nolo-cli 0.54.0 → 0.61.0-alpha.2

package/{chunk-YLCUU3WM.js → chunk-4NJEIGBQ.js}

@@ -1,15 +1,15 @@
 import {
   selectCurrentServer
-} from "./chunk-SVC3GGUA.js";
+} from "./chunk-VMWYBIXG.js";
 import {
   patch
-} from "./chunk-7NYDJAQ7.js";
-import {
-  selectIdentityToken
-} from "./chunk-XOJRZUPF.js";
+} from "./chunk-DNNSCSA2.js";
 import {
   getIsDesktopApp
 } from "./chunk-QOKAKDPE.js";
+import {
+  selectIdentityToken
+} from "./chunk-SEJ5OS6F.js";
 
 // packages/ai/agent/cliChatClient.ts
 function getCliChatRequestConfig(thunkApi) {

package/{chunk-K4NISKOR.js → chunk-5ZYA4JJ3.js}

@@ -1,12 +1,12 @@
 import {
   MODEL_LOOKUP_MAP
-} from "./chunk-SVDMVX4G.js";
+} from "./chunk-MMZ6ZTVO.js";
 import {
   ALL_MODELS
-} from "./chunk-NJUQBYFJ.js";
+} from "./chunk-FKFMIBTT.js";
 import {
   BUILTIN_NOLO_AGENT_KEY
-} from "./chunk-2WBFJHQA.js";
+} from "./chunk-L7B47JXR.js";
 
 // packages/ai/llm/getModelContextWindow.ts
 var DEFAULT_CONTEXT_WINDOW = 256e3;
@@ -34,6 +34,13 @@ function stripPreviewSuffix(model) {
   }
   return model;
 }
+function stripModelTag(model) {
+  const colonIndex = model.indexOf(":");
+  if (colonIndex > 0) {
+    return model.slice(0, colonIndex);
+  }
+  return model;
+}
 var fullModelMap = null;
 var getFullModelMap = () => {
   if (!fullModelMap) {
@@ -80,25 +87,48 @@ function fuzzyContextWindow(normalizedName) {
   if (normalizedName.includes("qwen3.7") || normalizedName.includes("qwen3p7")) return 262144;
   if (normalizedName.includes("minimax-m2") || normalizedName.includes("minimax_m2")) return 262144;
   if (normalizedName.includes("claude")) return 2e5;
+  if (normalizedName.includes("llama3.1") || normalizedName.includes("llama-3.1") || normalizedName.includes("llama3.2") || normalizedName.includes("llama-3.2") || normalizedName.includes("llama3.3") || normalizedName.includes("llama-3.3") || normalizedName.includes("qwen2.5") || normalizedName.includes("qwen-2.5") || normalizedName.includes("mistral-nemo") || normalizedName.includes("phi3.5") || normalizedName.includes("phi-3.5") || normalizedName.includes("phi4") || normalizedName.includes("phi-4")) {
+    return 131072;
+  }
+  if (normalizedName.includes("qwen2") || normalizedName.includes("qwen-2") || normalizedName.includes("mistral") || normalizedName.includes("mixtral")) {
+    return 32768;
+  }
+  if (normalizedName.includes("codellama") || normalizedName.includes("starcoder") || normalizedName.includes("phi3") || normalizedName.includes("phi-3")) {
+    return 16384;
+  }
+  if (normalizedName.includes("llama3") || normalizedName.includes("llama-3") || normalizedName.includes("gemma2") || normalizedName.includes("gemma-2") || normalizedName.includes("gemma")) {
+    return 8192;
+  }
+  if (normalizedName.includes("llama2") || normalizedName.includes("llama-2") || normalizedName.includes("llama")) {
+    return 4096;
+  }
   return void 0;
 }
 var getModelContextWindow = (modelName) => {
   if (!modelName) return DEFAULT_CONTEXT_WINDOW;
   const normalizedName = modelName.toLowerCase();
   const map = getFullModelMap();
-  const candidates = [modelName, normalizedName];
-  const strippedEffort = stripEffortSuffix(modelName);
-  if (strippedEffort !== modelName) {
-    candidates.push(strippedEffort, strippedEffort.toLowerCase());
-  }
-  const strippedPreview = stripPreviewSuffix(modelName);
-  if (strippedPreview !== modelName) {
-    candidates.push(strippedPreview, strippedPreview.toLowerCase());
+  const rawCandidates = [modelName, normalizedName];
+  const strippedTag = stripModelTag(modelName);
+  if (strippedTag !== modelName) {
+    rawCandidates.push(strippedTag, strippedTag.toLowerCase());
   }
-  if (strippedEffort !== modelName) {
-    const strippedBoth = stripPreviewSuffix(strippedEffort);
-    if (strippedBoth !== strippedEffort) {
-      candidates.push(strippedBoth, strippedBoth.toLowerCase());
+  const candidates = [];
+  for (const base of rawCandidates) {
+    candidates.push(base);
+    const strippedEffort = stripEffortSuffix(base);
+    if (strippedEffort !== base) {
+      candidates.push(strippedEffort, strippedEffort.toLowerCase());
+    }
+    const strippedPreview = stripPreviewSuffix(base);
+    if (strippedPreview !== base) {
+      candidates.push(strippedPreview, strippedPreview.toLowerCase());
+    }
+    if (strippedEffort !== base) {
+      const strippedBoth = stripPreviewSuffix(strippedEffort);
+      if (strippedBoth !== strippedEffort) {
+        candidates.push(strippedBoth, strippedBoth.toLowerCase());
+      }
     }
   }
   for (const candidate of candidates) {
@@ -152,7 +182,7 @@ var readCacheCreationInputTokens = (usage) => firstPositive(usage, [
 ]);
 
