feat(sdk): enumerate provider models of every modality in the model catalog (#4676)

* feat(sdk): enumerate provider models of every modality in the model catalog

The SDK model catalog (/api/sdk/v1/models) built its list from getAllModels,
which only enumerates language models from configured providers. Image, TTS,
music, ASR, video, and embedding models appeared only when they happened to be
in RECOMMENDED_MODELS — an SDK client asking for image_model compatibility got
exactly one entry (GPT Image 2) while the web editor's picker, which goes
through availableForKind, showed the full provider lists.

Add collectProviderCatalogModels: one pass per configured provider calling
each getAvailable*Models once, no task filtering, so text_to_image and
image_to_image capable models are both included. Per-list failures degrade to
an empty list instead of dropping the provider.

The catalog service merges these into the existing gathering behind a 60s
per-user TTL cache. Only the remote provider enumeration is cached — local
download state (HF cache scan, download manager) stays fresh on every call, so
a finished download still flips to ready_local immediately. Worker-scoped
catalogs are unchanged and never enumerate providers.

getAllModels and availableForKind are untouched; existing consumers keep
their exact behavior.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Author: 3 people

packages/websocket/src/sdk/sdk-model-catalog-service.ts

@@ -9,6 +9,7 @@ import {
   type SdkV1ModelCatalogQuery
 } from "@nodetool-ai/protocol/api-schemas/sdk-models-v1.js";
 import {
+  collectProviderCatalogModels,
   getAllModels,
   getAvailableProviderIds
 } from "../trpc/routers/models.js";
@@ -161,6 +162,43 @@ export function projectSdkModelCatalog(
   });
 }
 
+// The per-provider model enumeration fans out to remote provider APIs and is
+// the slow part of a catalog request. Its result is pure remote data (no
+// local download state), so a short TTL cache is safe and keeps repeated SDK
+// catalog polls fast.
+const PROVIDER_CATALOG_TTL_MS = 60_000;
+const providerCatalogCache = new Map<
+  string,
+  { at: number; models: readonly UnifiedModel[] }
+>();
+
+function pruneProviderCatalogCache(now: number): void {
+  for (const [key, value] of providerCatalogCache) {
+    if (now - value.at >= PROVIDER_CATALOG_TTL_MS) {
+      providerCatalogCache.delete(key);
+    }
+  }
+}
+
+async function getCachedProviderCatalogModels(
+  userId: string
+): Promise<readonly UnifiedModel[]> {
+  const now = Date.now();
+  pruneProviderCatalogCache(now);
+
+  const cached = providerCatalogCache.get(userId);
+  if (cached && now - cached.at < PROVIDER_CATALOG_TTL_MS) {
+    return cached.models;
+  }
+  const models = await collectProviderCatalogModels(userId);
+  providerCatalogCache.set(userId, { at: now, models });
+  return models;
+}
+
+export function clearProviderCatalogCache(): void {
+  providerCatalogCache.clear();
+}
+
 function dedupeCatalogModels(models: readonly UnifiedModel[]): UnifiedModel[] {
   const byKey = new Map<string, UnifiedModel>();
   for (const model of models) {
@@ -176,8 +214,12 @@ export async function getSdkV1ModelCatalog(args: {
   query: SdkV1ModelCatalogQuery;
   recommendedModels?: readonly UnifiedModel[];
   getWorkerModels?: () => Promise<readonly UnifiedModel[]>;
+  getProviderCatalogModels?: (
+    userId: string
+  ) => Promise<readonly UnifiedModel[]>;
 }): Promise<SdkV1ModelCatalog> {
   let availableModels: readonly UnifiedModel[];
+  let providerCatalogModels: readonly UnifiedModel[];
   let providerIds: readonly string[];
   if (args.query.scope === "worker") {
     if (!args.getWorkerModels) {
@@ -186,10 +228,14 @@ export async function getSdkV1ModelCatalog(args: {
       );
     }
     availableModels = await args.getWorkerModels();
+    providerCatalogModels = [];
     providerIds = [];
   } else {
-    [availableModels, providerIds] = await Promise.all([
+    [availableModels, providerCatalogModels, providerIds] = await Promise.all([
       getAllModels(args.userId),
+      (args.getProviderCatalogModels ?? getCachedProviderCatalogModels)(
+        args.userId
+      ),
       getAvailableProviderIds(args.userId)
     ]);
   }
@@ -199,6 +245,7 @@ export async function getSdkV1ModelCatalog(args: {
   ];
   const models = dedupeCatalogModels([
     ...availableModels,
+    ...providerCatalogModels,
     ...recommendedModels
   ]);
   const manager =

packages/websocket/src/trpc/routers/models.ts

@@ -928,6 +928,53 @@ async function collectProviderModelsForKind(
   return out;
 }
 
