mslxdff 0.1.88 → 0.1.89

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mslxdff",
-  "version": "0.1.88",
+  "version": "0.1.89",
   "description": "测试项目，请勿使用。",
   "type": "module",
   "bin": {

package/src/bench/cline-bench.js

@@ -1,54 +1,42 @@
 import { clineHeaders } from "../providers/cline/headers.js";
 import { computeMetrics } from "../metrics.js";
+import { createTransport } from "../transport/index.js";
+
+function sseContent(obj) {
+  const c = obj?.choices?.[0]?.delta?.content || obj?.choices?.[0]?.message?.content || "";
+  return typeof c === "string" ? c : "";
+}
 
 export async function clineBenchOne({ baseUrl, model, accessToken, prompt, maxTokens, timeoutMs, fetchImpl }) {
-  const controller = new AbortController();
-  const timer = setTimeout(() => controller.abort(new Error(`timeout ${timeoutMs}ms`)), timeoutMs);
-  const t0 = performance.now();
+  const tr = createTransport({ fetchImpl, keepAlive: false, retry: {}, timeoutMs });
   let ttfbMs = null;
+  let totalMs = null;
   let content = "";
   try {
-    const res = await fetchImpl(`${baseUrl}/api/v1/chat/completions`, {
+    const sid = `sess_bench_${Date.now()}`;
+    const res = await tr.request({
+      url: `${baseUrl}/api/v1/chat/completions`,
       method: "POST",
-      headers: { ...clineHeaders(`sess_bench_${Date.now()}`, accessToken), Accept: "text/event-stream" },
-      body: JSON.stringify({ model, messages: [{ role: "user", content: prompt }], stream: true, max_tokens: maxTokens, session_id: `sess_bench_${Date.now()}`, reasoning_effort: "high" }),
-      signal: controller.signal,
+      headers: { ...clineHeaders(sid, accessToken), Accept: "text/event-stream" },
+      body: { model, messages: [{ role: "user", content: prompt }], stream: true, max_tokens: maxTokens, session_id: sid, reasoning_effort: "high" },
+      stream: true,
     });
-    if (res instanceof Error) throw res;
     if (!res.ok) {
       let txt = "";
       try { txt = await res.text(); } catch {}
       const label = res.status === 401 ? "鉴权失败" : res.status === 429 ? "限流" : res.status >= 500 ? `上游错误 ${res.status}` : `HTTP ${res.status}`;
-      return { id: model, ok: false, status: res.status, label, error: txt.slice(0, 300), ttfbMs, totalMs: Math.round(performance.now() - t0), tps: null, charsPerSec: null, tokens: null };
+      return { id: model, ok: false, status: res.status, label, error: txt.slice(0, 300), ttfbMs, totalMs: res.totalMs, tps: null, charsPerSec: null, tokens: null };
     }
-    const reader = res.body.getReader();
-    const decoder = new TextDecoder();
-    let buf = "";
-    for (;;) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      if (ttfbMs === null) ttfbMs = Math.round(performance.now() - t0);
-      buf += decoder.decode(value, { stream: true });
-      let idx;
-      while ((idx = buf.indexOf("\n")) >= 0) {
-        const line = buf.slice(0, idx);
-        buf = buf.slice(idx + 1);
-        if (!line.startsWith("data:")) continue;
-        const payload = line.slice(5).trim();
-        if (!payload || payload === "[DONE]") continue;
-        try {
-          const j = JSON.parse(payload);
-          const c = j?.choices?.[0]?.delta?.content || j?.choices?.[0]?.message?.content || "";
-          if (typeof c === "string") content += c;
-        } catch {}
-      }
+    for await (const ev of res.stream()) {
+      try { content += sseContent(JSON.parse(ev)); } catch {}
     }
-    const totalMs = Math.round(performance.now() - t0);
+    ttfbMs = res.ttfbMs;
+    totalMs = res.totalMs;
     const chars = content.length;
     const { tps, charsPerSec } = computeMetrics({ ttfbMs, totalMs, promptTokens: null, completionTokens: null, chars });
     return { id: model, ok: true, status: 200, label: "成功", ttfbMs, totalMs, tps, charsPerSec, tokens: null, chars };
   } catch (e) {
     const msg = e?.message || String(e);
-    return { id: model, ok: false, label: /timeout|abort/i.test(msg) ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs: Math.round(performance.now() - t0), tps: null, charsPerSec: null, tokens: null };
-  } finally { clearTimeout(timer); }
+    return { id: model, ok: false, label: /timeout|abort/i.test(msg) ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
+  }
 }

package/src/bench/probe.js

@@ -1,9 +1,9 @@
 import { joinUrl } from "../providers/base.js";
 import { getCustomNormalizer } from "../providers/registry.js";
+import { createTransport } from "../transport/index.js";
 
