dsh-model 0.18.1 → 0.19.1

package/README.en.md

@@ -52,12 +52,14 @@ On a dsh-vps server, use `sudo dsh-model setup`. On Windows, run it in a normal
 |---|---|
 | **Endpoint** | Local and public addresses, each with a Copy button |
 | **Sources** | One switch per provider: on connects (and signs in if needed), off pauses and keeps the sign-in. Each row is tagged Free / Free · limited / Paid and shows plan usage or remaining credits |
-| **Models** | Pick which models of each source appear in dsh (5 suggested per source by default); the rest stay available through the endpoint |
+| **Models** | Pick which models of each source appear in dsh (5 suggested per source by default); the rest stay available through the endpoint; a "Test" button per model |
+| **Auto switch** | When a model is offered by several sources, dsh gets an "Auto switch (multiple sources)" group that moves to another source when one is rate-limited or failing |
+| **Recent failures** | Failed requests in the last 24 hours: which model, why, and what to do (no conversation content kept) |
 | **API keys** | OpenAI-style, one per device. Requests, success rate, latency and speed over 24 hours; copy, rotate, revoke |
 | **Model stats** | Success rate, latency and speed per model |
 | **About** | Current and latest npm version; one-click Update and Uninstall (on a VPS these need root, so the commands are shown instead) |
 
-If the service is not set up on this machine yet, the page says what is missing and offers a "Set up now" button, so you can also start by installing dsh-model from the dsh plugin marketplace and finish from this page.
+If the service is not set up on this machine yet, the page says what is missing and offers a "Set up now" button with live progress. A "Getting started" guide appears on a fresh install; a banner shows when the plugin was updated but dsh not restarted (one-click restart on dsh-vps); a low-quota notice appears in the corner of any dsh page. so you can also start by installing dsh-model from the dsh plugin marketplace and finish from this page.
 
 dsh-model does every sign-in itself. No provider CLI is installed and the server needs no browser:
 
@@ -74,6 +76,10 @@ If a source is signed in but has no usable plan (say, a Kimi account without Kim
 |---|---|---|
 | WorkBuddy / WorkBuddy AI | Free | Credits come with the account |
 | Ollama / LM Studio | Free | Models running on this machine; they appear once installed and started, no sign-in |
+| Gemini (Google AI Studio), Groq | Free · limited | Paste a free API key; daily / per-minute limits apply |
+| OpenRouter | Free · limited | `:free` models are free (about 50 requests a day) |
+| SiliconFlow | Free · limited | Some small models are free after identity verification |
+| DeepSeek official, any OpenAI-compatible service | Paid | Paste an API key (and a URL for other services) |
 | Codex | Paid | ChatGPT Plus / Pro |
 | Kimi | Paid | Kimi Code plan |
 | Grok | Paid | SuperGrok / X Premium+ |
@@ -92,6 +98,7 @@ dsh-model only connects; it provides no quota of its own.
 | `sources` | Sources: switch, cost, sign-in state, usage |
 | `source enable\|disable\|logout <source>` | Turn on (signing in if needed) / off / sign out |
 | `key list\|add\|revoke\|rotate <name>` | Access keys, one per device |
+| `failures` | Failed requests in the last 24 hours, with advice |
 | `stats` | Success rate, latency and speed by key, source and model |
 | `status` / `doctor [--e2e]` | Overview / health checks |
 | `models [sync]` | List models / sync them to dsh |

package/README.md

@@ -52,12 +52,14 @@ dsh-model setup
 |---|---|
 | **访问地址** | 本机地址和外网地址，各带「复制」按钮 |
 | **来源** | 每家一个开关：打开就接入（没登录会弹出登录框），关闭只停用、保留登录。每行标出「免费 / 免费·额度少 / 需付费」，下面显示套餐用量或剩余积分 |
-| **选模型** | 每个来源选哪些模型显示在 dsh 里，默认每家推荐 5 个，其余的经统一地址照样可用 |
+| **选模型** | 每个来源选哪些模型显示在 dsh 里，默认每家推荐 5 个，其余的经统一地址照样可用；每个模型旁边有「测一下」 |
+| **自动切换** | 同一个模型在几个来源都有时，dsh 里多一组「自动切换（多来源）」，一个来源限流或出错就自动换另一个 |
+| **最近失败** | 最近 24 小时失败的请求：哪个模型、为什么、该怎么办（不记录对话内容） |
 | **API key** | OpenAI 格式，每台设备一把。显示 24 小时内的请求数、成功率、延迟和速度，可复制、轮换、吊销 |
 | **模型统计** | 按模型显示成功率、延迟和速度 |
 | **关于** | 当前版本和 npm 上的最新版本；「更新」「卸载」一键完成（VPS 上要 root，这里给出命令） |
 
