fix(web,daemon): make max_tokens configurable (closes #29)

BYOK users on custom Anthropic-compatible providers (e.g. Xiaomi MiMo) hit the hardcoded 8192 cap and saw artifacts truncated mid-stream. - AppConfig.maxTokens with Settings input (EN/CN + 8 other locales) - ProxyStreamRequest.maxTokens contract field - anthropic, anthropic-compatible, and openai-compatible providers all forward cfg.maxTokens - /api/proxy/anthropic/stream and /api/proxy/stream payloads honor it, defaulting to 8192 when unset so prior clients are unaffected Original sketch by @mashu in #78 (50a9d14); rebased to the apps/web layout and extended to the proxy paths actually used when baseUrl is set, which is where #29's user actually traffics.
2026-07-07 06:34:10 +08:00 · 2026-05-02 11:51:14 +08:00
parent 3d306cb450
commit 6a3ae5f7dc
17 changed files with 58 additions and 5 deletions
--- a/apps/daemon/src/server.ts
+++ b/apps/daemon/src/server.ts
@@ -2051,7 +2051,7 @@ export async function startServer({ port = 7456, returnServer = false } = {}) {
  app.post('/api/proxy/anthropic/stream', async (req, res) => {
    /** @type {Partial<ProxyStreamRequest>} */
    const proxyBody = req.body || {};
-    const { baseUrl, apiKey, model, systemPrompt, messages } = proxyBody;
+    const { baseUrl, apiKey, model, systemPrompt, messages, maxTokens } = proxyBody;
    if (!baseUrl || !apiKey || !model) {
      return sendApiError(res, 400, 'BAD_REQUEST', 'baseUrl, apiKey, and model are required');
    }
@@ -2067,7 +2067,7 @@ export async function startServer({ port = 7456, returnServer = false } = {}) {

    const payload = {
      model,
-      max_tokens: 8192,
+      max_tokens: typeof maxTokens === 'number' && maxTokens > 0 ? maxTokens : 8192,
      stream: true,
      system: systemPrompt || '',
      messages: Array.isArray(messages)
@@ -2166,7 +2166,7 @@ export async function startServer({ port = 7456, returnServer = false } = {}) {
  app.post('/api/proxy/stream', async (req, res) => {
    /** @type {Partial<ProxyStreamRequest>} */
    const proxyBody = req.body || {};
-    const { baseUrl, apiKey, model, systemPrompt, messages } = proxyBody;
+    const { baseUrl, apiKey, model, systemPrompt, messages, maxTokens } = proxyBody;
    if (!baseUrl || !apiKey || !model) {
      return sendApiError(res, 400, 'BAD_REQUEST', 'baseUrl, apiKey, and model are required');
    }
@@ -2194,7 +2194,7 @@ export async function startServer({ port = 7456, returnServer = false } = {}) {

