fix: make max_tokens configurable (closes #29)

CONTRIBUTING.md

@@ -193,6 +193,28 @@ Bar for merging:
 
 ---
 
+## Updating model `max_tokens` metadata
+
+API-mode chat sends `max_tokens` to the upstream provider on every request. The web client picks that number from a three-tier lookup in [`apps/web/src/state/maxTokens.ts`](apps/web/src/state/maxTokens.ts):
+
+1. The user's explicit override in Settings, if set.
+2. Otherwise, the per-model default in [`apps/web/src/state/litellm-models.json`](apps/web/src/state/litellm-models.json) — a vendored slice of [BerriAI/litellm][litellm]'s `model_prices_and_context_window.json` (MIT). It covers ~2k chat models across Anthropic, OpenAI, DeepSeek, Groq, Together, Mistral, Gemini, Bedrock, Vertex, OpenRouter, and friends.
+3. Otherwise, `FALLBACK_MAX_TOKENS = 8192`.
+
+To pick up a newly-launched model, regenerate the vendored JSON:
+
+```bash
+node --experimental-strip-types scripts/sync-litellm-models.ts
+```
+
+The script fetches LiteLLM's catalog, filters to `mode: 'chat'` entries, projects each to its `max_output_tokens` (or `max_tokens` fallback), and writes a sorted snapshot. Commit the regenerated `litellm-models.json` alongside whatever PR triggered the refresh.
+
+The OVERRIDES table in `maxTokens.ts` is for the rare case where LiteLLM is missing or wrong for a model id we actually use — for example, `mimo-v2.5-pro` (LiteLLM only ships MiMo via the `openrouter/xiaomi/...` and `novita/xiaomimimo/...` aliases, neither of which matches the canonical id Xiaomi's direct API uses). Keep it small; everything that LiteLLM gets right belongs upstream.
+
+[litellm]: https://github.com/BerriAI/litellm
+
+---
+
 ## Localization maintenance
 
 German uses formal `Sie` because OD speaks to a mixed audience of solo creators, agencies, and engineering teams; until project feedback shows that an informal `du` voice fits better, formal German is the least surprising default. Locale PRs should translate UI chrome, core docs, and display-only gallery metadata in `apps/web/src/i18n/content.ts`, but should not translate `skills/`, `design-systems/`, or prompt bodies that agents execute. Those source prompts are maintained as workflow inputs, and keeping one source language avoids multiplying prompt QA across locales. When adding or renaming a skill, design system, or prompt template, update the German display metadata and run `pnpm --filter @open-design/web test`; `content.test.ts` fails if German display coverage drifts. Daemon errors, export filenames, and agent-generated artifact text are known limitations unless a PR explicitly scopes them.

CONTRIBUTING.zh-CN.md

@@ -192,6 +192,28 @@ design-systems/your-brand/
 
 ---
 
+## 更新模型 `max_tokens` 元数据
+
+API 模式下每次请求都会带 `max_tokens` 给上游。Web 端通过 [`apps/web/src/state/maxTokens.ts`](apps/web/src/state/maxTokens.ts) 的三层 lookup 决定这个数字：
+
+1. 用户在 Settings 里手填的覆盖值（如果有）。
+2. 否则用 [`apps/web/src/state/litellm-models.json`](apps/web/src/state/litellm-models.json) 里的 per-model 默认 —— 这是从 [BerriAI/litellm][litellm] 的 `model_prices_and_context_window.json`（MIT）摘的一份切片，覆盖约 2000 个 chat 模型，包括 Anthropic、OpenAI、DeepSeek、Groq、Together、Mistral、Gemini、Bedrock、Vertex、OpenRouter 等。
+3. 都 miss 就走 `FALLBACK_MAX_TOKENS = 8192`。
+
+新模型上线想吃到默认值，重新生成 vendored JSON：
+
+```bash
+node --experimental-strip-types scripts/sync-litellm-models.ts
+```
+
+脚本会拉 LiteLLM 的最新 catalog、过滤 `mode: 'chat'`、把每条投影到 `max_output_tokens`（缺失时 fallback 到 `max_tokens`），写成排好序的快照。把重新生成的 `litellm-models.json` 跟着触发它的 PR 一起提。
+
+`maxTokens.ts` 里的 OVERRIDES 表只用于 LiteLLM 没收 / 收错的 model id —— 比如 `mimo-v2.5-pro`（LiteLLM 只收了 `openrouter/xiaomi/...` 和 `novita/xiaomimimo/...` 两个 alias，model id 跟小米直接 API 用的不一样）。表要保持小：凡是 LiteLLM 已经对的，**不要**抄进来。
+
+[litellm]: https://github.com/BerriAI/litellm
+
+---
+
 ## 代码风格
 
