pleumrouter 0.5.30 → 0.5.31

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pleumrouter",
-  "version": "0.5.30",
+  "version": "0.5.31",
   "description": "Official PleumRouter terminal CLI — launch any coding agent against PleumRouter.",
   "type": "module",
   "bin": {

package/src/ansi.mjs

@@ -102,6 +102,10 @@ export function terminalWidth(fallback = 80) {
   return process.stdout.columns || fallback;
 }
 
+export function terminalHeight(fallback = 24) {
+  return process.stdout.rows || fallback;
+}
+
 export function divider(width = Math.min(terminalWidth(), 60), styler = style) {
   return styler.gray("─".repeat(Math.max(1, width)));
 }

package/src/interactive.mjs

@@ -109,6 +109,7 @@ export async function select(items, {
   return new Promise((resolve) => {
     let query = "";
     let cursor = Math.max(0, Math.min(initialCursor, Math.max(0, (items?.length ?? 1) - 1)));
+    let windowStart = 0;
     let lastRows = 0;
 
     const rowsFor = (s) => Math.max(1, Math.ceil(displayWidth(s) / Math.max(terminalWidth(), 1)));
@@ -121,15 +122,33 @@ export async function select(items, {
     const render = () => {
       const pool = matches(items, query);
       if (cursor >= pool.length) cursor = Math.max(0, pool.length - 1);
-      const shown = pool.slice(0, limit);
+      if (cursor < windowStart) {
+        windowStart = cursor;
+      } else if (cursor >= windowStart + limit) {
+        windowStart = cursor - limit + 1;
+      }
+      windowStart = Math.max(0, Math.min(windowStart, Math.max(0, pool.length - limit)));
+
+      const shown = pool.slice(windowStart, windowStart + limit);
       const lines = [
         `${style.plum("?")} ${style.bold(message)} ${style.dim(t("type to filter · ↑↓ move · Enter select · Esc cancel"))}`,
         `${style.gray("›")} ${query}`,
-        ...shown.map((it, i) =>
-          i === cursor ? `${style.plum("❯")} ${style.bold(formatItem(it))}` : `  ${style.dim(formatItem(it))}`,
-        ),
       ];
-      if (pool.length > shown.length) lines.push(style.dim(`  ${t("… {n} more — type to narrow down", { n: pool.length - shown.length })}`));
+      if (windowStart > 0) {
+        lines.push(style.dim(`  ${t("↑ {n} more above", { n: windowStart })}`));
+      }
+      lines.push(
+        ...shown.map((it, idx) => {
+          const itemIndex = windowStart + idx;
+          return itemIndex === cursor
+            ? `${style.plum("❯")} ${style.bold(formatItem(it))}`
+            : `  ${style.dim(formatItem(it))}`;
+        }),
+      );
+      const remainingBelow = pool.length - (windowStart + shown.length);
+      if (remainingBelow > 0) {
+        lines.push(style.dim(`  ${t("… {n} more — type to narrow down", { n: remainingBelow })}`));
+      }
       if (!pool.length) lines.push(style.dim(`  ${t("No matches")}`));
       erase();
       const frame = lines.join("\n");
@@ -158,12 +177,19 @@ export async function select(items, {
           const safeCursor = Math.max(0, Math.min(cursor, pool.length - 1));
           return finish(pool[safeCursor] ?? null);
         }
-        if (key.type === "up") cursor = Math.max(0, cursor - 1);
-        else if (key.type === "down") cursor = Math.min(limit - 1, cursor + 1);
-        else if (key.type === "backspace") query = query.slice(0, -1);
-        else if (key.type === "char") {
+        if (key.type === "up") {
+          cursor = Math.max(0, cursor - 1);
+        } else if (key.type === "down") {
+          const pool = matches(items, query);
+          cursor = Math.min(Math.max(0, pool.length - 1), cursor + 1);
+        } else if (key.type === "backspace") {
+          query = query.slice(0, -1);
+          cursor = 0;
+          windowStart = 0;
+        } else if (key.type === "char") {
           query += key.value;
           cursor = 0;
+          windowStart = 0;
         } else continue;
         dirty = true;
       }

