mslxdff 0.1.161 → 0.1.162

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (93) hide show
  1. package/bin/mslxdff.js +4 -4
  2. package/package.json +1 -1
  3. package/src/auto.js +254 -254
  4. package/src/bench/cline-bench.js +42 -42
  5. package/src/bench/probe.js +70 -70
  6. package/src/bench/report.js +162 -162
  7. package/src/bench/runner.js +77 -77
  8. package/src/bench/via-probe.js +124 -124
  9. package/src/bench/via-routes.js +87 -87
  10. package/src/bench/workbuddy-bench.js +54 -54
  11. package/src/chat/engine.js +160 -160
  12. package/src/chat/gateway.js +163 -163
  13. package/src/chat/orchestrator.js +234 -234
  14. package/src/chat/prompt.js +70 -70
  15. package/src/chat/repl.js +88 -88
  16. package/src/chat/terminal.js +135 -135
  17. package/src/chat/tools.js +306 -306
  18. package/src/chat-pipeline/index.js +126 -126
  19. package/src/chat-pipeline/policy.js +76 -76
  20. package/src/cli/commands/group.js +249 -249
  21. package/src/cli/commands/model/picks.js +50 -50
  22. package/src/cli/commands/provider/bench-via.js +247 -247
  23. package/src/cli/commands/provider/bench.js +141 -141
  24. package/src/cli/commands/provider/index.js +126 -126
  25. package/src/cli/commands/provider/models.js +142 -142
  26. package/src/cli/commands/provider/qwenwork-login.js +119 -119
  27. package/src/cli/commands/sync.js +232 -232
  28. package/src/cli/format.js +1 -0
  29. package/src/cli/status.js +279 -279
  30. package/src/daemon.js +101 -101
  31. package/src/model-capabilities/enrich.js +86 -86
  32. package/src/model-capabilities/index.js +183 -183
  33. package/src/model-capabilities/parse.js +70 -70
  34. package/src/model-trace.js +4 -3
  35. package/src/models.js +225 -225
  36. package/src/providers/cline/auth.js +228 -228
  37. package/src/providers/cline/chat.js +307 -307
  38. package/src/providers/cline.js +2 -2
  39. package/src/providers/dispatcher.js +26 -14
  40. package/src/providers/globalqwenwork/cosy.js +123 -123
  41. package/src/providers/globalqwenwork/crypto.js +218 -218
  42. package/src/providers/globalqwenwork/http.js +20 -20
  43. package/src/providers/globalqwenwork/rsa.js +54 -54
  44. package/src/providers/globalqwenwork/sse.js +268 -268
  45. package/src/providers/globalqwenwork/stream.js +130 -130
  46. package/src/providers/globalqwenwork/upstream.js +122 -122
  47. package/src/providers/keyring.js +60 -60
  48. package/src/providers/qoder/chat.js +193 -183
  49. package/src/providers/qoder/index.js +239 -230
  50. package/src/providers/qoder/sse.js +141 -103
  51. package/src/providers/qwenwork/account-store.js +133 -133
  52. package/src/providers/qwenwork/constants.js +67 -67
  53. package/src/providers/qwenwork/cosy.js +120 -120
  54. package/src/providers/qwenwork/crypto.js +218 -218
  55. package/src/providers/qwenwork/http.js +20 -20
  56. package/src/providers/qwenwork/index.js +327 -327
  57. package/src/providers/qwenwork/payload.js +142 -142
  58. package/src/providers/qwenwork/rsa.js +54 -54
  59. package/src/providers/qwenwork/sse.js +268 -268
  60. package/src/providers/qwenwork/stream.js +130 -130
  61. package/src/providers/qwenwork/upstream.js +120 -120
  62. package/src/providers/qwenwork.js +1 -1
  63. package/src/providers/registry.js +74 -74
  64. package/src/providers/workbuddy/chat.js +248 -248
  65. package/src/providers/workbuddy/reshape.js +152 -152
  66. package/src/providers/workbuddy.js +2 -2
  67. package/src/reasoning.js +32 -32
  68. package/src/routes/chat/gateway.js +46 -46
  69. package/src/routes/chat/relay-pipeline.js +276 -276
  70. package/src/routes/chat/via-route-handler.js +146 -146
  71. package/src/routes/hedge.js +255 -255