-还没在这台机器上装好服务时，这一页会说明缺什么，并提供「一键安装」按钮。所以也可以先从 dsh 的插件市场装 dsh-model，再在这一页里点安装。
+还没在这台机器上装好服务时，这一页会说明缺什么，并提供「一键安装」按钮，安装过程实时显示。刚装好时有「开始使用」向导；插件更新后还没重启 dsh 时页面顶部会提示（dsh-vps 上可一键重启）；额度快用完时，dsh 任何页面的右下角都会提醒。所以也可以先从 dsh 的插件市场装 dsh-model，再在这一页里点安装。
 
 登录全由 dsh-model 自己完成，服务器上不装各家的 CLI，也不需要浏览器：
 
@@ -74,6 +76,10 @@ dsh-model setup
 |---|---|---|
 | WorkBuddy / WorkBuddy AI | 免费 | 账号自带积分 |
 | Ollama / LM Studio | 免费 | 本机运行的模型。装好并启动后自动出现，不用登录 |
+| Gemini（Google AI Studio）、Groq | 免费·额度少 | 填免费 API key 即可，每天 / 每分钟有上限 |
+| OpenRouter | 免费·额度少 | 名字带 `:free` 的模型免费（每天约 50 次） |
+| 硅基流动 | 免费·额度少 | 实名认证后部分小模型免费 |
+| DeepSeek 官方、其他 OpenAI 兼容服务 | 需付费 | 填 API key（其他服务还要填地址） |
 | Codex | 需付费 | ChatGPT Plus / Pro |
 | Kimi | 需付费 | Kimi Code 套餐 |
 | Grok | 需付费 | SuperGrok / X Premium+ |
@@ -92,6 +98,7 @@ dsh-model 只负责转接，不提供额度。
 | `sources` | 来源列表：开关、费用、登录状态、用量 |
 | `source enable\|disable\|logout <来源>` | 打开（需要时登录）/ 关闭 / 退出登录 |
 | `key list\|add\|revoke\|rotate <名称>` | 访问 key，每台设备一把 |
+| `failures` | 最近 24 小时失败的请求和建议 |
 | `stats` | 按 key、来源、模型的成功率、延迟和速度 |
 | `status` / `doctor [--e2e]` | 总览 / 逐项自检 |
 | `models [sync]` | 列出模型 / 同步到 dsh |

package/dist/cli.js

@@ -14,7 +14,7 @@ import { supervise } from './service/windows.js';
 import { opencode } from './commands/opencode.js';
 import { workbuddy } from './commands/workbuddy.js';
 import { bridge } from './commands/bridge.js';
-import { enableInteractive, source, sources, stats } from './commands/sources.js';
+import { enableInteractive, failures, source, sources, stats } from './commands/sources.js';
 import { createContext } from './context.js';
 import { isDshModelError } from './errors.js';
 import { L, initLang } from './i18n.js';
@@ -68,6 +68,7 @@ function help() {
                                        打开（未登录就打印链接登录）/ 关闭（保留登录）/ 退出登录
   login <来源> / logout <来源>          同 source enable / source logout
   stats                                按 key / 来源 / 模型的成功率、延迟、速度
+  failures                             最近 24 小时失败的请求，以及该怎么办
   status                               一屏总览
   doctor [--e2e] [--all] [--model M]   逐项自检；--e2e 实测流式与工具调用
   connect-dsh [--dry-run] [--force]    写入 / 刷新 dsh 配置
@@ -99,6 +100,7 @@ Usage: dsh-model <command> [options]
                                        Turn on (prints a login link if needed) / off (keeps sign-in) / sign out
   login <source> / logout <source>     Same as source enable / source logout
   stats                                Success rate, latency and speed by key / source / model
+  failures                             Failed requests in the last 24 hours, and what to do
   status                               Overview
   doctor [--e2e] [--all] [--model M]   Health checks; --e2e tests streaming and tool calls
   connect-dsh [--dry-run] [--force]    Write / refresh dsh config
@@ -167,6 +169,8 @@ export async function main(argv) {
                 return await source(ctx, a1, a2, { acceptRisk: o['accept-risk'] });
             case 'stats':
                 return await stats(ctx);
+            case 'failures':
+                return await failures(ctx);
             case 'status':
                 return await status(ctx);
             case 'doctor':

package/dist/commands/self.js

@@ -129,6 +129,11 @@ export async function adminRun(ctx) {
     const req = await takeAdminRequest(ctx);
     if (!req)
         return 0;
+    if (req.action === 'restart-dsh') {
+        // dsh-vps 的 dsh 由 dsh-gate 管着：重启它（会断开当前会话，页面会自己重连）
+        const r = await run('systemctl', ['restart', 'dsh-gate']);
+        return r.code;
+    }
     const log = selfJobLog(ctx, req.action);
     const out = openSync(log, 'w', 0o600);
     if (ctx.owner)