package/src/locales/es.mjs

@@ -71,8 +71,16 @@ export const ES = {
     "Necesita GitHub Copilot CLI 2026-04 o posterior para el enrutamiento de proveedores BYOK.",
   "Gemini CLI speaks Google's native generateContent format — set the model after launch with `/model google/<id>`.":
     "Gemini CLI usa el formato nativo generateContent de Google — establece el modelo tras el inicio con `/model google/<id>`.",
-  "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. GEMINI_API_KEY alone is ignored. Gemini-family models only.":
-    "Escribe modelProvider=gemini en ~/.gemini/antigravity-cli/settings.json. GEMINI_API_KEY sola se ignora. Solo familia Gemini.",
+  "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. Routes any coding model via PleumRouter.":
+    "Escribe modelProvider=gemini en ~/.gemini/antigravity-cli/settings.json. Enruta cualquier modelo de codificación a través de PleumRouter.",
+  "↑ {n} more above": "↑ {n} más arriba",
+  "↓ {n} more below": "↓ {n} más abajo",
+  "Auto (best route)": "Auto (mejor ruta)",
+  "PleumRouter auto-routing — lowest price, lowest latency, automatic failover":
+    "Enrutamiento automático de PleumRouter: mejor precio, menor latencia y conmutación por error",
+  "Pin to {provider} (@{provider})": "Fijar al proveedor {provider} (@{provider})",
+  "Select routing provider": "Seleccionar proveedor de enrutamiento",
+  "... view all 200+ models": "... ver todos los 200+ modelos",
   "Uses an isolated ~/.pleum/pi-home (PI_CODING_AGENT_DIR). Your ~/.pi is untouched.":
     "Usa un ~/.pleum/pi-home aislado (PI_CODING_AGENT_DIR). No toca ~/.pi.",
   "Uses an isolated ~/.pleum/crush-home (CRUSH_GLOBAL_CONFIG). Your crush.json is untouched.":

package/src/locales/ja.mjs

@@ -72,8 +72,16 @@ export const JA = {
     "BYOK プロバイダールーティングに対応した 2026-04 以降の GitHub Copilot CLI が必要です。",
   "Gemini CLI speaks Google's native generateContent format — set the model after launch with `/model google/<id>`.":
     "Gemini CLI は Google のネイティブ generateContent 形式を使用します — 起動後に `/model google/<id>` でモデルを設定してください。",
-  "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. GEMINI_API_KEY alone is ignored. Gemini-family models only.":
-    "~/.gemini/antigravity-cli/settings.json に modelProvider=gemini を書き込みます。GEMINI_API_KEY だけでは無視されます。Gemini 系のみです。",
+  "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. Routes any coding model via PleumRouter.":
+    "~/.gemini/antigravity-cli/settings.json に modelProvider=gemini を書き込みます。あらゆるコーディングモデルを PleumRouter でルーティングします。",
+  "↑ {n} more above": "↑ 上にあと {n} 件",
+  "↓ {n} more below": "↓ 下にあと {n} 件",
+  "Auto (best route)": "Auto (最適な自動ルーティング)",
+  "PleumRouter auto-routing — lowest price, lowest latency, automatic failover":
+    "PleumRouter 自動ルーティング — 最安値・低遅延・自動フェイルオーバー",
+  "Pin to {provider} (@{provider})": "{provider} プロバイダーに固定 (@{provider})",
+  "Select routing provider": "ルーティングプロバイダーの選択",
+  "... view all 200+ models": "... 全 200+ モデルを表示",
   "Uses an isolated ~/.pleum/pi-home (PI_CODING_AGENT_DIR). Your ~/.pi is untouched.":
     "隔離した ~/.pleum/pi-home (PI_CODING_AGENT_DIR) だけを使います。~/.pi は触れません。",
   "Uses an isolated ~/.pleum/crush-home (CRUSH_GLOBAL_CONFIG). Your crush.json is untouched.":