    const payload = {
      model,
-      max_tokens: 8192,
+      max_tokens: typeof maxTokens === 'number' && maxTokens > 0 ? maxTokens : 8192,
      stream: true,
      ...(isMiMo ? { tool_choice: 'none', tools: [] } : {}),
      messages: [
--- a/apps/web/src/components/SettingsDialog.tsx
+++ b/apps/web/src/components/SettingsDialog.tsx
@@ -445,6 +445,21 @@ export function SettingsDialog({
                  ))}
                </select>
              </label>
+              <label className="field">
+                <span className="field-label">{t('settings.maxTokens')}</span>
+                <input
+                  type="number"
+                  min={1024}
+                  max={200000}
+                  step={1024}
+                  value={cfg.maxTokens ?? 8192}
+                  onChange={(e) => {
+                    const val = parseInt(e.target.value, 10);
+                    setCfg({ ...cfg, maxTokens: Number.isFinite(val) ? val : undefined });
+                  }}
+                />
+                <p className="hint">{t('settings.maxTokensHint')}</p>
+              </label>
              <p className="hint">{t('settings.apiHint')}</p>
            </section>
          )}
--- a/apps/web/src/i18n/locales/de.ts
+++ b/apps/web/src/i18n/locales/de.ts
@@ -73,6 +73,9 @@ export const de: Dict = {
  'settings.hide': 'Ausblenden',
  'settings.model': 'Modell',
  'settings.baseUrl': 'Base URL',
+  'settings.maxTokens': 'Max. Tokens',
+  'settings.maxTokensHint':
+    'Obergrenze für die Antwortlänge. Standard: 8192. Erhöhen (z. B. 32768), wenn ein einzelnes Design-Artefakt mitten im Stream abgeschnitten wird.',
  'settings.apiHint':
    'Aufrufe gehen direkt von diesem Browser an die festgelegte Base URL. Kein Proxy. Der Key verlässt localStorage nie.',
  'settings.skipForNow': 'Vorerst überspringen',
--- a/apps/web/src/i18n/locales/en.ts
+++ b/apps/web/src/i18n/locales/en.ts
@@ -73,6 +73,9 @@ export const en: Dict = {
  'settings.hide': 'Hide',
  'settings.model': 'Model',
  'settings.baseUrl': 'Base URL',
+  'settings.maxTokens': 'Max tokens',
+  'settings.maxTokensHint':
+    'Cap on the response length. Defaults to 8192. Raise it (e.g. 32768) when a single design artifact gets cut off mid-stream.',
  'settings.apiHint':
    'Calls go directly from this browser to the base URL you set. No proxy. The key never leaves localStorage.',
  'settings.skipForNow': 'Skip for now',
--- a/apps/web/src/i18n/locales/es-ES.ts
+++ b/apps/web/src/i18n/locales/es-ES.ts
@@ -73,6 +73,9 @@ export const esES: Dict = {
  'settings.hide': 'Ocultar',
  'settings.model': 'Modelo',
  'settings.baseUrl': 'URL base',
+  'settings.maxTokens': 'Tokens máx.',
+  'settings.maxTokensHint':
+    'Tope para la longitud de la respuesta. Por defecto 8192. Súbelo (p. ej. 32768) cuando un artefacto de diseño se trunca a mitad del stream.',
  'settings.apiHint':
    'Las llamadas van directamente desde este navegador a la URL base que indiques. Sin proxy. La clave nunca sale de localStorage.',
  'settings.skipForNow': 'Omitir por ahora',
--- a/apps/web/src/i18n/locales/fa.ts
+++ b/apps/web/src/i18n/locales/fa.ts
@@ -73,6 +73,9 @@ export const fa: Dict = {
  'settings.hide': 'پنهان',
  'settings.model': 'مدل',
  'settings.baseUrl': 'آدرس پایه',
+  'settings.maxTokens': 'حداکثر توکن',
+  'settings.maxTokensHint':
+    'سقف طول پاسخ. مقدار پیش‌فرض ۸۱۹۲. اگر یک artifact طراحی در میانهٔ استریم قطع می‌شود، آن را افزایش دهید (مثلاً ۳۲۷۶۸).',
  'settings.apiHint':
    'فراخوانی‌ها مستقیماً از این مرورگر به آدرس پایه‌ای که تعیین کرده‌اید ارسال می‌شوند. بدون پراکسی. کلید هرگز localStorage را ترک نمی‌کند.',
  'settings.skipForNow': 'فعلاً رد کنید',
--- a/apps/web/src/i18n/locales/ja.ts
+++ b/apps/web/src/i18n/locales/ja.ts
@@ -73,6 +73,9 @@ export const ja: Dict = {
  'settings.hide': '隠す',
  'settings.model': 'モデル',
  'settings.baseUrl': 'ベース URL',
+  'settings.maxTokens': '最大トークン',
+  'settings.maxTokensHint':
+    '応答長の上限。デフォルトは 8192。デザインの artifact がストリームの途中で切れる場合は、32768 などに引き上げてください。',
  'settings.apiHint':
    'リクエストはこのブラウザから設定したベース URL に直接送信されます。プロキシなし。キーは localStorage から外に出ません。',
  'settings.skipForNow': '今はスキップ',
--- a/apps/web/src/i18n/locales/pt-BR.ts
+++ b/apps/web/src/i18n/locales/pt-BR.ts
@@ -73,6 +73,9 @@ export const ptBR: Dict = {
  'settings.hide': 'Ocultar',
  'settings.model': 'Modelo',
  'settings.baseUrl': 'URL base',
+  'settings.maxTokens': 'Tokens máx.',
+  'settings.maxTokensHint':