package/dist/commands/sources.js

@@ -194,3 +194,22 @@ export async function stats(ctx) {
         info(dim(L('还没有请求记录', 'No requests recorded yet')));
     return 0;
 }
+/** dsh-model failures：最近 24 小时失败的请求和建议 */
+export async function failures(ctx) {
+    const list = await control(ctx, 'GET', '/failures');
+    if (isJsonMode()) {
+        printJson(list);
+        return 0;
+    }
+    if (!list.length) {
+        ok(L('最近 24 小时没有失败的请求', 'No failed requests in the last 24 hours'));
+        return 0;
+    }
+    for (const f of list) {
+        info(`${dim(f.at.replace('T', ' ').slice(0, 19))}  ${bold(f.model || '-')}  HTTP ${f.status}`);
+        info(`  ${f.kind === 'rate' || f.kind === 'upstream' ? yellow(f.advice) : red(f.advice)}`);
+        if (f.message)
+            info(dim(`  ${f.message}`));
+    }
+    return 0;
+}

package/dist/daemon/core.js

@@ -9,11 +9,13 @@ import { readFile } from 'node:fs/promises';
 import { join } from 'node:path';
 import { latestVersion, selfJobLog } from '../commands/self.js';
 import { LOCAL_SERVERS, loadLocalCatalog, refreshLocalCatalog } from '../integrations/local.js';
+import { loadKeyCatalog, refreshKeyCatalogs, removeApiKey, setApiKey } from '../integrations/apikeys.js';
 import { adminInstalled, adminRequestPath } from '../service/admin.js';
 import { atomicWrite } from '../util/fs.js';
 import { lang } from '../i18n.js';
 import { which } from '../util/exec.js';
 import { spawnDetached } from '../util/detach.js';
+import { classify, recentFailures } from './failures.js';
 import { PKG_ROOT, pkgVersion } from '../util/pkg.js';
 import { randomBytes } from 'node:crypto';
 import { currentBinary } from '../engine/install.js';
@@ -119,6 +121,9 @@ export class Daemon {
     }
     async refreshCatalogs() {
         let changed = await refreshLocalCatalog(this.ctx.paths.home, this.ctx.owner).catch(() => false);
+        const proxy = (await loadAll(this.ctx).catch(() => null))?.config.proxy;
+        if (await refreshKeyCatalogs(this.ctx, proxy).catch(() => false))
+            changed = true;
         for (const rt of this.runtimes) {
             try {
                 if (await refreshCatalog(this.ctx.paths.home, rt, this.ctx.owner))
@@ -177,6 +182,7 @@ export class Daemon {
         const cats = await loadCatalogs(this.ctx.paths.home);
         const secrets = await loadSecrets(this.ctx);
         const local = await loadLocalCatalog(this.ctx.paths.home);
+        const keyCat = await loadKeyCatalog(this.ctx.paths.home);
         // 订阅来源能服务的模型数：管理接口按凭据列模型（只数没停用的）
         const engineModels = new Map();
         try {
@@ -222,6 +228,19 @@ export class Daemon {
                     models: c?.signedIn ? c.models.length : 0,
                 };
             }
+            if (def.kind === 'apikey') {
+                const has = Boolean(secrets.apikeys?.[def.id]?.key);
+                const models = keyCat[def.id]?.models ?? [];
+                const free = models.filter((m) => m.free).length;
+                return {
+                    ...base,
+                    loggedIn: has,
+                    enabled: has && !disabled.has(def.id),
+                    ...(has ? { account: def.id === 'custom' ? (secrets.apikeys?.[def.id]?.base ?? '') : redactKey(secrets.apikeys[def.id].key) } : {}),
+                    ...(has && free ? { detail: L(`其中 ${free} 个免费`, `${free} of them free`) } : {}),
+                    models: has ? models.length : 0,
+                };
+            }
             if (def.kind === 'local') {
                 const st = local[def.id];
                 const srv = LOCAL_SERVERS.find((x) => x.id === def.id);
@@ -264,6 +283,9 @@ export class Daemon {
             if (!now.loggedIn || !now.models)
                 throw new DshModelError('local_not_running', now.detail ?? L(`${def.label} 没在运行`, `${def.label} is not running`), L(`先装好并启动它：${def.subscribeUrl}`, `Install and start it first: ${def.subscribeUrl}`));
         }
+        if (def.kind === 'apikey' && !state.loggedIn) {
+            throw new DshModelError('needs_key', L(`${def.label} 需要先填 API key`, `${def.label} needs an API key first`), def.subscribeUrl ? L(`在 ${def.subscribeUrl} 创建`, `Create one at ${def.subscribeUrl}`) : undefined);
+        }
         if (def.kind === 'opencode' && !state.loggedIn) {
             throw new DshModelError('needs_key', L('OpenCode Zen 需要 API key：用 opencode/key 提交', 'OpenCode Zen needs an API key: submit it via opencode/key'));
         }
@@ -308,6 +330,9 @@ export class Daemon {
             for (const c of credsFor(await m.credentials(), def))
                 await m.deleteCredential(c.name);
         }
+        else if (def.kind === 'apikey') {
+            await removeApiKey(this.ctx, def.id);
+        }
         else if (def.kind === 'local') {
