					return;
 				}
 				try {
@@ -18354,9 +18396,8 @@ var ServerLlmBackend = class ServerLlmBackend {
 					}
 					if (parsed.type === "content") {
 						const textChunk = parsed.text || "";
-						accumulatedText += textChunk;
-						if (parsed.usage || parsed.credits) lastUsageInfo = extractUsageInfo(parsed);
-						streamLogger.onContent(eventCount, textChunk, accumulatedText);
+						accumulator.onContent(textChunk, parsed.usage, parsed.credits);
+						streamLogger.onContent(eventCount, textChunk, "");
 					} else if (parsed.type === "tool_use") {
 						streamLogger.onCriticalEvent(eventCount, "TOOL_USE", `tools: ${parsed.tools?.length}`);
 						if (parsed.tools && parsed.tools.length > 0) for (const tool of parsed.tools) {
@@ -18368,14 +18409,8 @@ var ServerLlmBackend = class ServerLlmBackend {
 								logger.debug(`  Params: [Unable to stringify]`);
 							}
 						}
-						const textChunk = parsed.text || "";
-						if (textChunk) accumulatedText += textChunk;
-						if (parsed.tools && parsed.tools.length > 0) toolsUsed = parsed.tools;
-						if (parsed.thinking && parsed.thinking.length > 0) {
-							thinkingBlocks = parsed.thinking;
-							streamLogger.onCriticalEvent(eventCount, "THINKING", `${thinkingBlocks.length} thinking blocks`);
-						}
-						if (parsed.usage || parsed.credits) lastUsageInfo = extractUsageInfo(parsed);
+						accumulator.onToolUse(parsed.text || "", parsed.tools, parsed.thinking, parsed.usage, parsed.credits);
+						if (parsed.thinking && parsed.thinking.length > 0) streamLogger.onCriticalEvent(eventCount, "THINKING", `${parsed.thinking.length} thinking blocks`);
 					}
 				} catch (parseError) {
 					streamLogger.streamError(parseError);
@@ -18402,20 +18437,10 @@ var ServerLlmBackend = class ServerLlmBackend {
 			});
 			response.data.on("end", () => {
 				if (!receivedDone) {
-					const hasAccumulatedData = accumulatedText.trim().length > 0 || toolsUsed.length > 0;
-					logger.warn(`[ServerLlmBackend] Stream ended without [DONE] signal. Accumulated text: ${accumulatedText.length} chars, tools: ${toolsUsed.length}`);
-					if (hasAccumulatedData) {
-						const cleanedText = stripThinkingBlocks$1(accumulatedText);
-						streamLogger.streamComplete(accumulatedText);
-						if (toolsUsed.length > 0) {
-							const info = {
-								toolsUsed,
-								thinking: thinkingBlocks.length > 0 ? thinkingBlocks : void 0,
-								...lastUsageInfo
-							};
-							callback([cleanedText], info).then(() => resolve(), reject);
-						} else if (cleanedText) callback([cleanedText], lastUsageInfo).then(() => resolve(), reject);
-						else resolve();
+					logger.warn(`[ServerLlmBackend] Stream ended without [DONE] signal. Accumulated text: ${accumulator.accumulatedLength} chars, tools: ${accumulator.toolCount}`);
+					if (!accumulator.isEmpty()) {
+						streamLogger.streamComplete("");
+						accumulator.finalize(callback).then(() => resolve(), reject);
 					} else reject(/* @__PURE__ */ new Error("Stream ended prematurely without receiving any data. The server may be experiencing issues."));
 				} else logger.debug("[ServerLlmBackend] Stream ended, [DONE] handler will resolve");
 			});
@@ -18536,9 +18561,6 @@ var ServerLlmBackend = class ServerLlmBackend {
 };
 //#endregion
 //#region src/llm/WebSocketLlmBackend.ts
-function stripThinkingBlocks(text) {
-	return text.replace(/<think>[\s\S]*?<\/think>/g, "").trim();
-}
 /**
 * Hybrid HTTP + WebSocket LLM backend for CLI completions.