 function normalizeModelsPayload(json, baseUrl = "") {
   if (!json) return [];
-  // 可扩展：优先走注册表的定制化解析（如 cline.bot 仅 free）
   try {
     const custom = getCustomNormalizer(baseUrl);
     if (custom) {
@@ -46,27 +46,20 @@ export async function probeModels({
     if (!seen.has(norm)) { seen.add(norm); candidates.push(norm); }
   };
   if (modelsPath) push(modelsPath);
-  // 通用回退：/v1/models 与 /models
   push("/v1/models");
   push("/models");
-  // workbuddy 异形已在 defaultModelsPath 注入，若仍未命中则 candidates 已含
+  const tr = createTransport({ fetchImpl, keepAlive: false, retry: {}, timeoutMs });
   const tried = [];
   let lastError = "";
   for (const p of candidates) {
     const url = joinUrl(base, p);
     tried.push(url);
     try {
-      const controller = new AbortController();
-      const t = setTimeout(() => controller.abort(new Error(`probe timeout ${timeoutMs}ms`)), timeoutMs);
-      let res;
-      try {
-        res = await fetchImpl(url, { headers, signal: controller.signal });
-      } finally { clearTimeout(t); }
-      if (res instanceof Error) { lastError = res.message || String(res); continue; }
+      const res = await tr.request({ url, method: "GET", headers, stream: false });
       if (!res.ok) { lastError = `HTTP ${res.status}`; continue; }
       let json = {};
       try { json = await res.json(); } catch { json = {}; }
-      const raw = normalizeModelsPayload(json);
+      const raw = normalizeModelsPayload(json, baseUrl);
       const data = raw.map((m) => ({ id: toModelId(m), raw: m })).filter((x) => x.id).map((x) => ({ id: x.id, ...x.raw }));
       return { ok: true, data, tried, url, rawCount: raw.length };
     } catch (e) {

package/src/bench/runner.js

@@ -1,5 +1,6 @@
 import { joinUrl } from "../providers/base.js";
 import { computeMetrics, extractUsageFromJson } from "../metrics.js";
+import { createTransport } from "../transport/index.js";
 
 function extractInnerMessage(bodyText) {
   const t = String(bodyText || "");
@@ -37,43 +38,28 @@ export async function runOne({
   maxTokens = 32,
   timeoutMs = 30000,
   fetchImpl = globalThis.fetch,
-  clock = Date.now,
 } = {}) {
-  const started = clock();
-  const t0 = typeof performance !== "undefined" && performance.now ? performance.now() : started;
   if (!baseUrl) return { id: model, ok: false, error: "missing baseUrl", label: "配置错误", ttfbMs: null, totalMs: 0, tps: null, charsPerSec: null, tokens: null };
   if (!model) return { id: model, ok: false, error: "missing model", label: "配置错误", ttfbMs: null, totalMs: 0, tps: null, charsPerSec: null, tokens: null };
   if (!apiKey) return { id: model, ok: false, error: "missing apiKey", label: "未配置 Key", ttfbMs: null, totalMs: 0, tps: null, charsPerSec: null, tokens: null };
   const url = joinUrl(String(baseUrl).replace(/\/+$/, ""), chatPath);
   let rawModel = String(model || "").trim();
-  // allowlist 已是 raw（如 z-ai/glm-5.3-flash），仅当带供应商前缀时才剥
   if (providerId && rawModel.startsWith(`${providerId}/`)) rawModel = rawModel.slice(providerId.length + 1);
   const body = { model: rawModel, stream: false, messages: [{ role: "user", content: prompt }], max_tokens: maxTokens };
-  // workbuddy 需额外 workbuddy 透传头由调用方 headers 注入；此处直接透传
   const finalHeaders = { "Content-Type": "application/json", Accept: "application/json", ...headers };
   if (apiKey && !finalHeaders.Authorization) finalHeaders.Authorization = `Bearer ${apiKey}`;
-
-  let ttfbMs = null;
-  let res;
+  const tr = createTransport({ fetchImpl, keepAlive: false, retry: {}, timeoutMs });
   try {
-    const controller = new AbortController();
-    const timer = setTimeout(() => controller.abort(new Error(`timeout ${timeoutMs}ms`)), timeoutMs);
-    const fetchStart = typeof performance !== "undefined" && performance.now ? performance.now() : clock();
-    try {
-      res = await fetchImpl(url, { method: "POST", headers: finalHeaders, body: JSON.stringify(body), signal: controller.signal });
-    } finally { clearTimeout(timer); }
-    ttfbMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - fetchStart);
-    if (res instanceof Error) throw res;
-    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - t0);
+    const res = await tr.request({ url, method: "POST", headers: finalHeaders, body, stream: false });
+    const ttfbMs = res.ttfbMs;
+    const totalMs = res.totalMs;
     if (!res.ok) {
-      let txt = "";
-      try { txt = await res.text(); } catch {}
+      let txt = ""; try { txt = await res.text(); } catch {}
       const cls = classifyError(res.status, txt);
       const msg = extractInnerMessage(txt) || `HTTP ${res.status}`;
       return { id: model, providerId, ok: false, status: res.status, error: msg, label: cls.label, ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
     }
-    let json = {};
-    let txt = "";
+    let json = {}; let txt = "";
     try { txt = await res.text(); json = JSON.parse(txt); } catch { json = {}; }
     const usage = extractUsageFromJson(json);
     const content = json?.choices?.[0]?.message?.content || json?.choices?.[0]?.text || txt || "";
@@ -84,9 +70,8 @@ export async function runOne({
     const totalTokens = usage?.total_tokens ?? (promptTokens !== null && completionTokens !== null ? promptTokens + completionTokens : null);
     return { id: model, providerId, ok: true, status: res.status, ttfbMs, totalMs, tps, charsPerSec, tokens: { prompt: promptTokens, completion: completionTokens, total: totalTokens }, chars, label: "成功", raw: json };
   } catch (e) {
-    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - t0);
     const msg = e?.message || String(e);
     const isTimeout = /timeout|abort/i.test(msg);
-    return { id: model, providerId, ok: false, error: msg.slice(0, 300), label: isTimeout ? "超时" : "网络错误", ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
+    return { id: model, providerId, ok: false, error: msg.slice(0, 300), label: isTimeout ? "超时" : "网络错误", ttfbMs: null, totalMs: null, tps: null, charsPerSec: null, tokens: null };
   }
 }

