cascade-ai 0.12.14 → 0.12.16

package/dist/cli.cjs

@@ -3,6 +3,7 @@
 var Anthropic = require('@anthropic-ai/sdk');
 var OpenAI = require('openai');
 var genai = require('@google/genai');
+var dns = require('dns');
 var ink = require('ink');
 var commander = require('commander');
 var React2 = require('react');
@@ -24,7 +25,7 @@ var glob = require('glob');
 var util = require('util');
 var simpleGit = require('simple-git');
 var PDFDocument = require('pdfkit');
-var dns = require('dns/promises');
+var dns2 = require('dns/promises');
 var net = require('net');
 var index_js = require('@modelcontextprotocol/sdk/client/index.js');
 var stdio_js = require('@modelcontextprotocol/sdk/client/stdio.js');
@@ -63,6 +64,7 @@ function _interopNamespace(e) {
 
 var Anthropic__default = /*#__PURE__*/_interopDefault(Anthropic);
 var OpenAI__default = /*#__PURE__*/_interopDefault(OpenAI);
+var dns__default = /*#__PURE__*/_interopDefault(dns);
 var React2__default = /*#__PURE__*/_interopDefault(React2);
 var chalk9__default = /*#__PURE__*/_interopDefault(chalk9);
 var dotenv__default = /*#__PURE__*/_interopDefault(dotenv);
@@ -75,7 +77,7 @@ var _ignoreModule__namespace = /*#__PURE__*/_interopNamespace(_ignoreModule);
 var Database__default = /*#__PURE__*/_interopDefault(Database);
 var EventEmitter__default = /*#__PURE__*/_interopDefault(EventEmitter);
 var PDFDocument__default = /*#__PURE__*/_interopDefault(PDFDocument);
-var dns__default = /*#__PURE__*/_interopDefault(dns);
+var dns2__default = /*#__PURE__*/_interopDefault(dns2);
 var net__default = /*#__PURE__*/_interopDefault(net);
 var Spinner__default = /*#__PURE__*/_interopDefault(Spinner);
 var ora__default = /*#__PURE__*/_interopDefault(ora);
@@ -101,7 +103,7 @@ var __export = (target, all) => {
 var CASCADE_VERSION, CASCADE_CONFIG_FILE, CASCADE_DB_FILE, CASCADE_DASHBOARD_SECRET_FILE, GLOBAL_CONFIG_DIR, GLOBAL_DB_FILE, GLOBAL_KEYSTORE_FILE, GLOBAL_RUNTIME_DB_FILE, DEFAULT_DASHBOARD_PORT, DEFAULT_CONTEXT_LIMIT, DEFAULT_AUTO_SUMMARIZE_AT, MODELS, T1_MODEL_PRIORITY, T2_MODEL_PRIORITY, T3_MODEL_PRIORITY, VISION_MODEL_PRIORITY, COMPLEXITY_T2_COUNT, THEME_NAMES, DEFAULT_THEME, OLLAMA_BASE_URL, LM_STUDIO_BASE_URL, AZURE_BASE_URL_TEMPLATE, TOOL_NAMES, DEFAULT_APPROVAL_REQUIRED;
 var init_constants = __esm({
   "src/constants.ts"() {
-    CASCADE_VERSION = "0.12.14";
+    CASCADE_VERSION = "0.12.16";
     CASCADE_CONFIG_FILE = ".cascade/config.json";
     CASCADE_DB_FILE = ".cascade/memory.db";
     CASCADE_DASHBOARD_SECRET_FILE = ".cascade/dashboard-secret";
@@ -1207,6 +1209,18 @@ var init_gemini = __esm({
     };
   }
 });
+function preferIpv4Host(url) {
+  if (!url) return url;
+  return url.replace(/^(https?:\/\/)localhost(?=[:/]|$)/i, "$1127.0.0.1");
+}
+var init_net = __esm({
+  "src/utils/net.ts"() {
+    try {
+      dns__default.default.setDefaultResultOrder("ipv4first");
+    } catch {
+    }
+  }
+});
 