 *
@@ -18572,10 +18594,7 @@ var WebSocketLlmBackend = class {
 			const streamLogger = new StreamLogger(logger, "WebSocketLlmBackend", process.env.B4M_VERBOSE === "1", process.env.B4M_DEBUG_STREAM === "1");
 			streamLogger.streamStart();
 			let eventCount = 0;
-			let accumulatedText = "";
-			let lastUsageInfo = {};
-			let toolsUsed = [];
-			let thinkingBlocks = [];
+			const accumulator = new StreamAccumulator();
 			let settled = false;
 			const settle = (action) => {
 				if (settled) return;
@@ -18603,12 +18622,6 @@ var WebSocketLlmBackend = class {
 				}
 				options.abortSignal.addEventListener("abort", abortHandler, { once: true });
 			}
-			const updateUsage = (usage) => {
-				if (usage) lastUsageInfo = {
-					inputTokens: usage.inputTokens,
-					outputTokens: usage.outputTokens
-				};
-			};
 			this.wsManager.onRequest(requestId, (message) => {
 				if (options.abortSignal?.aborted) return;
 				const action = message.action;
@@ -18616,28 +18629,20 @@ var WebSocketLlmBackend = class {
 					eventCount++;
 					const chunk = message.chunk;
 					streamLogger.onEvent(eventCount, JSON.stringify(chunk));
-					const textChunk = chunk.text || "";
-					if (textChunk) accumulatedText += textChunk;
-					updateUsage(chunk.usage);
-					if (chunk.type === "content") streamLogger.onContent(eventCount, textChunk, accumulatedText);
-					else if (chunk.type === "tool_use") {
+					if (chunk.type === "content") {
+						accumulator.onContent(chunk.text || "", chunk.usage);
+						streamLogger.onContent(eventCount, chunk.text || "", "");
+					} else if (chunk.type === "tool_use") {
 						streamLogger.onCriticalEvent(eventCount, "TOOL_USE", `tools: ${chunk.tools?.length}`);
-						if (chunk.tools && chunk.tools.length > 0) toolsUsed = chunk.tools;
-						if (chunk.thinking && chunk.thinking.length > 0) thinkingBlocks = chunk.thinking;
+						accumulator.onToolUse(chunk.text || "", chunk.tools, chunk.thinking, chunk.usage);
 					}
 				} else if (action === "cli_completion_done") {
-					streamLogger.streamComplete(accumulatedText);
-					const cleanedText = stripThinkingBlocks(accumulatedText);
-					if (!cleanedText && toolsUsed.length === 0) {
+					streamLogger.streamComplete("");
+					if (accumulator.isEmpty()) {
 						settleResolve();
 						return;
 					}
-					const info = {
-						...lastUsageInfo,
-						...toolsUsed.length > 0 && { toolsUsed },
-						...thinkingBlocks.length > 0 && { thinking: thinkingBlocks }
-					};
-					callback([cleanedText], info).then(() => settleResolve()).catch((err) => settleReject(err));
+					accumulator.finalize(callback).then(() => settleResolve()).catch((err) => settleReject(err));
 				} else if (action === "cli_completion_error") {
 					const errorMsg = message.error || "Server error";
 					streamLogger.onCriticalEvent(eventCount, "ERROR", errorMsg);
@@ -18738,6 +18743,59 @@ var WebSocketLlmBackend = class {
 	}
 };
 //#endregion
+//#region src/llm/FallbackLlmBackend.ts
+/**
+* LLM backend decorator that provides model-level fallback routing.
+*
+* When the primary model fails (after the inner backend's own retries are exhausted),
+* FallbackLlmBackend tries the next model in the configured fallback chain.