  72. package/src/routes/models-route.js +167 -167
  73. package/src/routes/peers.js +273 -273
  74. package/src/routes/stream.js +386 -386
  75. package/src/runtime/bootstrap.js +45 -45
  76. package/src/runtime/provider-gate.js +34 -34
  77. package/src/runtime/providers-setup.js +165 -165
  78. package/src/runtime/server-lifecycle.js +5 -6
  79. package/src/server.js +64 -64
  80. package/src/state/schemas/allowlist.js +92 -92
  81. package/src/sync-opencode.js +280 -280
  82. package/src/transport/index.js +244 -244
  83. package/src/transport/pool.js +56 -56
  84. package/src/transport/retry.js +24 -24
  85. package/src/transport/sse.js +93 -93
  86. package/src/upstream-probe/display.js +52 -52
  87. package/src/upstream-probe/probe.js +49 -49
  88. package/src/upstream-probe/rotate.js +110 -110
  89. package/src/upstream-probe/start.js +45 -45
  90. package/src/upstream.js +289 -289
  91. package/docs/cli_help_mini.md +0 -135
  92. package/src/providers/AGENTS.md +0 -46
  93. package/src/routes/AGENTS.md +0 -37
package/src/upstream.js CHANGED
@@ -1,291 +1,291 @@
1
- import crypto from "node:crypto";
2
- import { mkdirSync } from "node:fs";
3
- import { appendFile } from "node:fs/promises";
4
- import { dirname, join } from "node:path";
5
- import { performance } from "node:perf_hooks";
6
- import { isFreeModel } from "./models.js";
7
- import { fmtShanghaiYMDHMS } from "./time.js";
8
- import { createTransport } from "./transport/index.js";
9
- import { isResponsesModel, chatToResponsesBody, toChatResponse, reshapeResponsesSse } from "./upstream-responses.js";
10
- import { ensureFreeLaneShape, aggregateChatSse } from "./free-lane.js";
11
- import { digestIdTail, genId, opencodeClientIdentity, opencodeUa } from "./opencode-identity.js";
12
-
1
+ import crypto from "node:crypto";
2
+ import { mkdirSync } from "node:fs";
3
+ import { appendFile } from "node:fs/promises";
4
+ import { dirname, join } from "node:path";
5
+ import { performance } from "node:perf_hooks";
6
+ import { isFreeModel } from "./models.js";
7
+ import { fmtShanghaiYMDHMS } from "./time.js";
8
+ import { createTransport } from "./transport/index.js";
9
+ import { isResponsesModel, chatToResponsesBody, toChatResponse, reshapeResponsesSse } from "./upstream-responses.js";
10
+ import { ensureFreeLaneShape, aggregateChatSse } from "./free-lane.js";
11
+ import { digestIdTail, genId, opencodeClientIdentity, opencodeUa } from "./opencode-identity.js";
12
+
13
13
  export { opencodeClientIdentity, opencodeUa };
14
- // 上游按 session 做粘性路由(实测:固定 session 两次请求均 ~1.2s;每次随机时可能撞冷机器 26s+)。
15
- // 客户端(opencode AI SDK 路径)不带会话标识 → 用对话首两条消息(system + 首条 user)哈希做稳定会话:
16
- // 同一会话多轮里这两条不变 ⇒ 路由亲和稳定;不同会话天然分散。
17
- function sessionFromMessages(messages) {
18
- try {
19
- const msgs = Array.isArray(messages) ? messages : [];
20
- const pick = (role) => {
21
- const m = msgs.find((x) => x?.role === role);
22
- if (!m) return "";
23
- return typeof m.content === "string" ? m.content : JSON.stringify(m.content ?? "");
24
- };
25
- const seed = `${pick("system")}|${pick("user")}`.slice(0, 4000);
26
- if (seed === "|") return null;
27
- return `ses_${digestIdTail(crypto.createHash("sha1").update(seed).digest())}`;
28
- } catch {
29
- return null;
30
- }
31
- }
32
- function envInt(name, fallback) {
33
- const v = Number(process.env[name]);
34
- return Number.isInteger(v) && v > 0 ? v : fallback;
35
- }
36
- function isPreheatDisabled() {
37
- const raw = process.env.MSLXDFF_PREHEAT;
38
- if (raw === undefined || raw === null || raw === "") return false;
39
- const s = String(raw).trim().toLowerCase();
40
- return s === "0" || s === "off" || s === "false" || s === "no" || s === "disable" || s === "disabled";
41
- }