package/src/bench/via-probe.js

@@ -1,5 +1,6 @@
 import { joinUrl } from "../providers/base.js";
 import { computeMetrics, extractUsageFromJson } from "../metrics.js";
+import { createTransport } from "../transport/index.js";
 
 function extractInnerMessage(bodyText) {
   const t = String(bodyText || "");
@@ -43,11 +44,9 @@ export async function viaProbe({
   relayBody,
   targetUrl,
 } = {}) {
-  const started = clock();
-  const t0 = typeof performance !== "undefined" && performance.now ? performance.now() : started;
   const base = String(peerUrl || "").replace(/\/+$/, "");
   if (!base) return { ok: false, label: "配置错误", error: "missing peerUrl", ttfbMs: null, totalMs: 0, tps: null, charsPerSec: null, tokens: null };
-  // 纯中继模式：A 把 targetUrl+headers+body 发给 B，B 原样 fetch 到上游（不查 B 本地 providerConfigs）
+  const tr = createTransport({ fetchImpl, keepAlive: false, retry: {}, timeoutMs });
   const rt = String(relayTarget || targetUrl || "").trim();
   if (rt) {
     const rh = relayHeaders && typeof relayHeaders === "object" ? relayHeaders : {};
@@ -56,15 +55,10 @@ export async function viaProbe({
     const relayHeadersOut = { "Content-Type": "application/json", Accept: "application/json" };
     if (token) relayHeadersOut.Authorization = `Bearer ${token}`;
     const payload = { targetUrl: rt, method: "POST", headers: rh, body: rb };
-    let ttfbMs = null;
     try {
-      const controller = new AbortController();
-      const timer = setTimeout(() => controller.abort(new Error(`timeout ${timeoutMs}ms`)), timeoutMs);
-      const fetchStart = typeof performance !== "undefined" && performance.now ? performance.now() : clock();
-      let res;
-      try { res = await fetchImpl(relayUrl, { method: "POST", headers: relayHeadersOut, body: JSON.stringify(payload), signal: controller.signal }); } finally { clearTimeout(timer); }
-      ttfbMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - fetchStart);
-      const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - t0);
+      const res = await tr.request({ url: relayUrl, method: "POST", headers: relayHeadersOut, body: payload, stream: false });
+      const ttfbMs = res.ttfbMs;
+      const totalMs = res.totalMs;
       if (!res.ok) {
         let txt = ""; try { txt = await res.text(); } catch {}
         const cls = classifyError(res.status, txt);
@@ -88,10 +82,9 @@ export async function viaProbe({
       const totalTokens = usage?.total_tokens ?? (pt !== null && ct !== null ? pt + ct : null);
       return { ok: true, status: relayStatus, label: "成功", ttfbMs, totalMs, tps, charsPerSec, tokens: { prompt: pt, completion: ct, total: totalTokens }, chars };
     } catch (e) {
-      const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - t0);
       const msg = e?.message || String(e);
       const isTimeout = /timeout|abort/i.test(msg);
-      return { ok: false, label: isTimeout ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
+      return { ok: false, label: isTimeout ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs: null, totalMs: null, tps: null, charsPerSec: null, tokens: null };
     }
   }
   if (!model) return { ok: false, label: "配置错误", error: "missing model", ttfbMs: null, totalMs: 0, tps: null, charsPerSec: null, tokens: null };
@@ -103,27 +96,17 @@ export async function viaProbe({
   if (token) headers.Authorization = `Bearer ${token}`;
   const sk = shareKeysHeader || shareKeys;
   if (sk) headers["x-mslxdff-share-keys"] = String(sk);
-  let ttfbMs = null;
   try {
-    const controller = new AbortController();
-    const timer = setTimeout(() => controller.abort(new Error(`timeout ${timeoutMs}ms`)), timeoutMs);
-    const fetchStart = typeof performance !== "undefined" && performance.now ? performance.now() : clock();
-    let res;
-    try {
-      res = await fetchImpl(url, { method: "POST", headers, body: JSON.stringify(body), signal: controller.signal });
-    } finally { clearTimeout(timer); }
-    ttfbMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - fetchStart);
-    if (res instanceof Error) throw res;
-    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - t0);
+    const res = await tr.request({ url, method: "POST", headers, body, stream: false });
+    const ttfbMs = res.ttfbMs;
+    const totalMs = res.totalMs;
     if (!res.ok) {
-      let txt = "";
-      try { txt = await res.text(); } catch {}
+      let txt = ""; try { txt = await res.text(); } catch {}
       const cls = classifyError(res.status, txt);
       const msg = extractInnerMessage(txt) || `HTTP ${res.status}`;
       return { ok: false, status: res.status, label: cls.label, error: msg, ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
     }
-    let json = {};
-    let txt = "";
+    let json = {}; let txt = "";
     try { txt = await res.text(); json = JSON.parse(txt); } catch { json = {}; }
     const usage = extractUsageFromJson(json);
     const content = json?.choices?.[0]?.message?.content || json?.choices?.[0]?.text || txt || "";
@@ -134,9 +117,8 @@ export async function viaProbe({
     const totalTokens = usage?.total_tokens ?? (promptTokens !== null && completionTokens !== null ? promptTokens + completionTokens : null);
     return { ok: true, status: res.status, label: "成功", ttfbMs, totalMs, tps, charsPerSec, tokens: { prompt: promptTokens, completion: completionTokens, total: totalTokens }, chars };
   } catch (e) {
-    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : clock()) - t0);
     const msg = e?.message || String(e);
     const isTimeout = /timeout|abort/i.test(msg);
-    return { ok: false, label: isTimeout ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
+    return { ok: false, label: isTimeout ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs: null, totalMs: null, tps: null, charsPerSec: null, tokens: null };
   }
 }

package/src/bench/workbuddy-bench.js

@@ -1,4 +1,5 @@
 import { computeMetrics } from "../metrics.js";
+import { createTransport } from "../transport/index.js";
 
 function buildWorkbuddyHeaders(apiKey, auth) {
   const h = {
@@ -16,55 +17,38 @@ function buildWorkbuddyHeaders(apiKey, auth) {
   return h;
 }
 