+*
+* Example chain: Opus → Sonnet → Haiku (graceful degradation under rate limits)
+*
+* Configured via `CliConfig.fallbackModels`. Wraps any `ICompletionBackend`,
+* fitting cleanly into the existing decorator pattern (NotifyingLlmBackend, etc.).
+*/
+var FallbackLlmBackend = class {
+	constructor(inner, fallbackModels, onFallback) {
+		this.inner = inner;
+		this.fallbackModels = fallbackModels;
+		this.onFallback = onFallback;
+	}
+	get currentModel() {
+		return this.inner.currentModel;
+	}
+	set currentModel(model) {
+		this.inner.currentModel = model;
+	}
+	async complete(model, messages, options, callback) {
+		if (options.abortSignal?.aborted) return this.inner.complete(model, messages, options, callback);
+		const modelsToTry = [model, ...this.fallbackModels.filter((m) => m !== model)];
+		let lastError;
+		for (let i = 0; i < modelsToTry.length; i++) {
+			const modelToTry = modelsToTry[i];
+			try {
+				await this.inner.complete(modelToTry, messages, options, callback);
+				return;
+			} catch (error) {
+				if (options.abortSignal?.aborted) throw error;
+				lastError = error instanceof Error ? error : new Error(String(error));
+				const nextModel = modelsToTry[i + 1];
+				if (nextModel) {
+					logger.warn(`[FallbackLlmBackend] Model "${modelToTry}" failed: ${lastError.message}`);
+					this.onFallback(modelToTry, nextModel, lastError);
+				}
+			}
+		}
+		throw lastError ?? /* @__PURE__ */ new Error("All fallback models exhausted");
+	}
+	pushToolMessages(messages, tool, result, thinkingBlocks) {
+		this.inner.pushToolMessages(messages, tool, result, thinkingBlocks);
+	}
+	async getModelInfo() {
+		return this.inner.getModelInfo();
+	}
+};
+//#endregion
 //#region src/ws/WebSocketConnectionManager.ts
 const useWsPolyfill = typeof globalThis.WebSocket === "undefined";
 const WS = useWsPolyfill ? WsWebSocket : globalThis.WebSocket;
@@ -20919,4 +20977,4 @@ function createGetFileStructureTool() {
 	};
 }
 //#endregion
-export { clearFeatureModuleTools as A, CommandHistoryStore as B, getEnvironmentName as C, DEFAULT_AGENT_MODEL as D, ALWAYS_DENIED_FOR_AGENTS as E, buildSkillsPromptSection as F, searchCommands as G, OAuthClient as H, isReadOnlyTool as I, searchFiles as J, mergeCommands as K, ReActAgent as L, setWebSocketToolExecutor as M, OllamaBackend as N, DEFAULT_MAX_ITERATIONS as O, buildCoreSystemPrompt as P, CustomCommandStore as R, getApiUrl as S, generateCliTools as T, hasFileReferences as U, SessionStore as V, processFileReferences as W, warmFileCache as Y, McpManager as _, createBackgroundAgentTools as a, extractCompactInstructions as b, AgentStore as c, parseAgentConfig as d, ApiClient as f, ServerLlmBackend as g, WebSocketLlmBackend as h, createWriteTodosTool as i, registerFeatureModuleTools as j, DEFAULT_THOROUGHNESS as k, SubagentOrchestrator as l, WebSocketConnectionManager as m, createFindDefinitionTool as n, BackgroundAgentManager as o, WebSocketToolExecutor as p, formatFileSize$1 as q, createTodoStore as r, createAgentDelegateTool as s, createGetFileStructureTool as t, createSkillTool as u, substituteArguments as v, PermissionManager as w, loadContextFiles as x, formatStep as y, CheckpointStore as z };