42
- // opencode 上游身份头单一来源(legacy 与 SDK 引擎共用,free 层门禁头必须一致)。
43
- // anonFirst 语义:authToken==="public" 时默认匿名(hermes referer);env 可显式覆盖。
44
- export function createOpencodeHeaderBuilder({ authToken = "public", env = process.env } = {}) {
45
- function isAnonFirst() {
46
- const raw = env.MSLXDFF_OPENCOD_ANON_FIRST ?? env.MSLXDFF_ANON_FIRST ?? env.MSLXDFF_OPENCOD_ANON;
47
- if (raw === undefined || raw === null || raw === "") return authToken === "public";
48
- const s = String(raw).trim().toLowerCase();
49
- if (s === "0" || s === "false" || s === "off" || s === "no" || s === "disable" || s === "disabled") return false;
50
- return authToken === "public";
51
- }
52
- const anonFirst = isAnonFirst();
53
- const baseHeaders = anonFirst
54
- ? {
55
- "Content-Type": "application/json",
56
- Authorization: "",
57
- "x-opencode-client": "desktop",
58
- "User-Agent": opencodeUa(),
59
- "HTTP-Referer": "https://hermes-agent.nousresearch.com",
60
- "X-Title": "Hermes Agent",
61
- }
62
- : {
63
- "Content-Type": "application/json",
64
- Authorization: `Bearer ${authToken}`,
65
- "x-opencode-client": "desktop",
66
- };
67
- // 无 messages 可哈希时的进程级兜底:至少 daemon 生命周期内稳定,不再每请求随机
68
- const FALLBACK_SESSION = genId("ses_");
69
- function buildHeaders(body, { anonymous = anonFirst, sessionId = null } = {}) {
70
- const isStream = body?.stream !== false;
71
- const session = sessionId || sessionFromMessages(body?.messages) || FALLBACK_SESSION;
72
- const base = {
73
- ...baseHeaders,
74
- Accept: isStream ? "text/event-stream" : "*/*",
75
- "User-Agent": opencodeUa(),
76
- "x-opencode-session": session,
77
- "x-opencode-request": genId("msg_"),
78
- "x-opencode-project": "global",
79
- };
80
- if (anonymous) {
81
- return { ...base, Authorization: "", "HTTP-Referer": "https://hermes-agent.nousresearch.com", "X-Title": "Hermes Agent" };
82
- }
83
- if (anonFirst) {
84
- const { "HTTP-Referer": _a, "X-Title": _b, ...rest } = base;
85
- return { ...rest, Authorization: `Bearer ${authToken}` };
86
- }
87
- return base;
88
- }
89
- return { buildHeaders, anonFirst };
90
- }
91
-
92
- export function createUpstreamClient({
93
- baseUrl = process.env.UPSTREAM_BASE_URL || "https://opencode.ai",
94
- authToken = process.env.UPSTREAM_AUTH_TOKEN || "public",
95
- connectTimeoutMs = Number(process.env.UPSTREAM_CONNECT_TIMEOUT_MS) || 30_000,
96
- retry = {
97
- network: { attempts: 2, delayMs: 300 },
98
- 429: { attempts: 1, delayMs: 100 },
99
- 502: { attempts: 1, delayMs: 100 },
100
- 503: { attempts: 1, delayMs: 100 },
101
- 504: { attempts: 1, delayMs: 100 },
102
- },
103
- fetchImpl,
104
- hooks,
105
- keepAlive = true,
106
- } = {}) {
107
- const { buildHeaders, anonFirst } = createOpencodeHeaderBuilder({ authToken });
108
- function shouldTryAnonFree() {
109
- const raw = process.env.MSLXDFF_FREE_ANON;
110
- if (raw === "0" || raw === "off" || raw === "false" || raw === "no") return false;
111
- return true;
112
- }
113
- function freeAnonLogFile() {
114
- return process.env.MSLXDFF_FREE_ANON_LOG || join(process.cwd(), "free-anon-extra.txt");
115
- }
116
- function logFreeAnon({ model, publicStatus, anonStatus, anonAttempts, hit, totalMs }) {
117
- const file = freeAnonLogFile();
118
- const line = `${fmtShanghaiYMDHMS(new Date())} model=${model} public=429 anonTries=${anonAttempts} anonStatus=${anonStatus ?? "none"} hit=${hit ? "YES额外额度" : "NO无额外"} totalMs=${totalMs} count=${hit ? "1" : "0"}\n`;
119
- try { mkdirSync(dirname(file), { recursive: true }); } catch {}
120
- appendFile(file, line).catch(() => {});
121
- try { if (process.env.MSLXDFF_DEBUG === "1") console.log(`[free-anon] ${line.trim()}`); } catch {}
122
- }
123
- let consecutiveHits = 0;
124
-
125