package/src/locales/ko.mjs

@@ -52,8 +52,16 @@ export const KO = {
     "BYOK provider 라우팅을 지원하는 2026-04 이후 버전의 GitHub Copilot CLI가 필요합니다.",
   "Gemini CLI speaks Google's native generateContent format — set the model after launch with `/model google/<id>`.":
     "Gemini CLI는 Google의 네이티브 generateContent 포맷을 사용합니다 — 실행 후 `/model google/<id>`로 모델을 설정하세요.",
-  "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. GEMINI_API_KEY alone is ignored. Gemini-family models only.":
-    "~/.gemini/antigravity-cli/settings.json 에 modelProvider=gemini 를 씁니다. GEMINI_API_KEY만 있으면 무시됩니다. Gemini 계열만 됩니다.",
+  "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. Routes any coding model via PleumRouter.":
+    "~/.gemini/antigravity-cli/settings.json 에 modelProvider=gemini 를 씁니다. 모든 코딩 모델을 PleumRouter로 라우팅합니다.",
+  "↑ {n} more above": "↑ 위에 {n}개 더",
+  "↓ {n} more below": "↓ 아래에 {n}개 더",
+  "Auto (best route)": "Auto (최적 자동 라우팅)",
+  "PleumRouter auto-routing — lowest price, lowest latency, automatic failover":
+    "PleumRouter 최적 자동 라우팅 — 최저가·최저지연·자동 페일오버",
+  "Pin to {provider} (@{provider})": "{provider} 제공자로 고정 (@{provider})",
+  "Select routing provider": "라우팅 제공자 선택",
+  "... view all 200+ models": "... 전체 200+ 모델 보기",
   "Uses an isolated ~/.pleum/pi-home (PI_CODING_AGENT_DIR). Your ~/.pi is untouched.":
     "격리된 ~/.pleum/pi-home (PI_CODING_AGENT_DIR)만 씁니다. ~/.pi는 건드리지 않습니다.",
   "Uses an isolated ~/.pleum/crush-home (CRUSH_GLOBAL_CONFIG). Your crush.json is untouched.":

package/src/locales/zh.mjs

@@ -71,8 +71,16 @@ export const ZH = {
     "需要 2026-04 或更新版本的 GitHub Copilot CLI 才支持 BYOK provider 路由。",
   "Gemini CLI speaks Google's native generateContent format — set the model after launch with `/model google/<id>`.":
     "Gemini CLI 使用 Google 原生 generateContent 格式 — 启动后用 `/model google/<id>` 设置模型。",
-  "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. GEMINI_API_KEY alone is ignored. Gemini-family models only.":
-    "会把 modelProvider=gemini 写入 ~/.gemini/antigravity-cli/settings.json。只设 GEMINI_API_KEY 会被忽略。仅 Gemini 系列。",
+  "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. Routes any coding model via PleumRouter.":
+    "将 modelProvider=gemini 写入 ~/.gemini/antigravity-cli/settings.json。通过 PleumRouter 路由所有编码模型。",
+  "↑ {n} more above": "↑ 上方还有 {n} 项",
+  "↓ {n} more below": "↓ 下方还有 {n} 项",
+  "Auto (best route)": "Auto (最佳自动路由)",
+  "PleumRouter auto-routing — lowest price, lowest latency, automatic failover":
+    "PleumRouter 自动路由 — 最优价格·超低延迟·自动故障转移",
+  "Pin to {provider} (@{provider})": "固定到 {provider} 提供商 (@{provider})",
+  "Select routing provider": "选择路由提供商",
+  "... view all 200+ models": "... 查看全部 200+ 个模型",
   "Uses an isolated ~/.pleum/pi-home (PI_CODING_AGENT_DIR). Your ~/.pi is untouched.":
     "只用隔离的 ~/.pleum/pi-home（PI_CODING_AGENT_DIR），不改 ~/.pi。",
   "Uses an isolated ~/.pleum/crush-home (CRUSH_GLOBAL_CONFIG). Your crush.json is untouched.":

package/src/menu.mjs

@@ -11,12 +11,12 @@
 import { readFileSync } from "node:fs";
 import { spawn } from "node:child_process";
 import { createInterface } from "node:readline";
-import { style, displayWidth, terminalWidth, hyperlink, sanitizeTerminalText } from "./ansi.mjs";
+import { style, displayWidth, terminalWidth, terminalHeight, hyperlink, sanitizeTerminalText } from "./ansi.mjs";
 import { keyFeeder, select } from "./interactive.mjs";
 import { readConfig, writeConfig, isLoggedIn, touchSession, getActiveContext } from "./config.mjs";
 import { RECIPES, MANUAL, DEFAULT_MODEL } from "./recipes.mjs";
 import { onPath, launch } from "./launch.mjs";
-import { fetchModels, matchModels, listModels, formatModelLabel } from "./models.mjs";
+import { fetchModels, matchModels, listModels, formatModelLabel, fetchCoreCodingModels, CORE_MODEL_PROVIDERS } from "./models.mjs";
 import { run } from "./run.mjs";
 import { login } from "./login.mjs";
 import { top } from "./top.mjs";
@@ -123,24 +123,81 @@ function saveLastSelection(id) {
   writeConfig(cfg);
 }
 
