@kenkaiiii/gg-core 5.88.1 → 5.90.0

package/dist/index.cjs

@@ -99,6 +99,7 @@ __export(index_exports, {
   kimiCodeBaseUrl: () => kimiCodeBaseUrl,
   kimiCodingHeaders: () => kimiCodingHeaders,
   localAuthStorageKey: () => localAuthStorageKey,
+  localModelShortName: () => localModelShortName,
   log: () => log,
   loginAnthropic: () => loginAnthropic,
   loginGemini: () => loginGemini,
@@ -106,6 +107,7 @@ __export(index_exports, {
   loginOpenAI: () => loginOpenAI,
   loginXai: () => loginXai,
   oauthStorageKey: () => oauthStorageKey,
+  ollamaServedContext: () => ollamaServedContext,
   openLog: () => openLog,
   parseLocalModelId: () => parseLocalModelId,
   pendingClaudeCodeVersionRefresh: () => import_gg_ai.pendingClaudeCodeVersionRefresh,
@@ -1525,6 +1527,7 @@ function isHiddenProvider(provider) {
 }
 
 // src/auth-storage.ts
+var import_gg_ai3 = require("@kenkaiiii/gg-ai");
 var MOONSHOT_OAUTH_KEY = "moonshot-oauth";
 var XAI_OAUTH_KEY = "xai-oauth";
 var DUAL_AUTH_PROVIDERS = [
@@ -1683,7 +1686,7 @@ var AuthStorage = class {
    */
   async setLocalEndpoint(endpointId, baseUrl, apiKey) {
     await this.setCredentials(`${LOCAL_AUTH_KEY_PREFIX}${endpointId}`, {
-      accessToken: apiKey && apiKey.length > 0 ? apiKey : "local",
+      accessToken: apiKey && apiKey.length > 0 ? apiKey : import_gg_ai3.LOCAL_NO_API_KEY,
       refreshToken: "",
       expiresAt: Date.now() + LOCAL_CREDENTIAL_LIFETIME_MS,
       baseUrl
@@ -1825,6 +1828,12 @@ var AuthStorage = class {
       latest[provider] = creds;
     });
   }
+  // Keeps local-endpoint keys: they're the user's own config, and a 401 there is often a busy gateway.
+  async clearRejectedApiKey(storageKey) {
+    if (storageKey.startsWith(LOCAL_AUTH_KEY_PREFIX)) return false;
+    await this.clearCredentials(storageKey);
+    return true;
+  }
   async clearCredentials(provider) {
     await this.mutateLatest((latest) => {
       delete latest[provider];
@@ -3028,11 +3037,12 @@ function clampThinkingForPlanMode(level) {
 }
 