- // 深模块:网络层委托 StreamingTransport
126
- const transport = createTransport({
127
- keepAlive,
128
- fetchImpl,
129
- timeoutMs: connectTimeoutMs,
130
- retry,
131
- hooks,
132
- });
133
-
134
- async function chat(body, opts = {}) {
135
- const isResp = isResponsesModel(body?.model);
136
- // responses 路径的 reqBody 无 messages,会话哈希必须基于原始 chat body 计算
137
- const sessionId = opts?.sessionId || sessionFromMessages(body?.messages) || null;
138
- const url = isResp ? `${baseUrl}/zen/v1/responses` : `${baseUrl}/zen/v1/chat/completions`;
139
- const reqBody = isResp ? chatToResponsesBody(body) : body;
140
- const clientWantsStream = body?.stream !== false;
141
- // zen 免费层 agent 形状门禁(2026-09-18):必须流式 + 含核心五工具名,否则 403 FreeTierError
142
- const freeLane = authToken === "public" && isFreeModel(body?.model);
143
- if (freeLane) ensureFreeLaneShape(reqBody, { responses: isResp });
144
- const laneDebug = freeLane && envInt("MSLXDFF_FREE_LANE_DEBUG", 0) === 1;
145
- if (laneDebug) {
146
- try { console.error(`[free-lane] send model=${body?.model} url=${url} stream=${reqBody.stream} tools=${(reqBody.tools || []).length} ua=${opencodeUa()} want=${clientWantsStream}`); } catch {}
147
- }
148
- // 非流式调用者实际拿到的是被强制流式的上游 → 聚合回 chat completion JSON
149
- async function agentJson(resp) {
150
- if (!(freeLane && !clientWantsStream && resp.ok)) return resp;
151
- const ct = resp.headers?.get?.("content-type") || "";
152
- if (!ct.includes("text/event-stream")) return resp;
153
- try {
154
- const out = await aggregateChatSse(resp);
155
- out._t = resp._t;
156
- return out;
157
- } catch {
158
- return resp;
159
- }
160
- }
161
- const t0 = performance.now();
162
-
163
- // 首发请求(transport 已处理 network/429 等重试)
164
- let res;
165
- try {
166
- res = await transport.request({
167
- url,
168
- method: "POST",
169
- headers: buildHeaders(reqBody, { sessionId }),
170
- body: reqBody,
171
- stream: reqBody.stream !== false,
172
- timeoutMs: connectTimeoutMs,
173
- retry,
174
- });
175
- } catch (e) {
176
- e._t = e._t || { attempts: [], waitMs: 0, totalMs: Math.round(performance.now() - t0) };
177
- throw e;
178
- }
179
- if (laneDebug) {
180
- try { console.error(`[free-lane] resp model=${body?.model} status=${res.status} ct=${res.headers?.get?.("content-type") || ""}`); } catch {}
14
+ // 上游按 session 做粘性路由(实测:固定 session 两次请求均 ~1.2s;每次随机时可能撞冷机器 26s+)。
15
+ // 客户端(opencode AI SDK 路径)不带会话标识 → 用对话首两条消息(system + 首条 user)哈希做稳定会话:
16
+ // 同一会话多轮里这两条不变 ⇒ 路由亲和稳定;不同会话天然分散。
17
+ function sessionFromMessages(messages) {
18
+ try {
19
+ const msgs = Array.isArray(messages) ? messages : [];
20
+ const pick = (role) => {
21
+ const m = msgs.find((x) => x?.role === role);
22
+ if (!m) return "";
23
+ return typeof m.content === "string" ? m.content : JSON.stringify(m.content ?? "");
24
+ };
25
+ const seed = `${pick("system")}|${pick("user")}`.slice(0, 4000);
26
+ if (seed === "|") return null;
27
+ return `ses_${digestIdTail(crypto.createHash("sha1").update(seed).digest())}`;
28
+ } catch {
29
+ return null;
30
+ }
31
+ }
32
+ function envInt(name, fallback) {
33
+ const v = Number(process.env[name]);
34
+ return Number.isInteger(v) && v > 0 ? v : fallback;
35
+ }
36
+ function isPreheatDisabled() {
37
+ const raw = process.env.MSLXDFF_PREHEAT;
38
+ if (raw === undefined || raw === null || raw === "") return false;
39
+ const s = String(raw).trim().toLowerCase();
40
+ return s === "0" || s === "off" || s === "false" || s === "no" || s === "disable" || s === "disabled";
41
+ }
42
+ // opencode 上游身份头单一来源(legacy 与 SDK 引擎共用,free 层门禁头必须一致)。
43
+ // anonFirst 语义:authToken==="public" 时默认匿名(hermes referer);env 可显式覆盖。
44
+ export function createOpencodeHeaderBuilder({ authToken = "public", env = process.env } = {}) {
