cascade-ai 0.43.0 → 0.46.0

package/dist/index.d.cts

@@ -831,6 +831,14 @@ declare class ModelSelector {
      * Absent for a provider ⇒ no validation ran ⇒ behave exactly as before.
      */
     private validatedIds;
+    /**
+     * Normalized ids that were added via discovery/registration but aren't in the
+     * bundled catalog — i.e. genuinely new models the provider reported (a Gemini
+     * flash released after the catalog was last hand-updated, an Ollama tag, etc.).
+     * getCandidatesForTier lets these compete in AUTO ranking so a newer,
+     * better-value model isn't invisible until someone edits constants.ts.
+     */
+    private discovered;
     constructor(availableProviders: Set<ProviderType>);
     addDynamicModel(model: ModelInfo): void;
     /** Record the ids a provider's API actually serves (discovery). Empty ⇒ ignored. */
@@ -1344,6 +1352,13 @@ declare class CascadeRouter extends EventEmitter {
     private providers;
     private stats;
     private tierModels;
+    /**
+     * Tiers the user pinned to a specific model in config (`models.tX` !== 'auto').
+     * An explicit pin always wins over Cascade Auto — selectModelForSubtask never
+     * re-selects a pinned tier per subtask. (tierModels is also filled with
+     * auto-picked defaults, so it alone can't distinguish "pinned" from "default".)
+     */
+    private explicitTierModels;
     private config;
     private sessionCostUsd;
     private runTokens;

package/dist/index.d.ts

@@ -831,6 +831,14 @@ declare class ModelSelector {
      * Absent for a provider ⇒ no validation ran ⇒ behave exactly as before.
      */
     private validatedIds;
+    /**
+     * Normalized ids that were added via discovery/registration but aren't in the
+     * bundled catalog — i.e. genuinely new models the provider reported (a Gemini
+     * flash released after the catalog was last hand-updated, an Ollama tag, etc.).
+     * getCandidatesForTier lets these compete in AUTO ranking so a newer,
+     * better-value model isn't invisible until someone edits constants.ts.
+     */
+    private discovered;
     constructor(availableProviders: Set<ProviderType>);
     addDynamicModel(model: ModelInfo): void;
     /** Record the ids a provider's API actually serves (discovery). Empty ⇒ ignored. */
@@ -1344,6 +1352,13 @@ declare class CascadeRouter extends EventEmitter {
     private providers;
     private stats;
     private tierModels;
+    /**
+     * Tiers the user pinned to a specific model in config (`models.tX` !== 'auto').
+     * An explicit pin always wins over Cascade Auto — selectModelForSubtask never
+     * re-selects a pinned tier per subtask. (tierModels is also filled with
+     * auto-picked defaults, so it alone can't distinguish "pinned" from "default".)
+     */
+    private explicitTierModels;
     private config;
     private sessionCostUsd;
     private runTokens;

package/dist/index.js

@@ -193,7 +193,7 @@ var init_audit_logger = __esm({
 });
 