+    'Limite para o comprimento da resposta. Padrão: 8192. Aumente (por exemplo, 32768) quando um artefato de design é truncado no meio do stream.',
  'settings.apiHint':
    'As chamadas vão direto deste navegador para a URL base definida. Sem proxy. A chave nunca sai do localStorage.',
  'settings.skipForNow': 'Pular por enquanto',
--- a/apps/web/src/i18n/locales/ru.ts
+++ b/apps/web/src/i18n/locales/ru.ts
@@ -73,6 +73,9 @@ export const ru: Dict = {
  'settings.hide': 'Скрыть',
  'settings.model': 'Модель',
  'settings.baseUrl': 'Базовый URL',
+  'settings.maxTokens': 'Макс. токенов',
+  'settings.maxTokensHint':
+    'Ограничение длины ответа. По умолчанию 8192. Увеличьте (например, до 32768), если один артефакт дизайна обрывается посреди стрима.',
  'settings.apiHint':
    'Вызовы идут напрямую из этого браузера на указанный вами базовый URL. Без прокси. Ключ никогда не покидает localStorage.',
  'settings.skipForNow': 'Пропустить сейчас',
--- a/apps/web/src/i18n/locales/tr.ts
+++ b/apps/web/src/i18n/locales/tr.ts
@@ -73,6 +73,9 @@ export const tr: Dict = {
  'settings.hide': 'Gizle',
  'settings.model': 'Model',
  'settings.baseUrl': 'Temel URL',
+  'settings.maxTokens': 'Maks. token',
+  'settings.maxTokensHint':
+    'Yanıt uzunluğu sınırı. Varsayılan 8192. Bir tasarım artifact akış ortasında kesildiğinde değeri yükseltin (örn. 32768).',
  'settings.apiHint':
    'Çağrılar, bu tarayıcıdan doğrudan belirlediğiniz temel URL’ye yönlendirilir. Vekil sunucu kullanılmaz. Anahtar hiçbir zaman localStorage’dan ayrılmaz.',
  'settings.skipForNow': 'Şimdilik atla',
--- a/apps/web/src/i18n/locales/zh-CN.ts
+++ b/apps/web/src/i18n/locales/zh-CN.ts
@@ -72,6 +72,9 @@ export const zhCN: Dict = {
  'settings.hide': '隐藏',
  'settings.model': '模型',
  'settings.baseUrl': 'Base URL',
+  'settings.maxTokens': '最大 tokens',
+  'settings.maxTokensHint':
+    '响应长度上限，默认 8192。当单次生成的 artifact 被截断时，可以调高（例如 32768）。',
  'settings.apiHint':
    '请求会从当前浏览器直连你设置的 Base URL，无中转代理。Key 只存放在 localStorage。',
  'settings.skipForNow': '暂时跳过',
--- a/apps/web/src/i18n/locales/zh-TW.ts
+++ b/apps/web/src/i18n/locales/zh-TW.ts
@@ -72,6 +72,9 @@ export const zhTW: Dict = {
  'settings.hide': '隱藏',
  'settings.model': '模型',
  'settings.baseUrl': 'Base URL',
+  'settings.maxTokens': '最大 tokens',
+  'settings.maxTokensHint':
+    '回應長度上限，預設 8192。當單次生成的 artifact 被截斷時，可以調高（例如 32768）。',
  'settings.apiHint':
    '請求會從當前瀏覽器直連你設定的 Base URL，無中轉代理。Key 只存放在 localStorage。',
  'settings.skipForNow': '暫時跳過',
--- a/apps/web/src/providers/anthropic-compatible.ts
+++ b/apps/web/src/providers/anthropic-compatible.ts
@@ -26,6 +26,7 @@ export async function streamMessageAnthropicProxy(
        model: cfg.model,
        systemPrompt: system,
        messages: history.map((m) => ({ role: m.role, content: m.content })),
+        maxTokens: cfg.maxTokens,
      }),
      signal,
    });
--- a/apps/web/src/providers/anthropic.ts
+++ b/apps/web/src/providers/anthropic.ts
@@ -57,7 +57,7 @@ export async function streamMessage(
    const stream = client.messages.stream(
      {
        model: cfg.model,
-        max_tokens: 8192,
+        max_tokens: cfg.maxTokens ?? 8192,
        system,
        messages: history.map((m) => ({ role: m.role, content: m.content })),
      },
--- a/apps/web/src/providers/openai-compatible.ts
+++ b/apps/web/src/providers/openai-compatible.ts
@@ -33,6 +33,7 @@ export async function streamMessageOpenAI(
        model: cfg.model,
        systemPrompt: system,
        messages: history.map((m) => ({ role: m.role, content: m.content })),
+        maxTokens: cfg.maxTokens,
      }),
      signal,
    });
--- a/apps/web/src/types.ts
+++ b/apps/web/src/types.ts
@@ -58,6 +58,9 @@ export interface AppConfig {
  // Pre-existing configs without this field fall through to the agent's
  // declared default.
  agentModels?: Record<string, AgentModelChoice>;
+  // Caps the upstream completion length in API mode. Defaults to 8192 when
+  // unset; raise it for providers (e.g. MiMo) that allow longer responses.
+  maxTokens?: number;
 }

 export type AgentEvent = PersistedAgentEvent;
--- a/packages/contracts/src/api/proxy.ts
+++ b/packages/contracts/src/api/proxy.ts
@@ -11,6 +11,9 @@ export interface ProxyStreamRequest {
  model: string;
  systemPrompt?: string;
  messages: ProxyMessage[];
+  // Caps the upstream completion length. Defaults to 8192 when unset so
+  // pre-existing clients keep their old behavior.
+  maxTokens?: number;
 }

 export interface ProxyStreamStartPayload {