@kenkaiiii/gg-core 5.91.0 → 5.91.2

package/dist/index.cjs

@@ -45,6 +45,7 @@ __export(index_exports, {
   MODELS: () => MODELS,
   MOONSHOT_OAUTH_KEY: () => MOONSHOT_OAUTH_KEY,
   NotLoggedInError: () => NotLoggedInError,
+  OPENROUTER_DEFAULT_BASE_URL: () => OPENROUTER_DEFAULT_BASE_URL,
   PLAN_MODE_THINKING_CAP: () => PLAN_MODE_THINKING_CAP,
   SubscriptionUsageError: () => SubscriptionUsageError,
   TelegramBot: () => TelegramBot,
@@ -86,6 +87,7 @@ __export(index_exports, {
   getModel: () => getModel,
   getModelsForProvider: () => getModelsForProvider,
   getNextThinkingLevel: () => getNextThinkingLevel,
+  getProviderCatalog: () => getProviderCatalog,
   getSessionId: () => getSessionId,
   getSummaryModel: () => getSummaryModel,
   getSupportedThinkingLevels: () => getSupportedThinkingLevels,
@@ -115,6 +117,7 @@ __export(index_exports, {
   ollamaServedContext: () => ollamaServedContext,
   openLog: () => openLog,
   parseLocalModelId: () => parseLocalModelId,
+  parseOpenRouterCatalog: () => parseOpenRouterCatalog,
   pendingClaudeCodeVersionRefresh: () => import_gg_ai.pendingClaudeCodeVersionRefresh,
   probeEndpoint: () => probeEndpoint,
   providerStorageKeys: () => providerStorageKeys,
@@ -128,6 +131,7 @@ __export(index_exports, {
   registerRuntimeModels: () => registerRuntimeModels,
   resample: () => resample,
   setProgressCallback: () => setProgressCallback,
+  setProviderCatalog: () => setProviderCatalog,
   toModelInfo: () => toModelInfo,
   transcribeVoice: () => transcribeVoice,
   usesOpenAICodexTransport: () => usesOpenAICodexTransport,
@@ -1334,9 +1338,13 @@ async function acquireLock(lockPath) {
           continue;
         }
       } catch {
-        await import_promises.default.unlink(lockPath).catch(() => {
-        });
-        continue;
+        const stat = await import_promises.default.stat(lockPath).catch(() => null);
+        if (!stat) continue;
+        if (Date.now() - stat.mtimeMs > STALE_TIMEOUT_MS) {
+          await import_promises.default.unlink(lockPath).catch(() => {
+          });
+          continue;
+        }
       }
       if (Date.now() - startTime > MAX_WAIT_MS) {
         await import_promises.default.unlink(lockPath).catch(() => {
@@ -2960,11 +2968,44 @@ function clearRuntimeModels(predicate) {
     if (predicate(model)) runtimeModels.delete(id);
   }
 }
+var providerCatalogs = /* @__PURE__ */ new Map();
+function setProviderCatalog(provider, models) {
+  if (!models) {
+    providerCatalogs.delete(provider);
+    return [];
+  }
+  const installed = /* @__PURE__ */ new Map();
+  for (const model of models) {
+    if (model.provider !== provider) continue;
+    if (MODELS.some((m) => m.id === model.id && m.provider !== provider)) continue;
+    installed.set(model.id, model);
+  }
+  providerCatalogs.set(provider, installed);
+  return [...installed.values()];
+}
+function getProviderCatalog(provider) {
+  const catalog = providerCatalogs.get(provider);
+  return catalog ? [...catalog.values()] : void 0;
+}
 function getAllModels() {