+function sseContent(obj) {
+  const c = obj?.choices?.[0]?.delta?.content || obj?.choices?.[0]?.message?.content || "";
+  return typeof c === "string" ? c : "";
+}
+
 export async function workbuddyBenchOne({ baseUrl, chatPath = "/v2/chat/completions", model, apiKey, auth, prompt = "hi", maxTokens = 5, timeoutMs = 30000, fetchImpl = globalThis.fetch }) {
-  const controller = new AbortController();
-  const timer = setTimeout(() => controller.abort(new Error(`timeout ${timeoutMs}ms`)), timeoutMs);
-  const t0 = typeof performance !== "undefined" && performance.now ? performance.now() : Date.now();
+  const tr = createTransport({ fetchImpl, keepAlive: false, retry: {}, timeoutMs });
   let ttfbMs = null;
+  let totalMs = null;
   let content = "";
   try {
     const url = String(baseUrl).replace(/\/+$/, "") + String(chatPath || "/v2/chat/completions");
     const headers = buildWorkbuddyHeaders(apiKey, auth);
     const rawModel = String(model || "").trim();
     const body = { model: rawModel, stream: true, messages: [{ role: "user", content: prompt }], max_tokens: maxTokens };
-    const res = await fetchImpl(url, { method: "POST", headers, body: JSON.stringify(body), signal: controller.signal });
-    if (res instanceof Error) throw res;
+    const res = await tr.request({ url, method: "POST", headers, body, stream: true });
     if (!res.ok) {
       let txt = "";
       try { txt = await res.text(); } catch {}
       const label = res.status === 401 ? "鉴权失败" : res.status === 402 ? "余额不足" : res.status === 429 ? "限流" : res.status >= 500 ? `上游错误 ${res.status}` : `HTTP ${res.status}`;
-      return { id: model, ok: false, status: res.status, label, error: txt.slice(0, 300), ttfbMs, totalMs: Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : Date.now()) - t0), tps: null, charsPerSec: null, tokens: null };
+      return { id: model, ok: false, status: res.status, label, error: txt.slice(0, 300), ttfbMs, totalMs: res.totalMs, tps: null, charsPerSec: null, tokens: null };
     }
-    // SSE stream parsing — 复用 cline 逻辑
-    const reader = res.body.getReader();
-    const decoder = new TextDecoder();
-    let buf = "";
-    for (;;) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      if (ttfbMs === null) ttfbMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : Date.now()) - t0);
-      buf += decoder.decode(value, { stream: true });
-      let idx;
-      while ((idx = buf.indexOf("\n")) >= 0) {
-        const line = buf.slice(0, idx);
-        buf = buf.slice(idx + 1);
-        if (!line.startsWith("data:")) continue;
-        const payload = line.slice(5).trim();
-        if (!payload || payload === "[DONE]") continue;
-        try {
-          const j = JSON.parse(payload);
-          const c = j?.choices?.[0]?.delta?.content || j?.choices?.[0]?.message?.content || "";
-          if (typeof c === "string") content += c;
-        } catch {}
-      }
+    for await (const ev of res.stream()) {
+      try { content += sseContent(JSON.parse(ev)); } catch {}
     }
-    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : Date.now()) - t0);
+    ttfbMs = res.ttfbMs;
+    totalMs = res.totalMs;
     const chars = content.length;
     const { tps, charsPerSec } = computeMetrics({ ttfbMs, totalMs, promptTokens: null, completionTokens: null, chars });
     return { id: model, ok: true, status: 200, label: "成功", ttfbMs, totalMs, tps, charsPerSec, tokens: null, chars };
   } catch (e) {
     const msg = e?.message || String(e);
-    const totalMs = Math.round((typeof performance !== "undefined" && performance.now ? performance.now() : Date.now()) - t0);
     return { id: model, ok: false, label: /timeout|abort/i.test(msg) ? "超时" : "网络错误", error: msg.slice(0, 300), ttfbMs, totalMs, tps: null, charsPerSec: null, tokens: null };
-  } finally { clearTimeout(timer); }
+  }
 }