45
+ function isAnonFirst() {
46
+ const raw = env.MSLXDFF_OPENCOD_ANON_FIRST ?? env.MSLXDFF_ANON_FIRST ?? env.MSLXDFF_OPENCOD_ANON;
47
+ if (raw === undefined || raw === null || raw === "") return authToken === "public";
48
+ const s = String(raw).trim().toLowerCase();
49
+ if (s === "0" || s === "false" || s === "off" || s === "no" || s === "disable" || s === "disabled") return false;
50
+ return authToken === "public";
51
+ }
52
+ const anonFirst = isAnonFirst();
53
+ const baseHeaders = anonFirst
54
+ ? {
55
+ "Content-Type": "application/json",
56
+ Authorization: "",
57
+ "x-opencode-client": "desktop",
58
+ "User-Agent": opencodeUa(),
59
+ "HTTP-Referer": "https://hermes-agent.nousresearch.com",
60
+ "X-Title": "Hermes Agent",
61
+ }
62
+ : {
63
+ "Content-Type": "application/json",
64
+ Authorization: `Bearer ${authToken}`,
65
+ "x-opencode-client": "desktop",
66
+ };
67
+ // 无 messages 可哈希时的进程级兜底:至少 daemon 生命周期内稳定,不再每请求随机
68
+ const FALLBACK_SESSION = genId("ses_");
69
+ function buildHeaders(body, { anonymous = anonFirst, sessionId = null } = {}) {
70
+ const isStream = body?.stream !== false;
71
+ const session = sessionId || sessionFromMessages(body?.messages) || FALLBACK_SESSION;
72
+ const base = {
73
+ ...baseHeaders,
74
+ Accept: isStream ? "text/event-stream" : "*/*",
75
+ "User-Agent": opencodeUa(),
76
+ "x-opencode-session": session,
77
+ "x-opencode-request": genId("msg_"),
78
+ "x-opencode-project": "global",
79
+ };
80
+ if (anonymous) {
81
+ return { ...base, Authorization: "", "HTTP-Referer": "https://hermes-agent.nousresearch.com", "X-Title": "Hermes Agent" };
181
82
  }
182
-
183
- // 匿名兜底:仅非 anonFirst 时,public 429 + free 模型才走 hermes 空头重试
184
- if (!anonFirst && res.status === 429 && isFreeModel(body?.model) && shouldTryAnonFree()) {
185
- const anonRetries = envInt("MSLXDFF_FREE_ANON_RETRIES", 3);
186
- const anonDelay = envInt("MSLXDFF_FREE_ANON_DELAY_MS", 1000);
187
- let anonRes = null;
188
- for (let i = 0; i < anonRetries; i++) {
189
- await new Promise((r) => setTimeout(r, anonDelay));
190
- try {
191
- anonRes = await transport.request({
192
- url,
193
- method: "POST",
194
- headers: buildHeaders(reqBody, { anonymous: true, sessionId }),
195
- body: reqBody,
196
- stream: reqBody.stream !== false,
197
- timeoutMs: connectTimeoutMs,
198
- retry,
199
- });
200
- } catch (e) {
201
- continue;
202
- }
203
- if (anonRes.status !== 429) {
204
- // responses 模型需转回 chat 形状(复用 upstream-responses)
205
- let outAnon = anonRes;
206
- if (isResp && anonRes.ok) {
207
- const ctAnon = anonRes.headers.get("content-type") || "";
208
- const isStreamAnon = (clientWantsStream || freeLane) && ctAnon.includes("text/event-stream");
209
- if (isStreamAnon) {
210
- outAnon = reshapeResponsesSse(anonRes, body.model);
211
- } else {
212
- try {
213
- const txt = await anonRes.text();
214
- const j = JSON.parse(txt);
215
- if (j && Array.isArray(j.output)) {
216
- outAnon = toChatResponse(anonRes, j);
217
- } else {
218
- outAnon = new Response(txt, { status: anonRes.status, headers: anonRes.headers });
219
- }
220
- } catch { outAnon = anonRes; }
221
- }
222
- }
223
- outAnon._t = { ...(outAnon._t || {}), anonTried: true, anonAttempts: i + 1, totalMs: Math.round(performance.now() - t0) };
224
- consecutiveHits += 1;
225
- logFreeAnon({ model: body?.model, publicStatus: 429, anonStatus: anonRes.status, anonAttempts: i + 1, hit: true, totalMs: outAnon._t.totalMs });
226
- if (consecutiveHits >= 2) {
227
- try { appendFile(freeAnonLogFile(), ` -> 连续额外额度 ${consecutiveHits} 次\n`).catch(() => {}); } catch {}
228
- }
229
- return await agentJson(outAnon);
230
- }
231
- }
232
- if (anonRes) {
233