+export async function pickProviderForModel(modelId, { input = process.stdin, output = process.stdout } = {}) {
+  if (!modelId || modelId === "__view_all__") return modelId;
+  const providers = CORE_MODEL_PROVIDERS[modelId] || [];
+  if (!providers.length) return modelId;
+
+  const options = [
+    {
+      id: null,
+      title: `⚡ ${t("Auto (best route)")}`,
+      desc: t("PleumRouter auto-routing — lowest price, lowest latency, automatic failover"),
+    },
+    ...providers.map((p) => ({
+      id: p.id,
+      title: p.name,
+      desc: t("Pin to {provider} (@{provider})", { provider: p.id }),
+    })),
+  ];
+
+  const chosen = await select(options, {
+    message: `${modelId} — ${t("Select routing provider")}`,
+    formatItem: (x) => `${x.title}  ${style.dim(x.desc || "")}`,
+    limit: 8,
+    input,
+    output,
+  });
+
+  if (!chosen) return null;
+  if (!chosen.id) return modelId;
+  return `${modelId}@${chosen.id}`;
+}
+
 export async function pickModelFor(item, defaultModel) {
-  let models;
-  try {
-    models = await fetchModels();
-  } catch (e) {
-    console.log(`${style.red("✗")} ${t("Failed to fetch model list: {error}", { error: sanitizeTerminalText(e.message) })}`);
-    return undefined;
+  for (;;) {
+    const core = fetchCoreCodingModels();
+    const items = [
+      ...core,
+      {
+        id: "__view_all__",
+        display_name: t("... view all 200+ models"),
+      },
+    ];
+    const targetId = defaultModel || DEFAULT_MODEL;
+    const initialIndex = items.findIndex((x) => x.id === targetId || targetId.startsWith(`${x.id}@`));
+    const m = await select(items, {
+      message: `${item.title} — ${t("select a model")}`,
+      formatItem: (x) => (x.id === "__view_all__" ? style.dim(x.display_name) : formatModelLabel(x)),
+      matches: matchModels,
+      limit: 11,
+      initialCursor: initialIndex >= 0 ? initialIndex : 0,
+    });
+    if (!m) return undefined;
+
+    let pickedModelId = m.id;
+    if (pickedModelId === "__view_all__") {
+      let allModels;
+      try {
+        allModels = await fetchModels();
+      } catch (e) {
+        console.log(`${style.red("✗")} ${t("Failed to fetch model list: {error}", { error: sanitizeTerminalText(e.message) })}`);
+        return undefined;
+      }
+      const allPicked = await select(allModels, {
+        message: `${item.title} — ${t("select a model")}`,
+        formatItem: (x) => formatModelLabel(x),
+        matches: matchModels,
+        limit: 10,
+      });
+      if (!allPicked) continue;
+      pickedModelId = allPicked.id;
+    }
+
+    const withProvider = await pickProviderForModel(pickedModelId);
+    if (!withProvider) continue;
+    return withProvider;
   }
-  const targetId = defaultModel || DEFAULT_MODEL;
-  const initialIndex = models.findIndex((x) => x.id === targetId);
-  const m = await select(models, {
-    message: `${item.title} — ${t("select a model")}`,
-    formatItem: (x) => formatModelLabel(x),
-    matches: matchModels,
-    limit: 10,
-    initialCursor: initialIndex >= 0 ? initialIndex : 0,
-  });
-  return m?.id;
 }
 
 // 미설치 도구: ollama처럼 확인 후 그 자리에서 설치(stdio 상속). 성공 시 true.
@@ -184,6 +241,7 @@ export async function runList(items, { header, footer, initialCursor = 0 } = {})
   if (!input.isTTY || !output.isTTY || typeof input.setRawMode !== "function") return undefined;
 
   let cursor = Math.max(0, Math.min(initialCursor, items.length - 1));
+  let windowStart = 0;
   let lastRows = 0;
 