             throw new DshModelError('no_logout', L(`${def.label} 不需要登录；不想用就关掉开关`, `${def.label} has no sign-in; turn the switch off instead`));
         }
@@ -321,6 +346,17 @@ export class Daemon {
         }
         await this.resync();
     }
+    /** 自带 API key 的来源：实测（拉模型列表）通过才保存，并打开它 */
+    async setSourceKey(id, key, baseUrl) {
+        const def = mustSource(id);
+        if (def.kind !== 'apikey')
+            throw new DshModelError('not_key_source', L(`${def.label} 不用 API key`, `${def.label} does not take an API key`));
+        const all = await loadAll(this.ctx);
+        const r = await setApiKey(this.ctx, def.id, key.trim(), baseUrl, all.config.proxy);
+        await this.setSourceDisabled(def.id, false);
+        await this.resync();
+        return r;
+    }
     async setOpencodeKey(key, opts = {}) {
         validateKeyShape(key);
         const all = await loadAll(this.ctx);
@@ -438,7 +474,8 @@ export class Daemon {
     async usage(refresh = false) {
         if (refresh)
             await this.refreshUsage();
-        return [...this.usageCache.values()];
+        // 带上来源名：dsh 里的额度提醒直接显示
+        return [...this.usageCache.values()].map((u) => ({ ...u, ...(findSource(u.source) ? { label: findSource(u.source).label } : {}) }));
     }
     // —— 登录会话 ——
     session(id) {
@@ -726,6 +763,50 @@ export class Daemon {
         log_(`daemon: started ${action}${pid ? ` (pid ${pid})` : ''}`);
         return { started: true, log };
     }
+    /** 「测一下」：经引擎发一句很短的话，看能不能用、多久回复（会用掉一点点额度） */
+    async testModel(model) {
+        if (!model || model.length > 200)
+            throw new DshModelError('bad_model', L('model 不合法', 'invalid model'));
+        const all = await loadAll(this.ctx);
+        const started = Date.now();
+        let res;
+        try {
+            res = await fetch(`http://127.0.0.1:${all.config.port}/v1/chat/completions`, {
+                method: 'POST',
+                headers: { Authorization: `Bearer ${dshKey(all.keys)}`, 'Content-Type': 'application/json' },
+                body: JSON.stringify({ model, max_tokens: 256, stream: false, messages: [{ role: 'user', content: L('只回复一个字：好', 'Reply with one word: ok') }] }),
+                signal: AbortSignal.timeout(90_000),
+            });
+        }
+        catch (error) {
+            return { ok: false, ms: Date.now() - started, advice: L('请求没有完成（超时或网络问题）', 'The request did not finish (timeout or network)'), message: String(error.message ?? error) };
+        }
+        const ms = Date.now() - started;
+        const text = await res.text();
+        if (res.ok) {
+            let reply = '';
+            try {
+                const m = JSON.parse(text).choices?.[0]?.message;
+                reply = (m?.content || (m?.reasoning_content ? L('（只返回了思考过程）', '(only reasoning returned)') : '')).trim().slice(0, 80);
+            }
+            catch {
+                // 不是 JSON 也算通了
+            }
+            return { ok: true, ms, reply };
+        }
+        let message = text.slice(0, 300);
+        let code;
+        try {
+            const e = JSON.parse(text).error ?? {};
+            code = e.code;
+            message = String(e.last_upstream_error || e.message || message).slice(0, 300);
+        }
+        catch {
+            // 原样
+        }
+        const c = classify({ status: res.status, ...(code ? { code } : {}), message });
+        return { ok: false, ms, status: res.status, advice: c.advice, message };
+    }
     /** 这个来源的全部模型，以及哪些显示在 dsh 里 */
     async sourceModels(id) {
         const def = mustSource(id);
@@ -837,6 +918,9 @@ export class Daemon {
                 if (method === 'POST')
                     return ok(await this.setSourceModels(seg[1], b.models === null ? null : b.models));
             }
+            if (method === 'POST' && seg[0] === 'apikey' && seg[1]) {
+                return ok(await this.setSourceKey(seg[1], String(b.key ?? ''), typeof b.baseUrl === 'string' ? b.baseUrl : undefined));
+            }
             if (method === 'POST' && path === '/opencode/key') {
                 return ok(await this.setOpencodeKey(String(b.key ?? ''), { skipVerify: b.skipVerify === true }));
             }
@@ -868,6 +952,10 @@ export class Daemon {
                 return ok(await this.endpoints());
             if (method === 'GET' && path === '/self')
                 return ok(await this.selfInfo());
+            if (method === 'GET' && path === '/failures')
+                return ok(await recentFailures(this.ctx.paths.logs));
+            if (method === 'POST' && path === '/models/test')
+                return ok(await this.testModel(String(b.model ?? '')));
             if (method === 'POST' && path === '/self/update')
                 return ok(await this.startSelfJob('update'));
             if (method === 'POST' && path === '/self/uninstall')