package/src/chat-pipeline/engine.js

@@ -0,0 +1,241 @@
+import { performance } from "node:perf_hooks";
+import { injectReasoningContent } from "../reasoning.js";
+import { runHook } from "../plugins.js";
+import { errMsg, json } from "../routes/helpers.js";
+import { hedgeDelayMs, shouldHedge } from "../routes/hedge.js";
+import { handleHedge } from "../routes/chat/hedge-handler.js";
+import { handleLocalRelay } from "../routes/chat/local-handler.js";
+import { handlePeerRelay } from "../routes/chat/peer-handler.js";
+import { handleBroadbandRelay } from "../routes/chat/broadband-handler.js";
+import { handleViaRoute } from "../routes/chat/via-route-handler.js";
+import { handleExhaustedLocal, handleExhaustedAll } from "../routes/chat/exhausted-handler.js";
+
+/**
+ * ExecutionEngine — 统一执行：auto 并发择优 → via-route → 串行 trial → hedge/local/peer/broadband/exhausted
+ * 承接原 gateway 的 Executor 段（order/race/回退/对冲全在此），gateway 仅做薄适配。
+ */
+export function createEngine(_deps = {}) {
+  async function run(_plan, state) {
+    let order = state.order;
+    const {
+      reqId, startedAt, req, res, body, policy,
+      useAuto, lockModel, requested, hops,
+      canFallback, canForwardPeers,
+      perf0, stages, mark, evt, logCall, logError, done, handlerCtx,
+      auto, upstream, peers, groups, bus, token, plugins, logs,
+    } = state;
+    const { shareKeys, workbuddyUid } = policy;
+
+    // ===== auto 首次并发择优 =====
+    if (useAuto && order.length > 1 && auto && !lockModel) {
+      const statuses = auto.statuses?.() ?? {};
+      const hasPriorSuccess = Object.values(statuses).some((e) => e && typeof e === "object" && e.status === "normal");
+      const nonCoolingOrder = order.filter((m) => { try { return !auto.isCooling(m); } catch { return true; } });
+      if (!hasPriorSuccess && nonCoolingOrder.length > 1) {
+        const concLimit = (() => {
+          const v = Number(process.env.MSLXDFF_AUTO_CONCURRENT);
+          if (Number.isInteger(v) && v > 0) return Math.min(v, nonCoolingOrder.length);
+          return Math.min(nonCoolingOrder.length, 5);
+        })();
+        let raceModels = nonCoolingOrder.slice(0, concLimit);
+        if (plugins?.length) {
+          const k = [];
+          for (const m of raceModels) {
+            const b = await runHook(plugins, "model:beforeTry", { reqId, requested, model: m, hops });
+            if (b.value === false || b.value?.skip) continue;
+            k.push(m);
+          }
+          raceModels = k;
+        }
+        if (!raceModels.length) {
+          order = order.filter((m) => !new Set(nonCoolingOrder.slice(0, concLimit)).has(m));
+          if (!order.length) {
+            await handleExhaustedAll({ res, body, lastErr: { model: requested, status: 502, message: "all concurrent candidates skipped by plugin" }, order: nonCoolingOrder.slice(0, concLimit), requested, handlerCtx: { ...handlerCtx, reqId, startedAt }, evt, logCall, mark, perf0, stages });
+            return;
+          }
+        } else {
+          evt("auto-concurrent-race", { reqId, models: raceModels, skippedFaulty: order.length - nonCoolingOrder.length, limit: concLimit });
+          const raceStart = performance.now();
+          const attempts = raceModels.map(async (m) => {
+            let f = { ...injectReasoningContent(m, body), model: m };
+            if (plugins?.length) {
+              const u = await runHook(plugins, "upstream:request", { reqId, requested, model: m, payload: f, stream: Boolean(body.stream) });
+              if (u.changed && u.value?.payload) f = u.value.payload;
+            }
+            let r = null;
+            try {
+              const o = {};
+              if (Object.keys(shareKeys).length) o.shareKeys = shareKeys;
+              if (workbuddyUid) o.workbuddyUid = workbuddyUid;
+              r = await upstream.chat(f, Object.keys(o).length ? o : undefined);
+            } catch (e) {