+/**
+ * Every remote model the user's configured providers can enumerate, across all
+ * non-language-model types (embeddings, images, TTS, music, ASR, video), for the
+ * SDK model catalog. Unlike `collectProviderModelsForKind` this makes one pass
+ * per provider (each `getAvailable*Models` called once, no task filtering) so
+ * text_to_image and image_to_image capable models are both included. Language
+ * models are not collected here — `getAllModels` already enumerates them.
+ */
+export async function collectProviderCatalogModels(
+  userId: string
+): Promise<UnifiedModel[]> {
+  const providerIds = await getAvailableProviderIds(userId);
+  const perProvider = await Promise.all(
+    providerIds.map((providerId) =>
+      safeProviderCall(
+        "catalogModels",
+        { provider: providerId, userId },
+        async () => {
+          const instance = await instantiateProvider(providerId, userId);
+          if (!instance) return [];
+          const collect = (
+            fetchModels: () => Promise<Parameters<typeof toUnifiedModel>[0][]>,
+            type: string
+          ) =>
+            safeProviderCall(
+              `catalogModels:${type}`,
+              { provider: providerId, userId },
+              async () => (await fetchModels()).map((m) => toUnifiedModel(m, type)),
+              [] as UnifiedModel[]
+            );
+          const lists = await Promise.all([
+            collect(() => instance.getAvailableImageModels(), "image_model"),
+            collect(() => instance.getAvailableEmbeddingModels(), "embedding_model"),
+            collect(() => instance.getAvailableTTSModels(), "tts_model"),
+            collect(() => instance.getAvailableMusicModels(), "music_model"),
+            collect(() => instance.getAvailableASRModels(), "asr_model"),
+            collect(() => instance.getAvailableVideoModels(), "video_model")
+          ]);
+          return lists.flat();
+        },
+        [] as UnifiedModel[]
+      )
+    )
+  );
+  return perProvider.flat();
+}
+
 function curatedForKind(kind: ModelSearchKind): UnifiedModel[] {
   const modality = KIND_TO_MODALITY[kind];
   // For text_generation/embedding/image/video, RECOMMENDED_MODELS entries are

packages/websocket/tests/sdk-model-catalog-provider-models.test.ts

@@ -0,0 +1,115 @@
+import { beforeEach, describe, expect, it, vi } from "vitest";
+import type { UnifiedModel } from "@nodetool-ai/protocol";
+
+const getAllModels = vi.fn<(userId: string) => Promise<UnifiedModel[]>>();
+const getAvailableProviderIds = vi.fn<(userId: string) => Promise<string[]>>();
+const collectProviderCatalogModels =
+  vi.fn<(userId: string) => Promise<UnifiedModel[]>>();
+
+vi.mock("../src/trpc/routers/models.js", () => ({
+  getAllModels: (userId: string) => getAllModels(userId),
+  getAvailableProviderIds: (userId: string) => getAvailableProviderIds(userId),
+  collectProviderCatalogModels: (userId: string) =>
+    collectProviderCatalogModels(userId)
+}));
+
+vi.mock("@nodetool-ai/huggingface", () => ({
+  getExistingDownloadManager: () => null
+}));
+
+import {
+  clearProviderCatalogCache,
+  getSdkV1ModelCatalog
+} from "../src/sdk/sdk-model-catalog-service.js";
+
+const query = { scope: "local" as const, limit: 200 };
+
+const languageModel: UnifiedModel = {
+  id: "gpt-test",
+  name: "GPT Test",
+  type: "language_model",
+  provider: "openai"
+};
+
+const falImageModel: UnifiedModel = {
+  id: "fal-ai/flux/schnell",
+  name: "FLUX.1 Schnell",
+  type: "image_model",
+  provider: "fal_ai"
+};
+
+const openaiImageModel: UnifiedModel = {
+  id: "gpt-image-2",
+  name: "GPT Image 2",
+  type: "image_model",
+  provider: "openai"
+};
+
+beforeEach(() => {
+  clearProviderCatalogCache();
+  getAllModels.mockReset().mockResolvedValue([languageModel]);
+  getAvailableProviderIds
+    .mockReset()
+    .mockResolvedValue(["openai", "fal_ai"]);
+  collectProviderCatalogModels
+    .mockReset()
+    .mockResolvedValue([falImageModel, openaiImageModel]);
+});
+
+describe("SDK model catalog provider models", () => {
+  it("includes provider-enumerated non-language models as ready_remote", async () => {
+    const catalog = await getSdkV1ModelCatalog({ userId: "alice", query });
+
+    const flux = catalog.entries.find(
+      (entry) => entry.id === "fal-ai/flux/schnell"
+    );
+    expect(flux).toMatchObject({
+      compatibility: "image_model",
+      availability: "ready_remote",
+      provider: "fal_ai"
+    });
+  });
+
+  it("dedupes provider-enumerated models against the recommended list", async () => {
+    const catalog = await getSdkV1ModelCatalog({ userId: "alice", query });
+
+    // gpt-image-2 exists both in RECOMMENDED_MODELS and in the provider
+    // enumeration; the catalog must carry it once per (type, provider, id).
+    const gptImage = catalog.entries.filter(
+      (entry) => entry.id === "gpt-image-2" && entry.provider === "openai"
+    );
+    expect(gptImage).toHaveLength(1);
+  });
+
+  it("caches the provider enumeration per user", async () => {
+    await getSdkV1ModelCatalog({ userId: "alice", query });
+    await getSdkV1ModelCatalog({ userId: "alice", query });
+    expect(collectProviderCatalogModels).toHaveBeenCalledTimes(1);
+
+    await getSdkV1ModelCatalog({ userId: "bob", query });
+    expect(collectProviderCatalogModels).toHaveBeenCalledTimes(2);
+    expect(collectProviderCatalogModels).toHaveBeenLastCalledWith("bob");
+  });
+
+  it("prefers an injected provider-catalog fetcher over the cache", async () => {
+    const injected = vi.fn().mockResolvedValue([falImageModel]);
+    await getSdkV1ModelCatalog({
+      userId: "alice",
+      query,
+      getProviderCatalogModels: injected
+    });
+    expect(injected).toHaveBeenCalledWith("alice");
+    expect(collectProviderCatalogModels).not.toHaveBeenCalled();
+  });
+
+  it("never enumerates providers for worker-scoped catalogs", async () => {
+    const catalog = await getSdkV1ModelCatalog({
+      userId: "alice",
+      query: { scope: "worker" as const, limit: 200 },
+      getWorkerModels: async () => [languageModel]
+    });
+    expect(collectProviderCatalogModels).not.toHaveBeenCalled();
+    expect(getAllModels).not.toHaveBeenCalled();
+    expect(catalog.scope).toBe("worker");
+  });
+});