package/dist/daemon/failures.js

@@ -0,0 +1,80 @@
+// daemon/failures.ts — 最近失败的请求（设计 §14.12）：从引擎写的 error-*.log 里取出时间、模型、状态码和上游的报错
+//
+// 引擎每个失败请求写一个 auth/logs/error-<路径>-<时间>-<id>.log，里面有完整的请求和响应。
+// 这里只取：时间、模型名、状态码、错误码和错误信息——请求内容（对话）一律不读出来、不往页面送。
+// 再按常见原因归类，给出"该怎么办"。
+import { readdir, readFile, stat } from 'node:fs/promises';
+import { join } from 'node:path';
+import { L } from '../i18n.js';
+const section = (text, name) => {
+    const i = text.indexOf(`=== ${name} ===`);
+    if (i < 0)
+        return '';
+    const rest = text.slice(i + name.length + 8);
+    const j = rest.indexOf('\n=== ');
+    return j < 0 ? rest : rest.slice(0, j);
+};
+/** 解析一个 error-*.log；看不懂返回 null。导出给测试 */
+export function parseErrorLog(text) {
+    const info = section(text, 'REQUEST INFO');
+    const at = /^Timestamp:\s*(\S+)/m.exec(info)?.[1];
+    // 只用正则取 model 字段，不解析、不保留请求体
+    const model = /"model"\s*:\s*"([^"]{1,200})"/.exec(section(text, 'REQUEST BODY'))?.[1] ?? '';
+    const resp = section(text, 'RESPONSE');
+    const status = Number(/^Status:\s*(\d{3})/m.exec(resp)?.[1]);
+    if (!at || !status)
+        return null;
+    // 响应体是最后一个 JSON：{"error":{"code","message","last_upstream_error"}}
+    let code;
+    let message = '';
+    const body = resp.slice(resp.lastIndexOf('\n{') + 1).trim();
+    try {
+        const e = JSON.parse(body).error ?? {};
+        code = e.code ?? e.type;
+        message = String(e.last_upstream_error || e.message || '');
+    }
+    catch {
+        message = body.slice(0, 300);
+    }
+    return { at, model, source: model.includes('/') ? model.split('/')[0] : '', status, ...(code ? { code } : {}), message: message.replace(/\s+/g, ' ').slice(0, 300) };
+}
+/** 按常见原因归类，给出该怎么办 */
+export function classify(f) {
+    const m = `${f.code ?? ''} ${f.message}`.toLowerCase();
+    if (f.status === 429 || /cooldown|rate|频繁|too many/.test(m)) {
+        return { kind: 'rate', advice: L('被限流：请求太频繁。等一会儿再试，或换一个来源的同款模型', 'Rate limited: too many requests. Wait a bit, or use the same model from another source') };
+    }
+    if (f.status === 402 || /payment|insufficient|quota|balance|credit|余额|额度|subscription|access_terminated/.test(m)) {
+        return { kind: 'quota', advice: L('额度用完或没有可用套餐：看一下这个来源的用量，或换一个来源', 'Out of quota or no usable plan: check this source\'s usage, or switch source') };
+    }
+    if (f.status === 401 || /unauthor|expired|invalid.*(key|token)|登录/.test(m)) {
+        return { kind: 'auth', advice: L('登录或 key 失效：在来源里重新登录，或换一把 key', 'Sign-in or key no longer valid: sign in again, or change the key') };
+    }
+    if (f.status === 404 || /model_not_found|unknown provider|not found|not supported/.test(m)) {
+        return { kind: 'model', advice: L('模型不存在或已下线：打开来源列表刷新一下，或换个模型', 'Model missing or retired: refresh the source list, or pick another model') };
+    }
+    if (f.status >= 500 || /timeout|timed out|econn|network|unavailable/.test(m)) {
+        return { kind: 'upstream', advice: L('上游出错或网络不通：稍后再试；如果一直这样，检查出站代理', 'Upstream error or network problem: try again later; if it persists, check the outbound proxy') };
+    }
+    return { kind: 'other', advice: L('看下面的原始报错', 'See the raw error below') };
+}
+/** 最近 24 小时的失败，新的在前，最多 limit 条 */
+export async function recentFailures(logsDir, limit = 20, sinceMs = 24 * 60 * 60 * 1000) {
+    let names = [];
+    try {
+        names = (await readdir(logsDir)).filter((n) => /^error-.*\.log$/.test(n));
+    }
+    catch {
+        return [];
+    }
+    const since = Date.now() - sinceMs;
+    const withTime = await Promise.all(names.map(async (n) => ({ n, t: (await stat(join(logsDir, n)).catch(() => null))?.mtimeMs ?? 0 })));
+    const recent = withTime.filter((x) => x.t >= since).sort((a, b) => b.t - a.t).slice(0, limit);
+    const out = [];
+    for (const { n } of recent) {
+        const p = parseErrorLog(await readFile(join(logsDir, n), 'utf8').catch(() => ''));
+        if (p)
+            out.push({ ...p, ...classify(p) });
+    }
+    return out;
+}