+              if (plugins?.length) runHook(plugins, "upstream:response", { reqId, requested, model: m, status: null, ok: false, error: errMsg(e), timing: e?._t ?? null }).catch(() => {});
+              return { model: m, ok: false, error: errMsg(e), status: 502, timing: e?._t ?? null };
+            }
+            if (plugins?.length) runHook(plugins, "upstream:response", { reqId, requested, model: m, status: r instanceof Error ? null : r?.status ?? null, ok: !(r instanceof Error) && r ? r.status < 400 : false, error: r instanceof Error ? errMsg(r) : null, timing: r?._t ?? null }).catch(() => {});
+            if (r && r.status >= 400) {
+              const a = r.status === 403 && r.headers?.get?.("x-mslxdff-allowlist") === "1";
+              if (a) return { model: m, ok: false, error: "allowlist", status: 403, allowlist: true };
+              return { model: m, ok: false, error: `upstream ${r.status}`, status: r.status, res: r, timing: r._t ?? null };
+            }
+            if (r instanceof Error) return { model: m, ok: false, error: errMsg(r), status: 502 };
+            return { model: m, ok: true, res: r, status: r.status, timing: r._t ?? null };
+          });
+          const results = await Promise.allSettled(attempts);
+          const okList = results.map((r, i) => ({ r, i, model: raceModels[i] }))
+            .filter(({ r }) => r.status === "fulfilled" && r.value?.ok)
+            .map(({ r, i, model }) => ({ model, idx: i, val: r.value, t: r.value.timing?.totalMs ?? r.value.timing?.ms ?? Number.MAX_SAFE_INTEGER }));
+          if (okList.length) {
+            okList.sort((a, b) => a.t - b.t);
+            const best = okList[0];
+            const winModel = best.model;
+            evt("auto-concurrent-win", { reqId, model: winModel, timing: best.val.timing, totalMs: Math.round(performance.now() - raceStart), tried: raceModels.length });
+            for (const { r, i } of results.map((r, i) => ({ r, i }))) {
+              const m = raceModels[i];
+              if (r.status === "fulfilled" && r.value?.ok) {
+                if (m === winModel) {
+                  const latencyMs = r.value.timing?.totalMs ?? Math.round(performance.now() - raceStart);
+                  await auto.recordOk(m, { latencyMs });
+                  try { const { savePreferredModel } = await import("../state.js"); savePreferredModel(m); evt("auto-concurrent-preferred", { reqId, model: m }); } catch {}
+                }
+              } else if (r.status === "fulfilled" && !r.value?.ok && !r.value?.allowlist) await auto.recordError(m, { status: r.value.status || 502 });
+              else if (r.status === "rejected") await auto.recordError(m, { status: 502 });
+            }
+            handlerCtx.model = winModel;
+            const lr = await handleLocalRelay({ upRes: best.val.res, model: winModel, body, order: raceModels, idx: best.idx, lastErr: null, requested, useAuto, lockModel, auto, handlerCtx, evt, logCall, logError, mark, perf0, stages, startedAt, plugins, res });
+            if (lr.handled) return;
+            if (!lr.lastErr) return;
+          } else {
+            evt("auto-concurrent-all-fail", { reqId, tried: raceModels.length, totalMs: Math.round(performance.now() - raceStart) });
+            for (const { r, i } of results.map((r, i) => ({ r, i }))) {
+              const m = raceModels[i];
+              if (r.status === "fulfilled" && !r.value?.ok && !r.value?.allowlist) await auto.recordError(m, { status: r.value.status || 502 });
+              else if (r.status === "rejected") await auto.recordError(m, { status: 502 });
+            }
+          }
+          const triedSet = new Set(raceModels);
+          order = order.filter((m) => !triedSet.has(m));
+          if (!order.length) {
+            const failedStatuses = results.map((r) => (r.status === "fulfilled" ? r.value?.status : null)).filter((s) => Number.isInteger(s));