-  return [...MODELS, ...runtimeModels.values()];
+  const synced = [...providerCatalogs.values()].flatMap((c) => [...c.values()]);
+  return [
+    ...MODELS.filter((m) => !providerCatalogs.has(m.provider)),
+    ...synced,
+    ...runtimeModels.values()
+  ];
 }
 function getModel(id) {
-  return MODELS.find((m) => m.id === id) ?? runtimeModels.get(id);
+  const fixed = MODELS.find((m) => m.id === id);
+  if (fixed && !providerCatalogs.has(fixed.provider)) return fixed;
+  return findSynced(id) ?? fixed ?? runtimeModels.get(id);
+}
+function findSynced(id) {
+  for (const catalog of providerCatalogs.values()) {
+    const hit = catalog.get(id);
+    if (hit) return hit;
+  }
+  return void 0;
 }
 function getModelsForProvider(provider) {
   return getAllModels().filter((m) => m.provider === provider);
@@ -3070,6 +3111,59 @@ function getSummaryModel(provider, currentModelId) {
   return tagged ?? getModel(currentModelId) ?? getDefaultModel(provider);
 }
 
+// src/openrouter-catalog.ts
+var OPENROUTER_DEFAULT_BASE_URL = "https://openrouter.ai/api/v1";
+var EFFORT_LADDER = ["low", "medium", "high", "xhigh", "max"];
+function positiveInt(value) {
+  return typeof value === "number" && Number.isSafeInteger(value) && value > 0 ? value : void 0;
+}
+function strings(value) {
+  return Array.isArray(value) ? value.filter((v) => typeof v === "string") : [];
+}
+function costTier(prompt) {
+  const perToken = typeof prompt === "string" ? Number(prompt) : NaN;
+  if (!Number.isFinite(perToken) || perToken < 5e-7) return "low";
+  return perToken < 3e-6 ? "medium" : "high";
+}
+function parseOpenRouterCatalog(body) {
+  const data = body?.data;
+  if (!Array.isArray(data)) return { models: [], skipped: 0 };
+  const models = [];
+  let skipped = 0;
+  const seen = /* @__PURE__ */ new Set();
+  for (const raw of data) {
+    const id = typeof raw?.id === "string" ? raw.id.trim() : "";
+    const params = strings(raw?.supported_parameters);
+    const contextWindow = positiveInt(raw?.top_provider?.context_length) ?? positiveInt(raw?.context_length);
+    if (!id || seen.has(id) || !params.includes("tools") || !contextWindow) {
+      skipped++;
+      continue;
+    }
+    seen.add(id);
+    const efforts = strings(raw.reasoning?.supported_efforts);
+    const ladder = EFFORT_LADDER.filter((rung) => efforts.includes(rung));
+    const supportsThinking = ladder.length > 0 || params.includes("reasoning") || params.includes("reasoning_effort");
+    const defaultEffort = raw.reasoning?.default_effort;
+    const maxOutput = positiveInt(raw.top_provider?.max_completion_tokens) ?? Math.min(contextWindow, 16384);
+    models.push({
+      id,
+      name: typeof raw.name === "string" && raw.name.trim() ? raw.name.trim() : id,
+      provider: "openrouter",
+      contextWindow,
+      maxOutputTokens: Math.min(maxOutput, contextWindow),
+      supportsThinking,
+      supportsImages: strings(raw.architecture?.input_modalities).includes("image"),
+      // Video needs a per-transport payload cap the catalog doesn't publish.