package/dist/daemon/usage.js

@@ -12,7 +12,7 @@ import { readFile } from 'node:fs/promises';
 import { join } from 'node:path';
 const TIMEOUT_MS = 15_000;
 /** 这几家暂不支持（见设计 §14.6） */
-export const USAGE_UNSUPPORTED = new Set(['antigravity', 'devin', 'meta', 'opencode', 'ollama', 'lmstudio']);
+export const USAGE_UNSUPPORTED = new Set(['antigravity', 'devin', 'meta', 'opencode', 'ollama', 'lmstudio', 'gemini', 'groq', 'openrouter', 'siliconflow', 'deepseek', 'custom']);
 async function getJson(url, headers) {
     const res = await fetch(url, { headers: { Accept: 'application/json', ...headers }, signal: AbortSignal.timeout(TIMEOUT_MS) });
     const text = await res.text();

package/dist/dsh/cli.js

@@ -5,7 +5,7 @@ import { join, win32 } from 'node:path';
 import { VPS_ROOT } from '../context.js';
 import { DshModelError } from '../errors.js';
 import { L } from '../i18n.js';
-import { run, runInherit, which } from '../util/exec.js';
+import { run, runTee, which } from '../util/exec.js';
 import { exists } from '../util/fs.js';
 const MAC_APPS = ['/Applications/DeepSeek Harness.app', join(homedir(), 'Applications/DeepSeek Harness.app')];
 /** Windows 上 DeepSeek Harness 桌面版可能装在的目录 */
@@ -109,10 +109,14 @@ async function profileStoreDir(ctx, profile) {
 }
 /** dsh plugin --profile <p> <pnpm-args...>，输出直接给用户看（pnpm 进度） */
 export async function dshPlugin(ctx, profile, pnpmArgs, proxy) {
+    return (await dshPluginTee(ctx, profile, pnpmArgs, proxy)).code;
+}
+/** 同 dshPlugin，另外返回输出（安装失败时要认出原因） */
+export async function dshPluginTee(ctx, profile, pnpmArgs, proxy) {
     const cli = await locateDshCli(ctx);
     const store = await profileStoreDir(ctx, profile);
     const w = wrap(ctx, cli, ['plugin', '--profile', profile, ...pnpmArgs, ...(store ? [`--config.store-dir=${store}`] : [])], proxy);
-    return runInherit(w.cmd, w.args, { env: { ...dshEnv(ctx, proxy), ...cli.env } });
+    return runTee(w.cmd, w.args, { env: { ...dshEnv(ctx, proxy), ...cli.env } });
 }
 /** profile 的 package.json 里有没有这个依赖（有就是用户或 dsh 已经装了） */
 export async function profileHasDependency(ctx, profile, name) {

package/dist/engine/compat.js

@@ -7,6 +7,7 @@ import { bridgeConfigPath, loadCatalogs } from '../bridge/runtime.js';
 import { loadSecrets } from '../secrets.js';
 import { readJson } from '../util/fs.js';
 import { LOCAL_SERVERS, loadLocalCatalog } from '../integrations/local.js';
+import { keyProvider, loadKeyCatalog } from '../integrations/apikeys.js';
 export const OPENCODE_BASE = 'https://opencode.ai/zen/v1';
 export const OPENCODE_PREFIX = 'opencode';
 export function opencodeModelsPath(ctx) {
@@ -68,6 +69,21 @@ export async function loadCompatUpstreams(ctx) {
             });
         }
     }
+    // 自带 API key 的来源：有 key、有模型、没被关掉就接进来（走出站代理，不 direct）
+    const keyCat = await loadKeyCatalog(ctx.paths.home);
+    for (const [id, s] of Object.entries(secrets.apikeys ?? {})) {
+        const p = keyProvider(id);
+        const models = keyCat[id]?.models ?? [];
+        if (!p || !models.length || disabled.has(id))
+            continue;
+        out.push({
+            name: id,
+            label: p.label,
+            baseUrl: s.base ?? p.base,
+            apiKey: s.key,
+            models: models.map((m) => ({ name: m.id, alias: `${id}/${m.id}`, displayName: `${p.label} · ${m.id}`, dshName: m.free ? `${m.id} · Free` : m.id, ...(m.free ? { rate: 0 } : {}), group: id, groupLabel: p.label })),
+        });
+    }
     // 本机模型服务（Ollama、LM Studio）：在运行、有模型、没被关掉就接进来
     const local = await loadLocalCatalog(ctx.paths.home);
     for (const s of LOCAL_SERVERS) {
@@ -83,7 +99,41 @@ export async function loadCompatUpstreams(ctx) {
             models: st.models.map((id) => ({ name: id, alias: `${s.id}/${id}`, displayName: `${s.label} · ${id}`, dshName: `${id} · 本地免费`, rate: 0, group: s.id, groupLabel: `${s.label}（本地）` })),
         });
     }
-    return out;
+    return addAutoModels(out);
+}
+export const AUTO_PREFIX = 'auto';
+/**
+ * 同名模型自动切换（设计 §14.13）：同一个模型名出现在两个以上来源时，在每个来源里再登记一个 auto/<名字>。
+ * 引擎把同一个 alias 的几个上游当成候选，一个限流或出错就换下一个（v8.0.13 实测：一个一直 429，请求照样成功）。
+ * 订阅来源（引擎原生 OAuth）不参与：计费方式不同，混在一起会让人搞不清用的是哪家。
+ */
+export function addAutoModels(upstreams) {
+    const where = new Map();
+    for (const u of upstreams) {
+        for (const name of new Set(u.models.map((m) => m.name)))
+            where.set(name, [...(where.get(name) ?? []), u]);
+    }
+    for (const [name, ups] of where) {
+        if (ups.length < 2)
+            continue;
+        const first = ups[0].models.find((m) => m.name === name);
+        const pretty = (first.dshName ?? name).split(' · ')[0];
+        const rates = ups.map((u) => u.models.find((m) => m.name === name)?.rate).filter((r) => r !== undefined);
+        const sources = ups.map((u) => u.label).join(' / ');
+        for (const u of ups) {
+            const own = u.models.find((m) => m.name === name);
+            u.models.push({
+                ...own,
+                alias: `${AUTO_PREFIX}/${name}`,
+                displayName: `${pretty}（${sources}）`,
+                dshName: `${pretty} · ${ups.length} 个来源`,
+                ...(rates.length ? { rate: Math.min(...rates) } : {}),
+                group: AUTO_PREFIX,
+                groupLabel: '自动切换（多来源）',
+            });
+        }
+    }
+    return upstreams;
 }
 /** 渲染成引擎 v8 的 api-keys.openai-compatibility 段 */
 export function renderCompat(upstreams) {