+            const lastStatus = failedStatuses[failedStatuses.length - 1] || failedStatuses[0] || 502;
+            const last = { model: raceModels[0] || requested, status: lastStatus, message: "all concurrent candidates failed" };
+            await handleExhaustedAll({ res, body, lastErr: last, order: raceModels, requested, handlerCtx: { ...handlerCtx, reqId, startedAt }, evt, logCall, mark, perf0, stages });
+            return;
+          }
+        }
+      }
+    }
+
+    // ===== VIA-ROUTE 单路径择路（显式锁模型，不并发） =====
+    let viaRouteLastErr = null;
+    if (!useAuto && requested && requested.includes("/") && canForwardPeers && !lockModel && peers) {
+      try {
+        const vr = await handleViaRoute({ model: requested, body, peers, handlerCtx, evt, logCall, logError, mark, perf0, stages, startedAt, plugins, res, requested, useAuto, lockModel, auto });
+        if (vr.handled) return;
+        if (vr.lastErr) viaRouteLastErr = vr.lastErr;
+      } catch (e) {
+        evt("via-route-exception", { reqId, model: requested, error: errMsg(e) });
+      }
+    }
+
+    // ===== 串行 trial =====
+    let lastErr = viaRouteLastErr;
+    for (let idx = 0; idx < order.length; idx++) {
+      const model = order[idx];
+      handlerCtx.model = model;
+      evt("model-try", { reqId, model, idx, remaining: order.length - idx });
+      if (plugins?.length) {
+        const bt = await runHook(plugins, "model:beforeTry", { reqId, requested, model, idx, hops });
+        for (const e of bt.errors) evt("plugin-hook-error", { reqId, hook: "model:beforeTry", plugin: e.plugin, error: e.error });
+        if (bt.value === false || bt.value?.skip === true) { evt("plugin-hook", { reqId, hook: "model:beforeTry", applied: true, skipped: model }); continue; }
+      }
+      let upRes = null;
+      let forwarded = { ...injectReasoningContent(model, body), model };
+      if (plugins?.length) {
+        const ur = await runHook(plugins, "upstream:request", { reqId, requested, model, payload: forwarded, stream: Boolean(body.stream) });
+        for (const e of ur.errors) evt("plugin-hook-error", { reqId, hook: "upstream:request", plugin: e.plugin, error: e.error });
+        if (ur.changed && ur.value?.payload && typeof ur.value.payload === "object") { forwarded = ur.value.payload; evt("plugin-hook", { reqId, hook: "upstream:request", applied: true, model, rewrittenModel: forwarded.model ?? null }); }
+      }
+      const tUp = performance.now();
+      evt("upstream-try", { reqId, model, attempt: idx + 1 });
+      try {
+        const chatOpts = {};
+        if (Object.keys(shareKeys).length) chatOpts.shareKeys = shareKeys;
+        if (workbuddyUid) chatOpts.workbuddyUid = workbuddyUid;
+        upRes = await upstream.chat(forwarded, Object.keys(chatOpts).length ? chatOpts : undefined);
+        evt("upstream-done", { reqId, model, ok: !(upRes instanceof Error) && upRes.status < 400, status: upRes instanceof Error ? null : upRes.status, timing: upRes._t ?? null, error: null });
+      } catch (err) {
+        if (auto) await auto.recordError(model, { message: errMsg(err) });
+        lastErr = { model, upstream: null, status: 502, message: errMsg(err) };
+        logError(model, 502, errMsg(err));
+        evt("upstream-error", { reqId, model, status: 502, message: errMsg(err), timing: err._t ?? { attempts: [], waitMs: 0, totalMs: Math.round(performance.now() - tUp) } });
+      }
+      if (plugins?.length) {
+        runHook(plugins, "upstream:response", {
+          reqId, requested, model,
+          status: upRes instanceof Error ? null : upRes instanceof Object ? (upRes.status ?? null) : null,
+          ok: !(upRes instanceof Error) && upRes ? upRes.status < 400 : false,
+          error: upRes instanceof Error ? errMsg(upRes) : null,
+          timing: upRes?._t ?? null,
+        }).catch(() => {});
+      }
+      mark(`up-${model}`);