 // packages/agent-runtime/emptyAssistantRepair.ts
-var EMPTY_ASSISTANT_REPAIR_PROMPT = "\u8BF7\u7ED9\u51FA\u660E\u786E\u7684\u6587\u5B57\u56DE\u7B54\u6216\u6267\u884C\u4E0B\u4E00\u6B65\uFF1A\u5982\u679C\u4EFB\u52A1\u5DF2\u5B8C\u6210\uFF0C\u8BF7\u76F4\u63A5\u603B\u7ED3\u7ED3\u679C\uFF1B\u5982\u679C\u9700\u8981\u8C03\u7528\u5DE5\u5177\uFF0C\u8BF7\u76F4\u63A5\u8F93\u51FA tool_calls\u3002\u8BF7\u5207\u52FF\u8FD4\u56DE\u7A7A\u5185\u5BB9\u3002";
+var EMPTY_ASSISTANT_REPAIR_PROMPT = "\u8BF7\u7ED9\u51FA\u660E\u786E\u7684\u6587\u5B57\u56DE\u7B54\u6216\u6267\u884C\u4E0B\u4E00\u6B65\uFF1A\u5982\u679C\u4EFB\u52A1\u5DF2\u5B8C\u6210\uFF0C\u8BF7\u76F4\u63A5\u603B\u7ED3\u7ED3\u679C\uFF1B\u5982\u679C\u9700\u8981\u8C03\u7528\u5DE5\u5177\uFF0C\u53EA\u80FD\u8C03\u7528\u672C\u6B21\u8BF7\u6C42\u4E2D\u5DF2\u58F0\u660E\u7684\u5DE5\u5177\u2014\u2014\u4F60\u60F3\u7528\u7684\u5DE5\u5177\u4E0D\u5728\u58F0\u660E\u5217\u8868\u4E2D\u65F6\uFF0C\u4E0D\u8981\u5C1D\u8BD5\u8C03\u7528\u5B83\uFF0C\u76F4\u63A5\u7528\u6587\u5B57\u8BF4\u660E\u8BE5\u80FD\u529B\u5F53\u524D\u4E0D\u53EF\u7528\u3002\u8BF7\u5207\u52FF\u8FD4\u56DE\u7A7A\u5185\u5BB9\u3002";
 var EMPTY_ASSISTANT_FALLBACK_MESSAGE = "\u6A21\u578B\u8FDE\u7EED\u8FD4\u56DE\u7A7A\u6D88\u606F\uFF0C\u5F53\u524D\u4EFB\u52A1\u672A\u5B8C\u6210\u3002\u8BF7\u91CD\u8BD5\u5F53\u524D\u6B65\u9AA4\uFF0C\u6216\u7ED9\u51FA\u66F4\u5177\u4F53\u7684\u4FEE\u6539\u8303\u56F4\u3002";
 var LENGTH_TRUNCATED_FALLBACK_MESSAGE = "\u8F93\u51FA\u8FBE\u5230\u957F\u5EA6\u4E0A\u9650\u88AB\u622A\u65AD\uFF0C\u5EFA\u8BAE\u7F29\u77ED\u4EFB\u52A1\u6216\u63D0\u9AD8\u8F93\u51FA\u4E0A\u9650\u3002";
 var STREAM_TRUNCATED_FALLBACK_MESSAGE = "\u4E0A\u6E38\u54CD\u5E94\u6D41\u5728\u6536\u5C3E\u524D\u88AB\u4E2D\u65AD\uFF08\u672A\u6536\u5230\u7ED3\u675F\u6807\u8BB0\uFF09\uFF0C\u672C\u8F6E\u8F93\u51FA\u4E0D\u5B8C\u6574\u3002\u8BF7\u91CD\u8BD5\u5F53\u524D\u6B65\u9AA4\u3002";
@@ -246,7 +276,7 @@ var FLASH_PROFILE = createProfile({
   tier: "flash",
   legacyAgentKey: BUILTIN_NOLO_AGENT_KEY,
   name: "nolo",
-  model: "glm-5-3-flash"
+  model: "deepseek-flash"
 });
 var AUTO_EXECUTION_PROFILES = {
   flash: FLASH_PROFILE,