package/dist/integrations/apikeys.js

@@ -0,0 +1,124 @@
+// integrations/apikeys.ts — 自带 API key 的来源（设计 §14.11）：Gemini、Groq、OpenRouter、硅基流动、DeepSeek 官方，以及任意 OpenAI 兼容服务
+//
+// 和 OpenCode 一样：key 存 dsh-model 自己的 secrets.json（0600），由引擎作为 openai-compatibility 上游持有；
+// 模型目录取自各家的 GET /models（带 key，能拉到就说明 key 可用，不花 token）。请求走 curl：它认代理，key 经 stdin 传、不进进程参数。
+import { join } from 'node:path';
+import { DshModelError } from '../errors.js';
+import { L } from '../i18n.js';
+import { loadSecrets, saveSecrets } from '../secrets.js';
+import { run } from '../util/exec.js';
+import { readJson, writeJson } from '../util/fs.js';
+export const KEY_PROVIDERS = [
+    { id: 'gemini', label: 'Gemini（Google AI Studio）', base: 'https://generativelanguage.googleapis.com/v1beta/openai', keyUrl: 'https://aistudio.google.com/apikey' },
+    { id: 'groq', label: 'Groq', base: 'https://api.groq.com/openai/v1', keyUrl: 'https://console.groq.com/keys' },
+    { id: 'openrouter', label: 'OpenRouter', base: 'https://openrouter.ai/api/v1', keyUrl: 'https://openrouter.ai/keys' },
+    { id: 'siliconflow', label: '硅基流动 SiliconFlow', base: 'https://api.siliconflow.cn/v1', keyUrl: 'https://cloud.siliconflow.cn/account/ak', listQuery: 'type=text&sub_type=chat' },
+    { id: 'deepseek', label: 'DeepSeek 官方', base: 'https://api.deepseek.com/v1', keyUrl: 'https://platform.deepseek.com/api_keys' },
+    { id: 'custom', label: '其他 OpenAI 兼容服务', base: '', keyUrl: '' },
+];
+export const keyProvider = (id) => KEY_PROVIDERS.find((p) => p.id === id);
+export const keyCatalogPath = (home) => join(home, 'apikey-models.json');
+export async function loadKeyCatalog(home) {
+    return (await readJson(keyCatalogPath(home))) ?? {};
+}
+export function normalizeBase(url) {
+    const u = url.trim().replace(/\/+$/, '').replace(/\/chat\/completions$/, '').replace(/\/models$/, '');
+    if (!/^https?:\/\/[^/\s]+/i.test(u))
+        throw new DshModelError('invalid_base_url', L(`地址不合法：${url}（例：https://api.example.com/v1）`, `Invalid URL: ${url} (e.g. https://api.example.com/v1)`));
+    return u;
+}
+/** 解析 GET /models 的返回；导出给测试 */
+export function parseModelList(providerId, body) {
+    const data = (body?.data ?? []);
+    const out = [];
+    for (const m of data) {
+        // Gemini 的 OpenAI 兼容接口返回 "models/gemini-2.5-flash"，调用时用不带前缀的名字
+        const id = String(m.id ?? '').replace(/^models\//, '');
+        if (!id)
+            continue;
+        const pricing = m.pricing;
+        const free = id.endsWith(':free') || (pricing ? Number(pricing.prompt) === 0 && Number(pricing.completion) === 0 : false);
+        out.push({ id, ...(free ? { free: true } : {}) });
+    }
+    // OpenRouter 有几百个模型：付费的照样能用，但先把免费的排前面，挑默认时就先挑它们
+    return out.sort((a, b) => Number(Boolean(b.free)) - Number(Boolean(a.free)) || a.id.localeCompare(b.id));
+}
+/** 用这把 key 拉模型列表：拉得到就是可用的 key（不花 token） */
+export async function fetchKeyModels(provider, key, base, proxy) {
+    const url = `${base}/models${provider.listQuery ? `?${provider.listQuery}` : ''}`;
+    const r = await run('curl', ['-s', '-m', '25', '-w', '\n%{http_code}', '-K', '-', ...(proxy ? ['-x', proxy] : []), url], {
+        input: `header = "Authorization: Bearer ${key.replace(/"/g, '')}"\n`,
+        timeoutMs: 40_000,
+    });
+    const lines = r.stdout.trimEnd().split('\n');
+    const status = Number(lines.pop()) || null;
+    const text = lines.join('\n');
+    if (status === 401 || status === 403)