+      if (upRes && upRes.status >= 400) {
+        const isAllowlistBlock = upRes.status === 403 && (upRes.headers?.get?.("x-mslxdff-allowlist") === "1");
+        if (isAllowlistBlock) {
+          let bodyText = null; try { bodyText = await upRes.clone().text(); } catch {}
+          let errBody = { error: `model not allowed for provider` };
+          try { errBody = bodyText ? JSON.parse(bodyText) : errBody; } catch { errBody = { error: bodyText || "model not allowed" }; }
+          if (useAuto) {
+            logError(model, 403, errBody.error || "model not allowed");
+            evt("upstream-error", { reqId, model, status: 403, message: errBody.error, timing: upRes._t ?? null, allowlist: true, skipped: true });
+            lastErr = { model, upstream: upRes, status: 403, message: errBody.error || "model not allowed" };
+            if (canFallback && idx < order.length - 1) { evt("fallback", { reqId, from: model, to: order[idx + 1] ?? null, reason: `allowlist skip ${errBody.error || "blocked"}` }); continue; }
+            return json(res, 403, errBody);
+          }
+          logError(model, 403, errBody.error || "model not allowed");
+          evt("upstream-error", { reqId, model, status: 403, message: errBody.error, timing: upRes._t ?? null, allowlist: true });
+          return json(res, 403, errBody);
+        }
+        if (auto) await auto.recordError(model, { status: upRes.status });
+        lastErr = { model, upstream: upRes, status: upRes.status, message: null };
+        logError(model, upRes.status, `upstream ${upRes.status}`);
+        evt("upstream-error", { reqId, model, status: upRes.status, message: null, timing: upRes._t ?? null });
+        upRes = null;
+      }
+      if (upRes) {
+        const isStream = Boolean(body.stream);
+        const d = hedgeDelayMs();
+        const hasPeers = Boolean(peers) && peers.ordered().length > 0;
+        const doHedge = shouldHedge({ isStream, canForwardPeers, hedgeDelayMs: d, hasPeers }) && upRes.status === 200 && upRes.body;
+        if (doHedge) {
+          const hr = await handleHedge({ upRes, model, body, order, idx, lastErr, requested, useAuto, lockModel, auto, peers, handlerCtx, evt, logCall, logError, mark, perf0, stages, startedAt, plugins, res, hedgeDelayMs: d });
+          if (hr.handled) return;
+          if (hr.lastErr) lastErr = hr.lastErr;
+          if (hr.upRes === null) upRes = null;
+          else if (hr.upRes) upRes = hr.upRes;
+        }
+        if (upRes) {
+          const lr = await handleLocalRelay({ upRes, model, body, order, idx, lastErr, requested, useAuto, lockModel, auto, handlerCtx, evt, logCall, logError, mark, perf0, stages, startedAt, plugins, res });
+          if (lr.handled) return;
+          if (lr.lastErr) { lastErr = lr.lastErr; continue; }
+          return;
+        }
+      }
+      if (canForwardPeers) {
+        const pr = await handlePeerRelay({ model, body, lastErr, requested, useAuto, lockModel, auto, peers, handlerCtx, evt, logCall, mark, perf0, stages, startedAt, plugins, res });
+        if (pr.handled) return;
+      }
+      if (groups) {
+        const br = await handleBroadbandRelay({ model, body, hops, lastErr, requested, useAuto, lockModel, auto, groups, token, bus, logs, handlerCtx, evt, mark, perf0, stages, res, startedAt, plugins });
+        if (br.handled) return;
+      }
+      if (canFallback) { evt("fallback", { reqId, from: model, to: order[idx + 1] ?? null, reason: lastErr?.message || `upstream ${lastErr?.status ?? 502}` }); continue; }
+      await handleExhaustedLocal({ res, body, lastErr, order, handlerCtx: { ...handlerCtx, model, reqId }, evt, logCall, mark, perf0, stages, done, requested, useAuto });
+      return;
+    }
+    await handleExhaustedAll({ res, body, lastErr, order, requested, handlerCtx: { ...handlerCtx, reqId, startedAt }, evt, logCall, mark, perf0, stages });
+  }
+
+  return { run };
+}