   const rowsFor = (s) => Math.max(1, Math.ceil(displayWidth(s) / Math.max(terminalWidth(), 1)));
@@ -192,28 +250,68 @@ export async function runList(items, { header, footer, initialCursor = 0 } = {})
     output.write("\r" + (lastRows > 1 ? `\x1b[${lastRows - 1}A` : "") + "\x1b[0J");
     lastRows = 0;
   };
+
+  const getMaxVisible = () => {
+    const h = terminalHeight();
+    return Math.max(3, Math.min(9, Math.floor((h - 6) / 2)));
+  };
+
   const render = () => {
+    const maxVisible = getMaxVisible();
+    if (cursor < windowStart) {
+      windowStart = cursor;
+    } else if (cursor >= windowStart + maxVisible) {
+      windowStart = cursor - maxVisible + 1;
+    }
+    windowStart = Math.max(0, Math.min(windowStart, Math.max(0, items.length - maxVisible)));
+
+    const visibleItems = items.slice(windowStart, windowStart + maxVisible);
     const lines = [
       header,
       "",
-      ...items.flatMap((it, i) =>
-        i === cursor
-          ? [`${style.plum("▸")} ${style.bold(it.title)}`, `    ${style.dim(it.desc)}`]
-          : [`  ${it.title}`, `    ${style.dim(it.desc)}`],
-      ),
-      "",
-      style.dim(footer),
     ];
+
+    if (windowStart > 0) {
+      lines.push(style.dim(`  ${t("↑ {n} more above", { n: windowStart })}`));
+    }
+
+    for (let idx = 0; idx < visibleItems.length; idx++) {
+      const i = windowStart + idx;
+      const it = visibleItems[idx];
+      if (i === cursor) {
+        lines.push(`${style.plum("▸")} ${style.bold(it.title)}`);
+        lines.push(`    ${style.dim(it.desc)}`);
+      } else {
+        lines.push(`  ${it.title}`);
+        lines.push(`    ${style.dim(it.desc)}`);
+      }
+    }
+
+    const remainingBelow = items.length - (windowStart + visibleItems.length);
+    if (remainingBelow > 0) {
+      lines.push(style.dim(`  ${t("↓ {n} more below", { n: remainingBelow })}`));
+    }
+
+    lines.push("");
+    lines.push(style.dim(footer));
+
     erase();
     output.write(lines.join("\n"));
     lastRows = lines.reduce((n, l) => n + rowsFor(l), 0);
   };
+
   const move = (dir) => {
     cursor = Math.max(0, Math.min(items.length - 1, cursor + dir));
   };
 
   return new Promise((resolve) => {
+    const onResize = () => {
+      render();
+    };
+    output.on("resize", onResize);
+
     const suspend = () => {
+      output.off("resize", onResize);
       input.off("data", onData);
       input.setRawMode(false);
       output.write("\x1b[?25h");
@@ -223,10 +321,12 @@ export async function runList(items, { header, footer, initialCursor = 0 } = {})
       input.setRawMode(true);
       input.resume();
       output.write("\x1b[?25l");
+      output.on("resize", onResize);
       input.on("data", onData);
       render();
     };
     const finish = (value) => {
+      output.off("resize", onResize);
       suspend();
       input.pause();
       resolve(value);

package/src/models.mjs

@@ -1,13 +1,199 @@
+import { homedir } from "node:os";
+import { join } from "node:path";
+import { readFileSync, writeFileSync, mkdirSync, statSync } from "node:fs";
 import { V1 } from "./config.mjs";
 import { style, padDisplay, divider, sanitizeTerminalText } from "./ansi.mjs";
 import { t } from "./i18n.mjs";
 