+export { DEFAULT_THOROUGHNESS as A, CheckpointStore as B, getApiUrl as C, ALWAYS_DENIED_FOR_AGENTS as D, generateCliTools as E, buildCoreSystemPrompt as F, processFileReferences as G, SessionStore as H, buildSkillsPromptSection as I, formatFileSize$1 as J, searchCommands as K, isReadOnlyTool as L, registerFeatureModuleTools as M, setWebSocketToolExecutor as N, DEFAULT_AGENT_MODEL as O, OllamaBackend as P, ReActAgent as R, loadContextFiles as S, PermissionManager as T, OAuthClient as U, CommandHistoryStore as V, hasFileReferences as W, warmFileCache as X, searchFiles as Y, ServerLlmBackend as _, createBackgroundAgentTools as a, formatStep as b, AgentStore as c, parseAgentConfig as d, ApiClient as f, WebSocketLlmBackend as g, FallbackLlmBackend as h, createWriteTodosTool as i, clearFeatureModuleTools as j, DEFAULT_MAX_ITERATIONS as k, SubagentOrchestrator as l, WebSocketConnectionManager as m, createFindDefinitionTool as n, BackgroundAgentManager as o, WebSocketToolExecutor as p, mergeCommands as q, createTodoStore as r, createAgentDelegateTool as s, createGetFileStructureTool as t, createSkillTool as u, McpManager as v, getEnvironmentName as w, extractCompactInstructions as x, substituteArguments as y, CustomCommandStore as z };

package/dist/{updateChecker-BEb2EBef.mjs → updateChecker-CRjs4C6H.mjs}

@@ -4,7 +4,7 @@ import { homedir } from "os";
 import path from "path";
 import axios from "axios";
 //#region package.json
-var version = "0.2.70";
+var version = "0.2.71-feat-7629-unified-streaming-model-fallback.22033+9b05427c0";
 //#endregion
 //#region src/utils/updateChecker.ts
 /**

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bike4mind/cli",
-  "version": "0.2.70",
+  "version": "0.2.71-feat-7629-unified-streaming-model-fallback.22033+9b05427c0",
   "type": "module",
   "description": "Interactive CLI tool for Bike4Mind with ReAct agents",
   "license": "UNLICENSED",
@@ -60,7 +60,7 @@
     "@opensearch-project/opensearch": "2.11.0",
     "@smithy/node-http-handler": "^4.5.0",
     "async-mutex": "^0.5.0",
-    "axios": "^1.13.6",
+    "axios": "1.14.0",
     "bcryptjs": "^3.0.2",
     "better-sqlite3": "^12.8.0",
     "cheerio": "1.0.0-rc.12",
@@ -115,11 +115,11 @@
     "zustand": "^4.5.4"
   },
   "devDependencies": {
-    "@bike4mind/agents": "0.4.7",
-    "@bike4mind/common": "2.78.0",
-    "@bike4mind/mcp": "1.33.26",
-    "@bike4mind/services": "2.71.1",
-    "@bike4mind/utils": "2.16.8",
+    "@bike4mind/agents": "0.4.8-feat-7629-unified-streaming-model-fallback.22033+9b05427c0",
+    "@bike4mind/common": "2.78.1-feat-7629-unified-streaming-model-fallback.22033+9b05427c0",
+    "@bike4mind/mcp": "1.33.27-feat-7629-unified-streaming-model-fallback.22033+9b05427c0",
+    "@bike4mind/services": "2.71.2-feat-7629-unified-streaming-model-fallback.22033+9b05427c0",
+    "@bike4mind/utils": "2.16.9-feat-7629-unified-streaming-model-fallback.22033+9b05427c0",
     "@types/better-sqlite3": "^7.6.13",
     "@types/jsonwebtoken": "^9.0.4",
     "@types/node": "^22.9.0",
@@ -136,5 +136,5 @@
   "optionalDependencies": {
     "@vscode/ripgrep": "^1.17.1"
   },
-  "gitHead": "3953e8b96bf559a76010be7880c2c1f74aec410d"
+  "gitHead": "9b05427c060904f024c75194742301784f85a0e3"
 }