package/{chunk-3Y2K5O2R.js → chunk-7A3RRUCQ.js}

@@ -15,15 +15,17 @@ import {
   processThinkChunk
 } from "./chunk-N7J63Q6C.js";
 import {
+  collectStablePromptPrefix,
   convertMessagesToResponsesInput,
   extractTextFromResponseOutput,
   extractToolCallsFromResponseOutput,
   getUsageRequestOptions,
   hasImageParts,
   shouldStripReasoningContentForOutbound,
+  stablePromptCacheKey,
   toOpenAiCompatibleMessages,
   toResponsesTools
-} from "./chunk-VZMUOPYG.js";
+} from "./chunk-XTJTN3RW.js";
 import {
   buildProviderExecutionPlan,
   canUsePlatformChatProvider,
@@ -33,13 +35,15 @@ import {
   resolvePlatformAuthToken,
   resolvePlatformServerUrl,
   resolveProviderTransportDecision
-} from "./chunk-KLIBF3BY.js";
+} from "./chunk-RDBARW6T.js";
 import {
-  clipCompactText
-} from "./chunk-BBTN4NQD.js";
+  clipCompactText,
+  isMeaningfulChatCompletionOutput,
+  isMeaningfulResponsesOutput
+} from "./chunk-Z4QSJILV.js";
 import {
   NOLO_CLIENT_VERSION_HEADER
-} from "./chunk-HPW5MNDQ.js";
+} from "./chunk-22BTGNSP.js";
 import {
   isRecord
 } from "./chunk-KUG52IAP.js";
@@ -78,33 +82,6 @@ function extractUsageFromSsePayload(parsed) {
   return void 0;
 }
 
-// packages/agent-runtime/promptCacheKey.ts
-function collectStablePromptPrefix(messages, fallbackPrompt) {
-  const systemTexts = [];
-  if (fallbackPrompt?.trim()) systemTexts.push(fallbackPrompt.trim());
-  for (const raw of messages) {
-    if (!raw || typeof raw !== "object") continue;
-    const message = raw;
-    const role = String(message.role ?? "");
-    if (role !== "system" && role !== "developer") continue;
-    if (typeof message.content === "string") {
-      const boundary = Number(message.stable_prefix_chars);
-      const text = Number.isFinite(boundary) && boundary > 0 ? message.content.slice(0, boundary) : message.content;
-      if (text.trim()) systemTexts.push(text.trim());
-    }
-  }
-  return systemTexts.join("\n\n");
-}
-function stablePromptCacheKey(parts, namespace) {
-  const value = JSON.stringify(parts);
-  let hash = 2166136261;
-  for (let index = 0; index < value.length; index += 1) {
-    hash ^= value.charCodeAt(index);
-    hash = Math.imul(hash, 16777619) >>> 0;
-  }
-  return `${namespace}-${hash.toString(16).padStart(8, "0")}`;
-}
-
 // packages/core/chat/providerFailureMessage.ts
 var RAW_BODY_CLIP = 400;
 function describeProviderFailure(args) {
@@ -266,6 +243,11 @@ function finalizeResponsesToolCalls(calls) {
 }
 