+export const CORE_MODEL_PROVIDERS = {
+  "deepseek-ai/DeepSeek-V4.1-Flash": [
+    { id: "deepseek", name: "DeepSeek (Official)" },
+    { id: "together", name: "Together AI" },
+    { id: "fireworks", name: "Fireworks AI" },
+    { id: "siliconflow", name: "SiliconFlow" },
+  ],
+  "claude-sonnet-5": [
+    { id: "anthropic", name: "Anthropic (Official)" },
+    { id: "aws_bedrock", name: "AWS Bedrock" },
+    { id: "gcp_vertex", name: "Google Cloud Vertex AI" },
+  ],
+  "claude-opus-5-5": [
+    { id: "anthropic", name: "Anthropic (Official)" },
+    { id: "aws_bedrock", name: "AWS Bedrock" },
+    { id: "gcp_vertex", name: "Google Cloud Vertex AI" },
+  ],
+  "gpt-6-astra": [
+    { id: "openai", name: "OpenAI (Official)" },
+    { id: "azure", name: "Microsoft Azure OpenAI" },
+  ],
+  "gpt-6-sol": [
+    { id: "openai", name: "OpenAI (Official)" },
+    { id: "azure", name: "Microsoft Azure OpenAI" },
+  ],
+  "gpt-6-luna": [
+    { id: "openai", name: "OpenAI (Official)" },
+    { id: "azure", name: "Microsoft Azure OpenAI" },
+  ],
+  "gemini-3.8-flash": [
+    { id: "google", name: "Google AI Studio (Official)" },
+    { id: "vertex", name: "Google Cloud Vertex AI" },
+  ],
+  "qwen3.8-max": [
+    { id: "alibaba", name: "Alibaba DashScope (Official)" },
+    { id: "together", name: "Together AI" },
+    { id: "siliconflow", name: "SiliconFlow" },
+  ],
+  "glm-5.3-flash": [
+    { id: "zhipu", name: "Zhipu AI BigModel (Official)" },
+    { id: "modelark", name: "ByteDance ModelArk" },
+  ],
+  "kimi-k3": [
+    { id: "moonshot", name: "Moonshot AI (Official)" },
+  ],
+};
+
+export const CORE_MODEL_DEFAULTS = [
+  {
+    id: "deepseek-ai/DeepSeek-V4.1-Flash",
+    display_name: "DeepSeek V4.1 Flash (Ultra-fast coding)",
+    maker: "DeepSeek",
+    context_length: 131072,
+    pricing: { prompt: "0.00000014", completion: "0.00000028" },
+  },
+  {
+    id: "claude-sonnet-5",
+    display_name: "Claude Sonnet 5 (Anthropic flagship coding)",
+    maker: "Anthropic",
+    context_length: 200000,
+    pricing: { prompt: "0.000003", completion: "0.000015" },
+  },
+  {
+    id: "claude-opus-5-5",
+    display_name: "Claude Opus 5.5 (Deep reasoning & architecture)",
+    maker: "Anthropic",
+    context_length: 200000,
+    pricing: { prompt: "0.000005", completion: "0.000025" },
+  },
+  {
+    id: "gpt-6-astra",
+    display_name: "GPT-6 Astra (OpenAI flagship reasoning)",
+    maker: "OpenAI",
+    context_length: 128000,
+    pricing: { prompt: "0.0000025", completion: "0.00001" },
+  },
+  {
+    id: "gpt-6-sol",
+    display_name: "GPT-6 Sol (Fast coding & logic)",
+    maker: "OpenAI",
+    context_length: 128000,
+    pricing: { prompt: "0.000001", completion: "0.000004" },
+  },
+  {
+    id: "gpt-6-luna",
+    display_name: "GPT-6 Luna (Lightweight coding)",
+    maker: "OpenAI",
+    context_length: 128000,
+    pricing: { prompt: "0.0000005", completion: "0.000002" },
+  },
+  {
+    id: "gemini-3.8-flash",
+    display_name: "Gemini 3.8 Flash (1M context fast multimodal)",
+    maker: "Google",
+    context_length: 1048576,
+    pricing: { prompt: "0.00000035", completion: "0.0000014" },
+  },
+  {