- anonRes._t = { ...(anonRes._t || {}), anonTried: true, anonAttempts: anonRetries, totalMs: Math.round(performance.now() - t0) };
234
- consecutiveHits = 0;
235
- logFreeAnon({ model: body?.model, publicStatus: 429, anonStatus: anonRes.status, anonAttempts: anonRetries, hit: false, totalMs: anonRes._t.totalMs });
236
- return anonRes;
237
- }
238
- }
239
-
240
- // responses 模型成功态转 chat(复用 upstream-responses)
241
- if (isResp && res.ok) {
242
- const ct = res.headers.get("content-type") || "";
243
- const isStream = (clientWantsStream || freeLane) && ct.includes("text/event-stream");
244
- if (isStream) {
245
- const transformed = reshapeResponsesSse(res, body.model);
246
- transformed._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
247
- return await agentJson(transformed);
248
- }
249
- try {
250
- const txt = await res.text();
251
- const j = JSON.parse(txt);
252
- if (j && Array.isArray(j.output)) {
253
- const transformed = toChatResponse(res, j);
254
- transformed._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
255
- return transformed;
256
- }
257
- const fallback = new Response(txt, { status: res.status, headers: res.headers });
258
- fallback._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
259
- return fallback;
260
- } catch {
261
- res._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
262
- return res;
263
- }
264
- }
265
- res._t = { ...(res._t || {}), totalMs: res._t?.totalMs ?? Math.round(performance.now() - t0) };
266
- return await agentJson(res);
267
- }
268
-
269
- async function preheat() {
270
- if (isPreheatDisabled()) return { ok: false, skipped: true, reason: "disabled" };
271
- const t0 = performance.now();
272
- try {
273
- const res = await transport.request({
274
- url: `${baseUrl}/zen/v1/models`,
275
- method: "GET",
276
- headers: buildHeaders({ stream: false }),
277
- stream: false,
278
- timeoutMs: 3000,
279
- });
280
- try { await res.text().catch(() => {}); } catch {}
281
- return { ok: res.ok, status: res.status, ms: Math.round(performance.now() - t0) };
282
- } catch (e) {
283
- return { ok: false, error: String(e?.message || e), ms: Math.round(performance.now() - t0) };
284
- }
285
- }
286
- async function close() {
287
- await transport.close();
288
- }
289
- const headers = buildHeaders({});
290
- return { chat, preheat, close, headers, buildHeaders, get dispatcher() { return transport.dispatcher; }, get agent() { return transport.agent; }, [Symbol.asyncDispose]: close };
291
- }
83
+ if (anonFirst) {
84
+ const { "HTTP-Referer": _a, "X-Title": _b, ...rest } = base;
85
+ return { ...rest, Authorization: `Bearer ${authToken}` };
86
+ }
87
+ return base;
88
+ }
89
+ return { buildHeaders, anonFirst };
90
+ }
91
+
92
+ export function createUpstreamClient({
93
+ baseUrl = process.env.UPSTREAM_BASE_URL || "https://opencode.ai",
94
+ authToken = process.env.UPSTREAM_AUTH_TOKEN || "public",
95
+ connectTimeoutMs = Number(process.env.UPSTREAM_CONNECT_TIMEOUT_MS) || 30_000,
96
+ retry = {
97
+ network: { attempts: 2, delayMs: 300 },
98
+ 429: { attempts: 1, delayMs: 100 },
99
+ 502: { attempts: 1, delayMs: 100 },
100
+ 503: { attempts: 1, delayMs: 100 },
101
+ 504: { attempts: 1, delayMs: 100 },
102
+ },
103
+ fetchImpl,
104
+ hooks,
105
+ keepAlive = true,
106
+ } = {}) {
107
+ const { buildHeaders, anonFirst } = createOpencodeHeaderBuilder({ authToken });
108
+ function shouldTryAnonFree() {
109
+ const raw = process.env.MSLXDFF_FREE_ANON;
110
+ if (raw === "0" || raw === "off" || raw === "false" || raw === "no") return false;
111
+ return true;
112
+ }
113
+ function freeAnonLogFile() {
114
+ return process.env.MSLXDFF_FREE_ANON_LOG || join(process.cwd(), "free-anon-extra.txt");
115
+ }
116