 // packages/agent-runtime/processChatCompletionDelta.ts
+function observeFirstMeaningfulChatCompletionDelta(state) {
+  if (!state.onMeaningfulDelta || state.sawMeaningfulDelta) return;
+  state.sawMeaningfulDelta = true;
+  state.onMeaningfulDelta();
+}
 function extractChatCompletionStreamError(parsed) {
   const raw = parsed?.error;
   if (!raw) return null;
@@ -308,6 +290,9 @@ function applyChatCompletionDelta(parsed, state) {
   }
   const delta = parsed?.choices?.[0]?.delta;
   if (!delta || typeof delta !== "object") return false;
+  if (isMeaningfulChatCompletionOutput(delta)) {
+    observeFirstMeaningfulChatCompletionDelta(state);
+  }
   const reasoningChunk = typeof delta.reasoning_content === "string" ? delta.reasoning_content : typeof delta.reasoning === "string" ? delta.reasoning : "";
   if (reasoningChunk) {
     state.reasoning += reasoningChunk;
@@ -380,6 +365,16 @@ async function* readSseFrames(response) {
       boundary = findSseFrameBoundary(buffer);
     }
   }
+  const remaining = decoder.decode();
+  if (remaining) {
+    buffer += remaining;
+    let boundary = findSseFrameBoundary(buffer);
+    while (boundary) {
+      yield buffer.slice(0, boundary.index);
+      buffer = buffer.slice(boundary.index + boundary.separatorLength);
+      boundary = findSseFrameBoundary(buffer);
+    }
+  }
   if (buffer.trim()) yield buffer;
 }
 function findSseFrameBoundary(buffer) {
@@ -405,6 +400,16 @@ async function* streamSseDataValues(response, parse) {
       if (parsed) yield parsed;
     }
   }
+  const remaining = decoder.decode();
+  if (remaining) {
+    buffer += remaining;
+    const lines = buffer.split("\n");
+    buffer = lines.pop() ?? "";
+    for (const line of lines) {
+      const parsed = parse(line);
+      if (parsed) yield parsed;
+    }
+  }
   if (buffer.trim()) {
     const parsed = parse(buffer);
     if (parsed) yield parsed;
@@ -419,6 +424,45 @@ async function readSseDataValues(response, parse) {
 }
 