+        throw new DshModelError('key_rejected', L(`${provider.label} 不认这把 key（HTTP ${status}）`, `${provider.label} rejected the key (HTTP ${status})`), provider.keyUrl ? L(`到 ${provider.keyUrl} 重新创建一把`, `Create a new one at ${provider.keyUrl}`) : undefined);
+    if (status !== 200) {
+        throw new DshModelError('key_check_failed', status ? L(`${provider.label} 返回 HTTP ${status}：${text.slice(0, 160)}`, `${provider.label} returned HTTP ${status}: ${text.slice(0, 160)}`) : L(`连不上 ${base}${proxy ? '' : '（需要代理？在设置里或 dsh-model setup --proxy <地址>）'}`, `Cannot reach ${base}${proxy ? '' : ' (proxy needed? dsh-model setup --proxy <url>)'}`));
+    }
+    let models = [];
+    try {
+        models = parseModelList(provider.id, JSON.parse(text));
+    }
+    catch {
+        throw new DshModelError('key_check_failed', L(`${provider.label} 的模型列表看不懂`, `Could not read the ${provider.label} model list`));
+    }
+    if (!models.length)
+        throw new DshModelError('key_no_models', L(`${provider.label} 没有返回任何模型`, `${provider.label} returned no models`));
+    return models;
+}
+/** 实测并保存 key、写模型目录 */
+export async function setApiKey(ctx, id, key, baseUrl, proxy) {
+    const provider = keyProvider(id);
+    if (!provider)
+        throw new DshModelError('unknown_source', L(`没有这个来源：${id}`, `No such source: ${id}`));
+    if (!key || /\s/.test(key) || key.length < 8)
+        throw new DshModelError('invalid_key', L('这看起来不是一把 API key', 'That does not look like an API key'));
+    const base = normalizeBase(provider.id === 'custom' ? (baseUrl ?? '') : provider.base);
+    const models = await fetchKeyModels(provider, key, base, proxy);
+    const secrets = await loadSecrets(ctx);
+    secrets.apikeys = { ...secrets.apikeys, [id]: { key, ...(provider.id === 'custom' ? { base } : {}), verifiedAt: new Date().toISOString() } };
+    await saveSecrets(ctx, secrets);
+    const cat = await loadKeyCatalog(ctx.paths.home);
+    cat[id] = { updatedAt: new Date().toISOString(), models };
+    await writeJson(keyCatalogPath(ctx.paths.home), cat, { owner: ctx.owner });
+    return { models: models.length, free: models.filter((m) => m.free).length };
+}
+export async function removeApiKey(ctx, id) {
+    const secrets = await loadSecrets(ctx);
+    if (secrets.apikeys?.[id]) {
+        delete secrets.apikeys[id];
+        await saveSecrets(ctx, secrets);
+    }
+    const cat = await loadKeyCatalog(ctx.paths.home);
+    if (cat[id]) {
+        delete cat[id];
+        await writeJson(keyCatalogPath(ctx.paths.home), cat, { owner: ctx.owner });
+    }
+}
+/** 定期刷新已配置来源的模型目录（模型上新 / 下线）；失败的保留上次的。返回是否有变化 */
+export async function refreshKeyCatalogs(ctx, proxy) {
+    const secrets = await loadSecrets(ctx);
+    const cat = await loadKeyCatalog(ctx.paths.home);
+    let changed = false;
+    for (const [id, s] of Object.entries(secrets.apikeys ?? {})) {
+        const provider = keyProvider(id);
+        if (!provider)
+            continue;
+        try {
+            const models = await fetchKeyModels(provider, s.key, s.base ?? provider.base, proxy);
+            if (JSON.stringify(models) !== JSON.stringify(cat[id]?.models)) {
+                cat[id] = { updatedAt: new Date().toISOString(), models };
+                changed = true;
+            }
+        }
+        catch {
+            // 网络 / 限流：留着上次的目录
+        }
+    }
+    if (changed)
+        await writeJson(keyCatalogPath(ctx.paths.home), cat, { owner: ctx.owner });
+    return changed;
+}