+      supportsVideo: false,
+      costTier: costTier(raw.pricing?.prompt),
+      maxThinkingLevel: ladder[ladder.length - 1] ?? "high",
+      ...ladder.length > 0 ? { thinkingLevels: ladder } : {},
+      ...typeof defaultEffort === "string" && ladder.includes(defaultEffort) ? { defaultThinkingLevel: defaultEffort } : {}
+    });
+  }
+  return { models, skipped };
+}
+
 // src/thinking-level.ts
 var OPENAI_GPT_THINKING_LEVELS = ["medium", "high", "xhigh"];
 var OPENAI_GPT_56_THINKING_LEVELS = [
@@ -3137,6 +3231,8 @@ function getSupportedThinkingLevels(provider, model) {
     const maxIndex2 = LOCAL_THINKING_LEVELS.indexOf(info.maxThinkingLevel);
     return maxIndex2 === -1 ? LOCAL_THINKING_LEVELS.slice(0, 3) : LOCAL_THINKING_LEVELS.slice(0, maxIndex2 + 1);
   }
+  const declared = getModel(model)?.thinkingLevels;
+  if (declared && declared.length > 0) return declared;
   const maxLevel = getMaxThinkingLevel(model);
   if (isAnthropicAdaptiveModel(provider, model)) {
     const levels2 = isAnthropicXhighModel(provider, model) ? ANTHROPIC_XHIGH_THINKING_LEVELS : ANTHROPIC_ADAPTIVE_THINKING_LEVELS;
@@ -3292,7 +3388,9 @@ async function probeEndpoint(endpoint, { timeoutMs = DEFAULT_PROBE_TIMEOUT_MS, s
   const entries = (list.data ?? []).filter(
     (entry) => typeof entry.id === "string" && entry.id.length > 0 && !NON_CHAT_ID_PATTERN.test(entry.id)
   );
-  const models = await enrich(endpoint, entries, { timeoutMs, signal });
+  const enriched = await enrich(endpoint, entries, { timeoutMs, signal });
+  const pinned = contextLength(endpoint.contextLength);
+  const models = pinned === void 0 ? enriched : enriched.map((m) => ({ ...m, contextWindow: pinned, contextWindowKnown: true }));
   log("INFO", "local-models", `Probed ${endpoint.label}`, {
     baseUrl: endpoint.baseUrl,
     models: String(models.length)
@@ -3318,7 +3416,7 @@ async function enrichCustom(endpoint, entries, options) {
   return generic;
 }
 function genericModel(endpoint, entry) {
-  const declared = contextLength(entry.max_model_len) ?? contextLength(entry.context_length);
+  const declared = contextLength(entry.max_model_len) ?? contextLength(entry.context_length) ?? contextLength(entry.meta?.n_ctx);
   return {
     rawId: entry.id,
     endpointId: endpoint.id,
@@ -3498,7 +3596,7 @@ function toModelInfo(model, endpoint) {
 }
 var cache;
 function cacheKey(endpoints) {
-  return endpoints.map((e) => `${e.id}@${e.baseUrl}`).join("|");
+  return endpoints.map((e) => `${e.id}@${e.baseUrl}#${e.contextLength ?? ""}`).join("|");
 }
 async function discoverLocalModels(endpoints = DEFAULT_LOCAL_ENDPOINTS, options = {}) {
   const key = cacheKey(endpoints);
@@ -4457,6 +4555,7 @@ function createAutoUpdater(config) {
   MODELS,
   MOONSHOT_OAUTH_KEY,
   NotLoggedInError,
+  OPENROUTER_DEFAULT_BASE_URL,
   PLAN_MODE_THINKING_CAP,
   SubscriptionUsageError,
   TelegramBot,
@@ -4498,6 +4597,7 @@ function createAutoUpdater(config) {
   getModel,
   getModelsForProvider,
   getNextThinkingLevel,
+  getProviderCatalog,
   getSessionId,
   getSummaryModel,
   getSupportedThinkingLevels,
@@ -4527,6 +4627,7 @@ function createAutoUpdater(config) {
   ollamaServedContext,
   openLog,
   parseLocalModelId,
+  parseOpenRouterCatalog,
   pendingClaudeCodeVersionRefresh,
   probeEndpoint,
   providerStorageKeys,
@@ -4540,6 +4641,7 @@ function createAutoUpdater(config) {
   registerRuntimeModels,
   resample,
   setProgressCallback,
+  setProviderCatalog,
   toModelInfo,
   transcribeVoice,
   usesOpenAICodexTransport,