 // packages/agent-runtime/platformChatProvider.ts
+var PlatformStreamInterruptedError = class extends Error {
+  code = "PLATFORM_STREAM_INTERRUPTED";
+  name = "PlatformStreamInterrupted";
+  phase;
+  httpStatus;
+  frameCount;
+  textChars;
+  sawDone;
+  sawUsage;
+  sawFinishReason;
+  elapsedMs;
+  constructor(args) {
+    super(
+      args.message ?? `Platform LLM stream interrupted (${args.phase}, frames=${args.frameCount ?? 0}, chars=${args.textChars ?? 0})`,
+      args.cause !== void 0 ? { cause: args.cause } : void 0
+    );
+    this.name = "PlatformStreamInterrupted";
+    this.phase = args.phase;
+    this.httpStatus = args.httpStatus ?? (args.phase === "streaming" ? 200 : 0);
+    this.frameCount = args.frameCount ?? 0;
+    this.textChars = args.textChars ?? 0;
+    this.sawDone = args.sawDone ?? false;
+    this.sawUsage = args.sawUsage ?? false;
+    this.sawFinishReason = args.sawFinishReason ?? false;
+    this.elapsedMs = args.elapsedMs ?? 0;
+    if (args.cause !== void 0 && this.cause === void 0) {
+      this.cause = args.cause;
+    }
+  }
+};
+function isAbortOrTimeoutError(error) {
+  if (typeof DOMException !== "undefined" && error instanceof DOMException) {
+    return error.name === "AbortError" || error.name === "TimeoutError";
+  }
+  if (error instanceof Error) {
+    return error.name === "AbortError" || error.name === "TimeoutError";
+  }
+  return false;
+}
 var CHAT_PROXY_PATH = "/api/v1/chat";
 function shouldDisableThinking(providerConfig) {
   return providerConfig.provider.toLowerCase() === "mimo" || /xiaomimimo\.com/i.test(providerConfig.endpoint);
@@ -545,11 +589,12 @@ function buildPlatformChatCompletionRequest(args) {
       headers: {
         "Content-Type": "application/json",
         Authorization: `Bearer ${args.providerConfig.authToken}`,
+        ...args.requestId ? { "x-request-id": args.requestId } : {},
         // 流式请求显式放弃压缩协商。实测记录（2026-09-01，glm-5-3-flash 经
         // nolo.chat）：某个时间窗内 SSE 响应带 content-encoding: gzip 返回，
         // 整条流被攒成 1 个 chunk，首字节 = 总生成时长（5.8/10.6/16.3s）；
         // 同一分钟交替发起的 identity 请求为 123-138 chunks、首字节
-        // 1.2-3.6s。即压缩中间件一旦介入，流式的全部 TTFT 收益归零。
+        // 1.2-3.6s。即压缩中间件一旦介入，流式的首个输出收益归零。
         // 该窗口之后线上不再压缩 SSE（同一探针 6/6 未压缩），具体是哪一跳
         // 压的没有定论；边缘侧已把流式路径排除出 encode
         // （configureCaddyProxy.sh），这里是不依赖任何中间跳行为的客户端兜底
@@ -709,15 +754,24 @@ function processPlatformChatSseEvent(event, state) {
       if (dsml.content) {
         state.content += dsml.content;
         state.onTextDelta?.(dsml.content);
+        if (isMeaningfulResponsesOutput(parsed)) {
+          observeFirstMeaningfulChatCompletionDelta(state);
+        }
       }
       continue;
     }
     if (eventType === "response.reasoning_text.delta" && typeof parsed?.delta === "string") {
       state.reasoning += parsed.delta;
       state.onReasoningDelta?.(parsed.delta);
+      if (isMeaningfulResponsesOutput(parsed)) {
+        observeFirstMeaningfulChatCompletionDelta(state);
+      }
       continue;
     }
     if (eventType === "response.output_item.added" || eventType === "response.output_item.done" || eventType === "response.function_call_arguments.delta" || eventType === "response.function_call_arguments.done") {
+      if (isMeaningfulResponsesOutput(parsed)) {
+        observeFirstMeaningfulChatCompletionDelta(state);
+      }
       applyResponsesToolEvent(state.responsesToolCalls, parsed);
       continue;
     }
@@ -727,6 +781,8 @@ function processPlatformChatSseEvent(event, state) {
   }
 }
 async function readPlatformChatSseCompletion(args) {
+  const startedAtMs = Date.now();
+  let frameCount = 0;
   const state = {
     content: "",
     reasoning: "",
@@ -735,6 +791,7 @@ async function readPlatformChatSseCompletion(args) {
     usesResponsesApi: args.usesResponsesApi,
     onTextDelta: args.onTextDelta,
     onReasoningDelta: args.onReasoningDelta,
+    ...args.onMeaningfulDelta ? { onMeaningfulDelta: args.onMeaningfulDelta } : {},