 // src/local-models.ts
+var import_gg_ai4 = require("@kenkaiiii/gg-ai");
 var DEFAULT_LOCAL_ENDPOINTS = [
   { id: "ollama", label: "Ollama", baseUrl: "http://127.0.0.1:11434/v1", kind: "ollama" }
 ];
 var FALLBACK_CONTEXT_WINDOW = 4096;
-var LOCAL_API_KEY_PLACEHOLDER = "local";
+var LOCAL_API_KEY_PLACEHOLDER = import_gg_ai4.LOCAL_NO_API_KEY;
 var DEFAULT_PROBE_TIMEOUT_MS = 1200;
 var ENRICH_CONCURRENCY = 6;
 var CACHE_TTL_MS = 3e4;
@@ -3048,6 +3058,9 @@ function parseLocalModelId(id) {
   if (slash <= 0 || slash === rest.length - 1) return void 0;
   return { endpointId: rest.slice(0, slash), rawId: rest.slice(slash + 1) };
 }
+function localModelShortName(rawId) {
+  return rawId.slice(rawId.lastIndexOf("/") + 1) || rawId;
+}
 function isLocalModelId(id) {
   return parseLocalModelId(id) !== void 0;
 }
@@ -3058,8 +3071,9 @@ function endpointRoot(baseUrl) {
   return baseUrl.replace(/\/+$/, "").replace(/\/v1$/, "");
 }
 function authHeaders(endpoint) {
+  const key = (0, import_gg_ai4.localBearerKey)(endpoint.apiKey);
   return {
-    Authorization: `Bearer ${endpoint.apiKey ?? LOCAL_API_KEY_PLACEHOLDER}`,
+    ...key ? { Authorization: `Bearer ${key}` } : {},
     Accept: "application/json"
   };
 }
@@ -3121,10 +3135,22 @@ async function enrich(endpoint, entries, options) {
   if (endpoint.kind === "lmstudio") return enrichLmStudio(endpoint, entries, options);
   if (endpoint.kind === "ollama") return enrichOllama(endpoint, entries, options);
   if (endpoint.kind === "llamacpp") return enrichLlamaCpp(endpoint, entries, options);
+  if (endpoint.kind === "custom") return enrichCustom(endpoint, entries, options);
   return entries.map((entry) => genericModel(endpoint, entry));
 }
+async function enrichCustom(endpoint, entries, options) {
+  const generic = entries.map((entry) => genericModel(endpoint, entry));
+  if (generic.every((model) => model.contextWindowKnown)) return generic;
+  const root = endpointRoot(endpoint.baseUrl);
+  const props = await fetchJson(`${root}/props`, endpoint, options);
+  const nCtx = contextLength(props?.default_generation_settings?.n_ctx);
+  if (nCtx !== void 0) return llamaCppModels(endpoint, entries, nCtx);
+  const lmStudio = await fetchJson(`${root}/api/v1/models`, endpoint, options);
+  if (Array.isArray(lmStudio?.models)) return enrichLmStudio(endpoint, entries, options, lmStudio);
+  return generic;
+}
 function genericModel(endpoint, entry) {
-  const declared = contextLength(entry.max_model_len);
+  const declared = contextLength(entry.max_model_len) ?? contextLength(entry.context_length);
   return {
     rawId: entry.id,
     endpointId: endpoint.id,
@@ -3151,9 +3177,7 @@ async function enrichOllama(endpoint, entries, options) {
     });
     const caps = Array.isArray(show?.capabilities) ? show.capabilities : [];
     if (caps.includes("embedding") && !caps.includes("completion")) return void 0;
-    const ctx = contextLength(
-      runningModels.find((model) => model?.name === entry.id || model?.model === entry.id)?.context_length
-    );
+    const ctx = runningContext(runningModels, entry.id) ?? modelfileNumCtx(show?.parameters);
     return {
       rawId: entry.id,
       endpointId: endpoint.id,
@@ -3168,12 +3192,41 @@ async function enrichOllama(endpoint, entries, options) {
   });
   return enriched.filter((model) => model !== void 0);
 }
+function runningContext(models, rawId) {
+  return contextLength(
+    models.find((model) => model?.name === rawId || model?.model === rawId)?.context_length
+  );
+}
+function modelfileNumCtx(parameters) {
+  const match = /^num_ctx\s+(\d+)\s*$/m.exec(parameters ?? "");
+  return match ? contextLength(Number(match[1])) : void 0;
+}
+async function ollamaServedContext(endpoint, rawId, {
+  load = false,
+  timeoutMs = 12e4,
+  signal
+} = {}) {
+  const root = endpointRoot(endpoint.baseUrl);
+  if (load) {
+    await fetchJson(`${root}/api/generate`, endpoint, {
+      timeoutMs,
+      signal,
+      method: "POST",
+      body: { model: rawId }
+    });
+  }
+  const running = await fetchJson(`${root}/api/ps`, endpoint, {
+    timeoutMs: DEFAULT_PROBE_TIMEOUT_MS,
+    signal
+  });
+  return runningContext(Array.isArray(running?.models) ? running.models : [], rawId);
+}
 function contextLength(value) {
   return typeof value === "number" && Number.isSafeInteger(value) && value > 0 ? value : void 0;
 }
-async function enrichLmStudio(endpoint, entries, options) {
+async function enrichLmStudio(endpoint, entries, options, fetched) {
   const root = endpointRoot(endpoint.baseUrl);
-  const detail = await fetchJson(`${root}/api/v1/models`, endpoint, options);
+  const detail = fetched ?? await fetchJson(`${root}/api/v1/models`, endpoint, options);
   const byId = /* @__PURE__ */ new Map();
   if (Array.isArray(detail?.models)) {
     for (const model of detail.models) {
@@ -3231,12 +3284,17 @@ async function enrichLlamaCpp(endpoint, entries, options) {
     endpoint,
     options
   );
-  const nCtx = contextLength(props?.default_generation_settings?.n_ctx);
-  const known = nCtx !== void 0;
+  return llamaCppModels(
+    endpoint,
+    entries,
+    contextLength(props?.default_generation_settings?.n_ctx)
+  );
+}
+function llamaCppModels(endpoint, entries, nCtx) {
   return entries.map((entry) => ({
     ...genericModel(endpoint, entry),
-    contextWindow: known ? nCtx : FALLBACK_CONTEXT_WINDOW,
-    contextWindowKnown: known
+    contextWindow: nCtx ?? FALLBACK_CONTEXT_WINDOW,
+    contextWindowKnown: nCtx !== void 0
   }));
 }
 async function mapLimited(items, limit, fn) {
@@ -4285,6 +4343,7 @@ function createAutoUpdater(config) {
   kimiCodeBaseUrl,
   kimiCodingHeaders,
   localAuthStorageKey,
+  localModelShortName,
   log,
   loginAnthropic,
   loginGemini,
@@ -4292,6 +4351,7 @@ function createAutoUpdater(config) {
   loginOpenAI,
   loginXai,
   oauthStorageKey,
+  ollamaServedContext,
   openLog,
   parseLocalModelId,
   pendingClaudeCodeVersionRefresh,