 // src/providers/ollama.ts
 var ollama_exports = {};
@@ -1221,6 +1235,7 @@ var TOOL_CAPABLE_FAMILIES, OllamaProvider;
 var init_ollama = __esm({
   "src/providers/ollama.ts"() {
     init_constants();
+    init_net();
     init_base();
     TOOL_CAPABLE_FAMILIES = [
       "llama3.1",
@@ -1238,7 +1253,7 @@ var init_ollama = __esm({
       baseUrl;
       constructor(config, model) {
         super(config, model);
-        this.baseUrl = config.baseUrl ?? OLLAMA_BASE_URL;
+        this.baseUrl = preferIpv4Host(config.baseUrl ?? OLLAMA_BASE_URL);
       }
       async generate(options) {
         const chunks = [];
@@ -1435,37 +1450,61 @@ var OpenAICompatibleProvider;
 var init_openai_compatible = __esm({
   "src/providers/openai-compatible.ts"() {
     init_openai();
+    init_net();
     OpenAICompatibleProvider = class extends OpenAIProvider {
       constructor(config, model) {
         super(config, model);
         this.client = new OpenAI__default.default({
           apiKey: config.apiKey ?? "not-required",
-          baseURL: config.baseUrl
+          baseURL: preferIpv4Host(config.baseUrl)
         });
       }
+      modelsUrl() {
+        const base = (preferIpv4Host(this.config.baseUrl) ?? "").replace(/\/+$/, "");
+        return base + "/models";
+      }
+      authHeaders() {
+        const h = { Accept: "application/json" };
+        if (this.config.apiKey) h["Authorization"] = `Bearer ${this.config.apiKey}`;
+        return h;
+      }
+      // Discover models with a tolerant direct GET instead of the OpenAI SDK's typed
+      // `models.list()`. Local servers (llama.cpp / LM Studio / vLLM) return
+      // non-standard `/v1/models` payloads — an extra `models` array, filesystem
+      // path ids (`C:\…\model.gguf`) — that can make the SDK's typed pagination
+      // throw, which previously surfaced as a misleading "endpoint unreachable".
+      // A plain fetch + lenient parse is robust and reports the real HTTP error.
       async listModels() {
-        try {
-          const response = await this.client.models.list();
-          return response.data.map((m) => ({
-            id: m.id,
-            name: m.id,
-            provider: "openai-compatible",
-            contextWindow: 32e3,
-            isVisionCapable: false,
-            inputCostPer1kTokens: 0,
-            outputCostPer1kTokens: 0,
-            maxOutputTokens: 4e3,
-            supportsStreaming: true,
-            isLocal: false
-          }));
-        } catch {
-          return [this.model];
-        }
+        const res = await fetch(this.modelsUrl(), { headers: this.authHeaders() });
+        if (!res.ok) throw new Error(`models endpoint ${this.modelsUrl()} returned HTTP ${res.status}`);
+        const body = await res.json();
+        const raw = Array.isArray(body?.data) ? body.data : Array.isArray(body?.models) ? body.models : [];
+        const ids = raw.map((m) => {
+          if (m && typeof m === "object") {
+            const o = m;
+            const v = o["id"] ?? o["name"] ?? o["model"];
+            return typeof v === "string" ? v : void 0;
+          }
+          return typeof m === "string" ? m : void 0;
+        }).filter((x) => typeof x === "string" && x.length > 0);
+        if (ids.length === 0) return [this.model];
+        return ids.map((id) => ({
+          id,
+          name: id,
+          provider: "openai-compatible",
+          contextWindow: 32e3,
+          isVisionCapable: false,
+          inputCostPer1kTokens: 0,
+          outputCostPer1kTokens: 0,
+          maxOutputTokens: 4e3,
+          supportsStreaming: true,
+          isLocal: false
+        }));
       }
       async isAvailable() {
         try {
-          await this.client.models.list();
-          return true;
+          const res = await fetch(this.modelsUrl(), { headers: this.authHeaders() });
+          return res.ok;
         } catch {
           return false;
         }
@@ -4556,7 +4595,8 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter__default.default {
       for (const m of models) {
         this.selector.addDynamicModel(m);
       }
-    } catch {
+    } catch (err) {
+      console.warn("[router] OpenAI-compatible model discovery failed:", err instanceof Error ? err.message : err);
     }
   }
   ensureProvider(model, configs) {
@@ -8943,7 +8983,7 @@ async function assertPublicUrl(rawUrl) {
   }
   let addresses;
   try {
-    const records = await dns__default.default.lookup(host, { all: true });
+    const records = await dns2__default.default.lookup(host, { all: true });
     addresses = records.map((r) => r.address);
   } catch {
     throw new SsrfBlockedError(`Could not resolve host "${host}".`);