+    id: "qwen3.8-max",
+    display_name: "Qwen 3.8 Max (Alibaba flagship coding)",
+    maker: "Alibaba",
+    context_length: 131072,
+    pricing: { prompt: "0.0000016", completion: "0.0000064" },
+  },
+  {
+    id: "glm-5.3-flash",
+    display_name: "GLM 5.3 Flash (Zhipu AI high-speed coding)",
+    maker: "Zhipu AI",
+    context_length: 128000,
+    pricing: { prompt: "0.0000001", completion: "0.0000002" },
+  },
+  {
+    id: "kimi-k3",
+    display_name: "Kimi K3 (Moonshot 256K long-context coding)",
+    maker: "Moonshot",
+    context_length: 262144,
+    pricing: { prompt: "0.0000006", completion: "0.0000024" },
+  },
+];
+
+let memoryModelsCache = null;
+let memoryCacheExpiry = 0;
+const CACHE_TTL_MS = 60 * 60 * 1000; // 1 hour
+
+function getCacheFilePath() {
+  return join(homedir(), ".pleum", "models-cache.json");
+}
+
+export function fetchCoreCodingModels() {
+  const cacheFile = getCacheFilePath();
+  let liveList = memoryModelsCache;
+  if (!liveList) {
+    try {
+      liveList = JSON.parse(readFileSync(cacheFile, "utf8"));
+    } catch {}
+  }
+  const byId = new Map((liveList || []).map((m) => [m.id, m]));
+  return CORE_MODEL_DEFAULTS.map((def) => {
+    const live = byId.get(def.id);
+    return live ? { ...def, ...live } : { ...def };
+  });
+}
+
 // GET /v1/models 는 공개(인증 불필요). 슬러그는 provider/model 형식.
-export async function fetchModels() {
-  const res = await fetch(`${V1}/models`);
-  if (!res.ok) throw new Error(t("Failed to fetch model list ({status})", { status: res.status }));
-  const data = await res.json();
-  return data.data ?? [];
+export async function fetchModels({ forceRefresh = false } = {}) {
+  const now = Date.now();
+  if (!forceRefresh && memoryModelsCache && now < memoryCacheExpiry) {
+    return memoryModelsCache;
+  }
+  const cacheFile = getCacheFilePath();
+  if (!forceRefresh && !memoryModelsCache) {
+    try {
+      const stats = statSync(cacheFile);
+      if (now - stats.mtimeMs < CACHE_TTL_MS) {
+        const cached = JSON.parse(readFileSync(cacheFile, "utf8"));
+        if (Array.isArray(cached) && cached.length > 0) {
+          memoryModelsCache = cached;
+          memoryCacheExpiry = stats.mtimeMs + CACHE_TTL_MS;
+          return cached;
+        }
+      }
+    } catch {}
+  }
+
+  try {
+    const res = await fetch(`${V1}/models`);
+    if (!res.ok) throw new Error(t("Failed to fetch model list ({status})", { status: res.status }));
+    const data = await res.json();
+    const list = data.data ?? [];
+    if (Array.isArray(list) && list.length > 0) {
+      memoryModelsCache = list;
+      memoryCacheExpiry = now + CACHE_TTL_MS;
+      try {
+        mkdirSync(join(homedir(), ".pleum"), { recursive: true, mode: 0o700 });
+        writeFileSync(cacheFile, JSON.stringify(list), { mode: 0o600 });
+      } catch {}
+    }
+    return list;
+  } catch (err) {
+    try {
+      const cached = JSON.parse(readFileSync(cacheFile, "utf8"));
+      if (Array.isArray(cached) && cached.length > 0) {
+        memoryModelsCache = cached;
+        memoryCacheExpiry = now + 60 * 1000;
+        return cached;
+      }
+    } catch {}
+    throw err;
+  }
 }
 