 格式我们不抠（保存时跑 Prettier 就行），但有两条不能让 —— 因为它们出现在提示词栈和用户可见的 API 里：

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: [

apps/web/src/components/SettingsDialog.tsx

@@ -10,6 +10,11 @@ import {
   renderModelOptions,
 } from './modelOptions';
 import { KNOWN_PROVIDERS } from '../state/config';
+import {
+  MAX_MAX_TOKENS,
+  MIN_MAX_TOKENS,
+  modelMaxTokensDefault,
+} from '../state/maxTokens';
 import type { AgentInfo, AppConfig, AppVersionInfo, ExecMode } from '../types';
 import { MEDIA_PROVIDERS } from '../media/models';
 import type { MediaProvider } from '../media/models';
@@ -445,6 +450,27 @@ export function SettingsDialog({
                   ))}
                 </select>
               </label>
+              <label className="field">
+                <span className="field-label">{t('settings.maxTokens')}</span>
+                <input
+                  type="number"
+                  min={MIN_MAX_TOKENS}
+                  max={MAX_MAX_TOKENS}
+                  step={MIN_MAX_TOKENS}
+                  placeholder={String(modelMaxTokensDefault(cfg.model))}
+                  value={cfg.maxTokens ?? ''}
+                  onChange={(e) => {
+                    const raw = e.target.value.trim();
+                    if (raw === '') {
+                      setCfg({ ...cfg, maxTokens: undefined });
+                      return;
+                    }
+                    const val = parseInt(raw, 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>
           )}

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 (optional)',
+  'settings.maxTokensHint':
+    'Obergrenze für die Antwortlänge. Jedes Modell hat einen abgestimmten Standardwert (im Platzhalter sichtbar); leer lassen, um ihn zu verwenden, oder eine Zahl eingeben, um ihn zu überschreiben.',
   'settings.apiHint':
     'Aufrufe gehen direkt von diesem Browser an die festgelegte Base URL. Kein Proxy. Der Key verlässt localStorage nie.',
   'settings.skipForNow': 'Vorerst überspringen',

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 (optional)',
+  'settings.maxTokensHint':
+    'Cap on the response length. Each model has a tuned default (shown as a placeholder); leave blank to use it, or enter a number to override.',
   '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',

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. (opcional)',
+  'settings.maxTokensHint':
+    'Tope para la longitud de la respuesta. Cada modelo tiene un valor por defecto ajustado (visible en el placeholder); déjalo vacío para usarlo o introduce un número para anularlo.',
   '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',

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 (اختیاری)',
+  'settings.maxTokensHint':
+    'سقف طول پاسخ. هر مدل مقدار پیش‌فرض تنظیم‌شدهٔ خود را دارد (در placeholder نمایش داده می‌شود)؛ برای استفاده از آن خالی بگذارید، یا برای جایگزینی، عددی وارد کنید.',
   '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',

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（任意）',
+  'settings.maxTokensHint':
+    '応答長の上限。各モデルにチューニング済みのデフォルト値があります（プレースホルダーに表示）。空のままにすればそれを使用し、数値を入力すれば上書きされます。',
   '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',