+ function logFreeAnon({ model, publicStatus, anonStatus, anonAttempts, hit, totalMs }) {
117
+ const file = freeAnonLogFile();
118
+ const line = `${fmtShanghaiYMDHMS(new Date())} model=${model} public=429 anonTries=${anonAttempts} anonStatus=${anonStatus ?? "none"} hit=${hit ? "YES额外额度" : "NO无额外"} totalMs=${totalMs} count=${hit ? "1" : "0"}\n`;
119
+ try { mkdirSync(dirname(file), { recursive: true }); } catch {}
120
+ appendFile(file, line).catch(() => {});
121
+ try { if (process.env.MSLXDFF_DEBUG === "1") console.log(`[free-anon] ${line.trim()}`); } catch {}
122
+ }
123
+ let consecutiveHits = 0;
124
+
125
+ // 深模块:网络层委托 StreamingTransport
126
+ const transport = createTransport({
127
+ keepAlive,
128
+ fetchImpl,
129
+ timeoutMs: connectTimeoutMs,
130
+ retry,
131
+ hooks,
132
+ });
133
+
134
+ async function chat(body, opts = {}) {
135
+ const isResp = isResponsesModel(body?.model);
136
+ // responses 路径的 reqBody 无 messages,会话哈希必须基于原始 chat body 计算
137
+ const sessionId = opts?.sessionId || sessionFromMessages(body?.messages) || null;
138
+ const url = isResp ? `${baseUrl}/zen/v1/responses` : `${baseUrl}/zen/v1/chat/completions`;
139
+ const reqBody = isResp ? chatToResponsesBody(body) : body;
140
+ const clientWantsStream = body?.stream !== false;
141
+ // zen 免费层 agent 形状门禁(2026-09-18):必须流式 + 含核心五工具名,否则 403 FreeTierError
142
+ const freeLane = authToken === "public" && isFreeModel(body?.model);
143
+ if (freeLane) ensureFreeLaneShape(reqBody, { responses: isResp });
144
+ const laneDebug = freeLane && envInt("MSLXDFF_FREE_LANE_DEBUG", 0) === 1;
145
+ if (laneDebug) {
146
+ try { console.error(`[free-lane] send model=${body?.model} url=${url} stream=${reqBody.stream} tools=${(reqBody.tools || []).length} ua=${opencodeUa()} want=${clientWantsStream}`); } catch {}
147
+ }
148
+ // 非流式调用者实际拿到的是被强制流式的上游 → 聚合回 chat completion JSON
149
+ async function agentJson(resp) {
150
+ if (!(freeLane && !clientWantsStream && resp.ok)) return resp;
151
+ const ct = resp.headers?.get?.("content-type") || "";
152
+ if (!ct.includes("text/event-stream")) return resp;
153
+ try {
154
+ const out = await aggregateChatSse(resp);
155
+ out._t = resp._t;
156
+ return out;
157
+ } catch {
158
+ return resp;
159
+ }
160
+ }
161
+ const t0 = performance.now();
162
+
163
+ // 首发请求(transport 已处理 network/429 等重试)
164
+ let res;
165
+ try {
166
+ res = await transport.request({
167
+ url,
168
+ method: "POST",
169
+ headers: buildHeaders(reqBody, { sessionId }),
170
+ body: reqBody,
171
+ stream: reqBody.stream !== false,
172
+ timeoutMs: connectTimeoutMs,
173
+ retry,
174
+ });
175
+ } catch (e) {
176
+ e._t = e._t || { attempts: [], waitMs: 0, totalMs: Math.round(performance.now() - t0) };
177
+ throw e;
178
+ }
179
+ if (laneDebug) {
180
+ try { console.error(`[free-lane] resp model=${body?.model} status=${res.status} ct=${res.headers?.get?.("content-type") || ""}`); } catch {}
181
+ }
182
+
183
+ // 匿名兜底:仅非 anonFirst 时,public 429 + free 模型才走 hermes 空头重试
184
+ if (!anonFirst && res.status === 429 && isFreeModel(body?.model) && shouldTryAnonFree()) {
185
+ const anonRetries = envInt("MSLXDFF_FREE_ANON_RETRIES", 3);
186
+ const anonDelay = envInt("MSLXDFF_FREE_ANON_DELAY_MS", 1000);
187
+ let anonRes = null;
188
+ for (let i = 0; i < anonRetries; i++) {
189
+ await new Promise((r) => setTimeout(r, anonDelay));
190
+ try {
191
+ anonRes = await transport.request({
192
+ url,
193
+ method: "POST",
194
+ headers: buildHeaders(reqBody, { anonymous: true, sessionId }),
195
+ body: reqBody,
196
+ stream: reqBody.stream !== false,
197
+ timeoutMs: connectTimeoutMs,
198
+ retry,
199
+ });
200
+ } catch (e) {
201
+ continue;
202
+ }
203
+ if (anonRes.status !== 429) {
204
+ // responses 模型需转回 chat 形状(复用 upstream-responses)