 // id/표시명 부분일치 필터 — run 피커·REPL /models·목록이 공유하는 단일 구현.

package/src/recipes.mjs

@@ -7,12 +7,12 @@ export const CORE_MODELS = [
   "claude-sonnet-5",
   "claude-opus-5-5",
   "gpt-6-astra",
-  "claude-fable-5-1",
+  "gpt-6-sol",
+  "gpt-6-luna",
   "gemini-3.8-flash",
+  "qwen3.8-max",
   "glm-5.3-flash",
   "kimi-k3",
-  "gpt-6-sol",
-  "gpt-6-luna",
 ];
 
 // 도구별 설정의 진실 원천. 출처: frontend .../docs/cookbook/{codex,claude-code,...} + 에이전트별 공식 문서.
@@ -48,6 +48,8 @@ export const RECIPES = {
       ANTHROPIC_AUTH_TOKEN: key,
       // /model 이 GET {ANTHROPIC_BASE_URL}/v1/models 를 치게 한다. 기본값은 꺼져 있다.
       CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY: "1",
+      ANTHROPIC_SMALL_MODEL: "anthropic/deepseek-ai/DeepSeek-V4.1-Flash",
+      ANTHROPIC_DEFAULT_HAIKU_MODEL: "anthropic/deepseek-ai/DeepSeek-V4.1-Flash",
     }),
     args: (m) => ["--model", (m || DEFAULT_MODEL).startsWith("anthropic/") ? (m || DEFAULT_MODEL) : `anthropic/${m || DEFAULT_MODEL}`],
   },
@@ -63,6 +65,55 @@ export const RECIPES = {
     env: (key) => ({ PLEUM_API_KEY: key }),
     args: () => [],
   },
+  agy: {
+    label: "Antigravity CLI",
+    bin: "agy",
+    type: "agy-settings",
+    install: "https://antigravity.google/cli/install.sh",
+    note: "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. Routes any coding model via PleumRouter.",
+    attribution: {
+      title: "Antigravity CLI",
+      referer: "https://www.antigravity.google/docs/cli/install",
+    },
+    env: (key) => ({
+      GOOGLE_GEMINI_BASE_URL: ROOT,
+      GEMINI_API_KEY: key,
+      GOOGLE_API_KEY: key,
+    }),
+    args: (m) => ["--model", m || DEFAULT_MODEL],
+  },
+  opencode: {
+    label: "opencode",
+    bin: "opencode",
+    type: "env",
+    install: "npm i -g opencode-ai@latest",
+    // opencode는 커스텀 엔드포인트를 opencode.json provider 블록으로만 읽는다 — env var는 없다.
+    // 대신 OPENCODE_CONFIG_CONTENT(파일 대신 JSON 문자열을 그대로 읽는 env var)로 파일 없이 주입한다.
+    note: "Generates a temporary PleumRouter provider config for this run only — your own opencode.json is untouched.",
+    attribution: {
+      title: "opencode",
+      referer: "https://opencode.ai",
+    },
+    env: (key, model) => {
+      const active = model || DEFAULT_MODEL;
+      const models = Object.fromEntries(CORE_MODELS.map((id) => [id, {}]));
+      models[active] = {};
+      return {
+        PLEUM_API_KEY: key,
+        OPENCODE_CONFIG_CONTENT: JSON.stringify({
+          provider: {
+            pleum: {
+              npm: "@ai-sdk/openai-compatible",
+              name: "PleumRouter",
+              options: { baseURL: V1, apiKey: "{env:PLEUM_API_KEY}" },
+              models,
+            },
+          },
+        }),
+      };
+    },
+    args: (m) => ["--model", `pleum/${m || DEFAULT_MODEL}`],
+  },
   aider: {
     label: "Aider",
     bin: "aider",
@@ -101,38 +152,6 @@ export const RECIPES = {
     env: (key) => ({ OPENAI_API_BASE: V1, OPENAI_API_KEY: key }),
     args: () => [],
   },
-  opencode: {
-    label: "opencode",
-    bin: "opencode",
-    type: "env",
-    install: "npm i -g opencode-ai@latest",
-    // opencode는 커스텀 엔드포인트를 opencode.json provider 블록으로만 읽는다 — env var는 없다.
-    // 대신 OPENCODE_CONFIG_CONTENT(파일 대신 JSON 문자열을 그대로 읽는 env var)로 파일 없이 주입한다.
-    note: "Generates a temporary PleumRouter provider config for this run only — your own opencode.json is untouched.",
-    attribution: {
-      title: "opencode",
-      referer: "https://opencode.ai",
-    },
-    env: (key, model) => {
-      const active = model || DEFAULT_MODEL;
-      const models = Object.fromEntries(CORE_MODELS.map((id) => [id, {}]));
-      models[active] = {};
-      return {
-        PLEUM_API_KEY: key,
-        OPENCODE_CONFIG_CONTENT: JSON.stringify({
-          provider: {
-            pleum: {
-              npm: "@ai-sdk/openai-compatible",
-              name: "PleumRouter",
-              options: { baseURL: V1, apiKey: "{env:PLEUM_API_KEY}" },
-              models,
-            },
-          },
-        }),
-      };
-    },
-    args: (m) => ["--model", `pleum/${m || DEFAULT_MODEL}`],
-  },
   omp: {
     label: "omp (Oh My Pi)",
     bin: "omp",
@@ -211,23 +230,6 @@ export const RECIPES = {
     }),
     args: (m) => ["--model", (m || DEFAULT_MODEL).startsWith("google/") ? (m || DEFAULT_MODEL) : `google/${m || DEFAULT_MODEL}`],
   },
-  agy: {
-    label: "Antigravity CLI",
-    bin: "agy",
-    type: "agy-settings",
-    install: "https://antigravity.google/cli/install.sh",
-    note: "Writes modelProvider=gemini into ~/.gemini/antigravity-cli/settings.json. GEMINI_API_KEY alone is ignored. Gemini-family models only.",
-    attribution: {
-      title: "Antigravity CLI",
-      referer: "https://www.antigravity.google/docs/cli/install",
-    },
-    env: (key) => ({
-      GOOGLE_GEMINI_BASE_URL: ROOT,
-      GEMINI_API_KEY: key,
-      GOOGLE_API_KEY: key,
-    }),
-    args: (m) => ["--model", (m || DEFAULT_MODEL).startsWith("google/") ? (m || DEFAULT_MODEL) : `google/${m || DEFAULT_MODEL}`],
-  },
   pi: {
     label: "Pi",
     bin: "pi",