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. (opcional)',
+  'settings.maxTokensHint':
+    'Limite para o comprimento da resposta. Cada modelo tem um valor padrão ajustado (visível no placeholder); deixe em branco para usá-lo ou insira um número para substituí-lo.',
   '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',

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 (опционально)',
+  'settings.maxTokensHint':
+    'Ограничение длины ответа. У каждой модели свой настроенный дефолт (виден в плейсхолдере); оставьте поле пустым, чтобы использовать его, или введите число, чтобы переопределить.',
   'settings.apiHint':
     'Вызовы идут напрямую из этого браузера на указанный вами базовый URL. Без прокси. Ключ никогда не покидает localStorage.',
   'settings.skipForNow': 'Пропустить сейчас',

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 (isteğe bağlı)',
+  'settings.maxTokensHint':
+    'Yanıt uzunluğu sınırı. Her modelin ayarlanmış bir varsayılanı vardır (yer tutucuda görünür); kullanmak için boş bırakın, üzerine yazmak için bir sayı girin.',
   'settings.apiHint':
     'Çağrılar, bu tarayıcıdan doğrudan belirlediğiniz temel URL’ye yönlendirilir. Vekil sunucu kullanılmaz. Anahtar hiçbir zaman localStorage’dan ayrılmaz.',
   'settings.skipForNow': 'Şimdilik atla',

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（可选）',
+  'settings.maxTokensHint':
+    '响应长度上限。每个 model 有调优过的默认值（在 placeholder 里显示），留空即使用，输入数字则覆盖。',
   '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',

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（可選）',
+  'settings.maxTokensHint':
+    '回應長度上限。每個 model 有調過的預設值（在 placeholder 顯示），留空即使用，輸入數字則覆蓋。',
   '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',

apps/web/src/providers/anthropic-compatible.ts

@@ -1,3 +1,4 @@
+import { effectiveMaxTokens } from '../state/maxTokens';
 import type { AppConfig, ChatMessage } from '../types';
 import type { StreamHandlers } from './anthropic';
 import { parseSseFrame } from './sse';
@@ -26,6 +27,7 @@ export async function streamMessageAnthropicProxy(
         model: cfg.model,
         systemPrompt: system,
         messages: history.map((m) => ({ role: m.role, content: m.content })),
+        maxTokens: effectiveMaxTokens(cfg),
       }),
       signal,
     });

apps/web/src/providers/anthropic.ts

@@ -8,6 +8,7 @@
  * your own backend.
  */
 import Anthropic from '@anthropic-ai/sdk';
+import { effectiveMaxTokens } from '../state/maxTokens';
 import type { AppConfig, ChatMessage } from '../types';
 import { streamMessageAnthropicProxy } from './anthropic-compatible';
 import { isOpenAICompatible, streamMessageOpenAI } from './openai-compatible';
@@ -57,7 +58,7 @@ export async function streamMessage(
     const stream = client.messages.stream(
       {
         model: cfg.model,
-        max_tokens: 8192,
+        max_tokens: effectiveMaxTokens(cfg),
         system,
         messages: history.map((m) => ({ role: m.role, content: m.content })),
       },

apps/web/src/providers/openai-compatible.ts

@@ -5,6 +5,7 @@
  * Routes through the daemon proxy to avoid browser CORS issues.
  * BYOK — the key stays on the user's machine.
  */
+import { effectiveMaxTokens } from '../state/maxTokens';
 import type { AppConfig, ChatMessage } from '../types';
 import type { StreamHandlers } from './anthropic';
 import { parseSseFrame } from './sse';
@@ -33,6 +34,7 @@ export async function streamMessageOpenAI(
         model: cfg.model,
         systemPrompt: system,
         messages: history.map((m) => ({ role: m.role, content: m.content })),
+        maxTokens: effectiveMaxTokens(cfg),
       }),
       signal,
     });