 // src/constants.ts
-var CASCADE_VERSION = "0.43.0";
+var CASCADE_VERSION = "0.46.0";
 var CASCADE_CONFIG_DIR = ".cascade";
 var CASCADE_MD_FILE = "CASCADE.md";
 var CASCADE_IGNORE_FILE = ".cascadeignore";
@@ -1797,12 +1797,21 @@ var ModelSelector = class {
    * Absent for a provider ⇒ no validation ran ⇒ behave exactly as before.
    */
   validatedIds = /* @__PURE__ */ new Map();
+  /**
+   * Normalized ids that were added via discovery/registration but aren't in the
+   * bundled catalog — i.e. genuinely new models the provider reported (a Gemini
+   * flash released after the catalog was last hand-updated, an Ollama tag, etc.).
+   * getCandidatesForTier lets these compete in AUTO ranking so a newer,
+   * better-value model isn't invisible until someone edits constants.ts.
+   */
+  discovered = /* @__PURE__ */ new Set();
   constructor(availableProviders) {
     this.availableProviders = availableProviders;
     this.availableModels = new Map(Object.entries(MODELS));
   }
   addDynamicModel(model) {
     this.availableModels.set(model.id, model);
+    if (!(model.id in MODELS)) this.discovered.add(normalizeModelId(model.id));
   }
   /** Record the ids a provider's API actually serves (discovery). Empty ⇒ ignored. */
   setValidatedModels(provider, ids) {
@@ -1910,6 +1919,19 @@ var ModelSelector = class {
         candidates.push(model);
       }
     }
+    if (this.discovered.size) {
+      const tierProviders = /* @__PURE__ */ new Set();
+      for (const key of priority) {
+        const m = MODELS[key];
+        if (m) tierProviders.add(m.provider);
+      }
+      const have = new Set(candidates.map((c) => c.id));
+      for (const model of this.availableModels.values()) {
+        if (have.has(model.id)) continue;
+        if (!this.discovered.has(normalizeModelId(model.id))) continue;
+        if (tierProviders.has(model.provider) && this.isUsable(model)) candidates.push(model);
+      }
+    }
     const localOnly = this.availableProviders.size > 0 && Array.from(this.availableProviders).every((p) => p === "ollama");
     if (localOnly) {
       for (const model of this.availableModels.values()) {
@@ -2626,6 +2648,14 @@ var FAMILY_MATCHERS = [
   [/gemini-?1\.5-pro/i, "gemini-1.5-pro"],
   [/gemini-?2\.0-flash-lite/i, "gemini-flash-lite"],
   [/gemini-?2\.0-flash/i, "gemini-2.0-flash"],
+  // Generic Gemini fallbacks — a model released after this table was written
+  // (e.g. a Gemini 3.x flash discovered live from the provider) still gets a
+  // sensible score by its class instead of a neutral 0.5, so it can compete in
+  // AUTO ranking. Ordered after the specific rows above; most-specific first.
+  [/gemini.*flash.?lite/i, "gemini-flash-lite"],
+  [/gemini.*pro/i, "gemini-2.5-pro"],
+  [/gemini.*flash/i, "gemini-2.5-flash"],
+  [/gemini/i, "gemini-2.5-flash"],
   [/codellama|code-llama|starcoder|stable-code/i, "codellama"],
   [/deepseek/i, "deepseek"],
   [/qwen/i, "qwen"],
@@ -2717,6 +2747,13 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter {
     costByFeature: {}
   };
   tierModels = /* @__PURE__ */ new Map();
+  /**
+   * Tiers the user pinned to a specific model in config (`models.tX` !== 'auto').
+   * An explicit pin always wins over Cascade Auto — selectModelForSubtask never
+   * re-selects a pinned tier per subtask. (tierModels is also filled with
+   * auto-picked defaults, so it alone can't distinguish "pinned" from "default".)
+   */
+  explicitTierModels = /* @__PURE__ */ new Set();
   config;
   sessionCostUsd = 0;
   // Per-run accounting for the hard per-task cap. Reset by beginRun() at the
@@ -2794,6 +2831,7 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter {
         );
       }
       this.tierModels.set(tier, model);
+      this.explicitTierModels.add(tier);
       this.ensureProvider(model, config.providers);
     }
     const azureRanked = [...azureDeployments].sort((a, b) => inferModelCapability(b.id) - inferModelCapability(a.id));
@@ -3234,6 +3272,7 @@ var CascadeRouter = class _CascadeRouter extends EventEmitter {
    * Pure heuristic — no extra LLM call.
    */
   async selectModelForSubtask(tier, text, opts) {
+    if (this.explicitTierModels.has(tier)) return this.tierModels.get(tier) ?? null;
     if (!this.config?.cascadeAuto || !this.taskAnalyzer || !text.trim()) return null;
     try {
       return await this.taskAnalyzer.selectModel(text, tier, this.selector, opts);