     completedResponsesPayload: void 0,
     responsesToolCalls: createResponsesToolAccumulator(),
     dsmlToolCallState: createDsmlParserState(),
@@ -745,8 +802,35 @@ async function readPlatformChatSseCompletion(args) {
     finishReason: void 0,
     sawDone: false
   };
-  for await (const frame of readSseFrames(args.response)) {
-    processPlatformChatSseEvent(frame, state);
+  try {
+    for await (const frame of readSseFrames(args.response)) {
+      frameCount += 1;
+      processPlatformChatSseEvent(frame, state);
+    }
+  } catch (cause) {
+    if (cause instanceof PlatformStreamInterruptedError) {
+      throw cause;
+    }
+    if (isAbortOrTimeoutError(cause)) {
+      throw cause;
+    }
+    const elapsedMs = Date.now() - startedAtMs;
+    const textChars = state.content.length;
+    const sawDone = state.sawDone;
+    const sawUsage = Boolean(state.usage);
+    const sawFinishReason = Boolean(state.finishReason);
+    throw new PlatformStreamInterruptedError({
+      message: cause instanceof Error ? cause.message : "Platform chat SSE stream interrupted",
+      phase: "streaming",
+      httpStatus: 200,
+      frameCount,
+      textChars,
+      sawDone,
+      sawUsage,
+      sawFinishReason,
+      elapsedMs,
+      cause
+    });
   }
   flushChatCompletionStream(state);
   if (state.usesResponsesApi) {
@@ -762,6 +846,8 @@ async function readPlatformChatSseCompletion(args) {
     if (dsml.content) state.content += dsml.content;
   }
   throwIfChatCompletionStreamFailed(state);
+  const streamComplete = state.sawDone || Boolean(state.usage) || Boolean(state.finishReason);
+  const streamTruncated = !streamComplete;
   if (state.usesResponsesApi) {
     const response = state.completedResponsesPayload;
     const completedToolCalls = response ? extractToolCallsFromResponseOutput(response) : [];
@@ -777,13 +863,13 @@ async function readPlatformChatSseCompletion(args) {
     );
     const tool_calls2 = completedToolCalls.length > 0 ? completedToolCalls : normalizedDsmlCalls.length > 0 ? normalizedDsmlCalls : completionsToolCalls.length > 0 ? completionsToolCalls : finalizeResponsesToolCalls(state.responsesToolCalls);
     const finalUsage2 = mergeBillingIntoUsage(state.usage, state.billing);
-    const streamComplete2 = state.sawDone || !!state.usage || !!state.finishReason;
     return {
       content: normalizedFinal.content || state.content,
       ...state.reasoning ? { reasoning_content: state.reasoning } : {},
       ...tool_calls2.length > 0 ? { tool_calls: tool_calls2 } : {},
       ...finalUsage2 ? { usage: finalUsage2 } : {},
-      ...streamComplete2 ? { stream_complete: true } : {},
+      ...streamComplete ? { stream_complete: true } : {},
+      ...streamTruncated ? { streamTruncated: true } : {},
       // finish_reason 供 executePlatformChatCompletion 穿透给 localLoop 的
       // 空轮/截断判定；此前 reader 从不回传，流式路径恒缺。
       ...state.finishReason ? { finish_reason: state.finishReason } : {}
@@ -791,13 +877,13 @@ async function readPlatformChatSseCompletion(args) {
   }
   const tool_calls = finalizeAccumulatedToolCalls(state.accumulatedToolCalls);
   const finalUsage = mergeBillingIntoUsage(state.usage, state.billing);
-  const streamComplete = state.sawDone || !!state.usage || !!state.finishReason;
   return {
     content: state.content,
     ...state.reasoning ? { reasoning_content: state.reasoning } : {},
     ...tool_calls.length > 0 ? { tool_calls } : {},
     ...finalUsage ? { usage: finalUsage } : {},
     ...streamComplete ? { stream_complete: true } : {},
+    ...streamTruncated ? { streamTruncated: true } : {},
     ...state.finishReason ? { finish_reason: state.finishReason } : {}
   };
 }
@@ -841,8 +927,7 @@ export {
   createResponsesToolAccumulator,
   applyResponsesToolEvent,
   finalizeResponsesToolCalls,
-  collectStablePromptPrefix,
-  stablePromptCacheKey,
+  PlatformStreamInterruptedError,
   resolvePlatformChatProviderConfig,
   buildPlatformChatCompletionRequest,
   parsePlatformChatCompletionData,