205
+ let outAnon = anonRes;
206
+ if (isResp && anonRes.ok) {
207
+ const ctAnon = anonRes.headers.get("content-type") || "";
208
+ const isStreamAnon = (clientWantsStream || freeLane) && ctAnon.includes("text/event-stream");
209
+ if (isStreamAnon) {
210
+ outAnon = reshapeResponsesSse(anonRes, body.model);
211
+ } else {
212
+ try {
213
+ const txt = await anonRes.text();
214
+ const j = JSON.parse(txt);
215
+ if (j && Array.isArray(j.output)) {
216
+ outAnon = toChatResponse(anonRes, j);
217
+ } else {
218
+ outAnon = new Response(txt, { status: anonRes.status, headers: anonRes.headers });
219
+ }
220
+ } catch { outAnon = anonRes; }
221
+ }
222
+ }
223
+ outAnon._t = { ...(outAnon._t || {}), anonTried: true, anonAttempts: i + 1, totalMs: Math.round(performance.now() - t0) };
224
+ consecutiveHits += 1;
225
+ logFreeAnon({ model: body?.model, publicStatus: 429, anonStatus: anonRes.status, anonAttempts: i + 1, hit: true, totalMs: outAnon._t.totalMs });
226
+ if (consecutiveHits >= 2) {
227
+ try { appendFile(freeAnonLogFile(), ` -> 连续额外额度 ${consecutiveHits} 次\n`).catch(() => {}); } catch {}
228
+ }
229
+ return await agentJson(outAnon);
230
+ }
231
+ }
232
+ if (anonRes) {
233
+ anonRes._t = { ...(anonRes._t || {}), anonTried: true, anonAttempts: anonRetries, totalMs: Math.round(performance.now() - t0) };
234
+ consecutiveHits = 0;
235
+ logFreeAnon({ model: body?.model, publicStatus: 429, anonStatus: anonRes.status, anonAttempts: anonRetries, hit: false, totalMs: anonRes._t.totalMs });
236
+ return anonRes;
237
+ }
238
+ }
239
+
240
+ // responses 模型成功态转 chat(复用 upstream-responses)
241
+ if (isResp && res.ok) {
242
+ const ct = res.headers.get("content-type") || "";
243
+ const isStream = (clientWantsStream || freeLane) && ct.includes("text/event-stream");
244
+ if (isStream) {
245
+ const transformed = reshapeResponsesSse(res, body.model);
246
+ transformed._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
247
+ return await agentJson(transformed);
248
+ }
249
+ try {
250
+ const txt = await res.text();
251
+ const j = JSON.parse(txt);
252
+ if (j && Array.isArray(j.output)) {
253
+ const transformed = toChatResponse(res, j);
254
+ transformed._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
255
+ return transformed;
256
+ }
257
+ const fallback = new Response(txt, { status: res.status, headers: res.headers });
258
+ fallback._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
259
+ return fallback;
260
+ } catch {
261
+ res._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
262
+ return res;
263
+ }
264
+ }
265
+ res._t = { ...(res._t || {}), totalMs: res._t?.totalMs ?? Math.round(performance.now() - t0) };
266
+ return await agentJson(res);
267
+ }
268
+
269
+ async function preheat() {
270
+ if (isPreheatDisabled()) return { ok: false, skipped: true, reason: "disabled" };
271
+ const t0 = performance.now();
272
+ try {
273
+ const res = await transport.request({
274
+ url: `${baseUrl}/zen/v1/models`,
275
+ method: "GET",
276
+ headers: buildHeaders({ stream: false }),
277
+ stream: false,
278
+ timeoutMs: 3000,
279
+ });
280
+ try { await res.text().catch(() => {}); } catch {}
281
+ return { ok: res.ok, status: res.status, ms: Math.round(performance.now() - t0) };
282
+ } catch (e) {
283
+ return { ok: false, error: String(e?.message || e), ms: Math.round(performance.now() - t0) };
284
+ }
285
+ }
286
+ async function close() {
287
+ await transport.close();
288
+ }
289
+ const headers = buildHeaders({});
290
+ return { chat, preheat, close, headers, buildHeaders, get dispatcher() { return transport.dispatcher; }, get agent() { return transport.agent; }, [Symbol.asyncDispose]: close };
291
+ }