feat(providers): curated OpenRouter embeddings catalog + specialty merge in live discovery (#6976) (#6994)

* feat(providers): curated OpenRouter embeddings catalog + specialty merge in live discovery (#6976)

OpenRouter serves embeddings via a dedicated OpenAI-compatible
/api/v1/embeddings endpoint that is omitted from /v1/models, and the
embeddingRegistry entry for it was stale (3 legacy ids). Meanwhile
providerModelsConfig gives openrouter a live discovery config, so
buildApiDiscoveryResponse's success path returned only the live chat
catalog verbatim — the specialty (embeddings/rerank) static catalog was
only ever merged in on the no-config local_catalog fallback, so OpenRouter
embeddings never surfaced through model discovery.

Refreshed the curated openrouter embeddingRegistry lineup (ids verified
against https://openrouter.ai/docs/api/reference/embeddings and the
collections page) and added a scoped, additive merge
(mergeSpecialtyCatalogIntoLiveModels, allowlisted to openrouter) that folds
embeddings/rerank entries from getStaticModelsForProvider() into the live
discovery response, deduped by id. Scoped as an allowlist rather than a
blanket merge because some providers (e.g. Gemini) already return
embedding models directly from their live /v1/models endpoint, where a
blind merge would risk stale/duplicate entries.

* test(providers): type the models discovery payload instead of any (#6976)

no-explicit-any is an error under tests/ (#6218), so the 4 `any`
usages in the new discovery assertions failed the max-warnings-0
lint gate. Replace them with an explicit ModelsResponseBody shape —
type-only change, all 13 assertions unchanged and still passing.

* test(providers): type the openrouter merge assertion callback (#6976)

The new #6976 assertion added a 56th explicit `any` to this file,
one over the 55 frozen in config/quality/eslint-suppressions.json,
tripping the max-warnings-0 lint gate. Type the callback param
instead of raising the frozen count — the debt ratchet only decreases.
All 59 tests still pass.

changelog.d/features/6976-openrouter-embeddings.md

@@ -0,0 +1 @@
+- **feat(providers):** refresh the curated OpenRouter embeddings catalog (`open-sse/config/embeddingRegistry.ts`) with the current lineup — `openai/text-embedding-3-small`/`-large`, `qwen/qwen3-embedding-8b`/`-4b`, `baai/bge-m3`, `mistralai/mistral-embed-2312`, `google/gemini-embedding-001` — and fold curated embedding/rerank entries into OpenRouter's live model-discovery response (`src/app/api/providers/[id]/models/route.ts`), additively and deduped by id, so they no longer only appear on the no-config `local_catalog` fallback. OpenRouter serves embeddings via a dedicated `/api/v1/embeddings` endpoint (omitted from `/v1/models`), so the live-discovery success path previously returned chat models only ([#6976](https://github.com/diegosouzapw/OmniRoute/issues/6976)). Regression guard: `tests/unit/openrouter-embeddings-catalog-6976.test.ts`.

open-sse/config/embeddingRegistry.ts

@@ -187,6 +187,12 @@ export const EMBEDDING_PROVIDERS: Record<string, EmbeddingProvider> = {
     ],
   },
 
+  // #6976 — OpenRouter serves embeddings via a dedicated OpenAI-compatible
+  // /api/v1/embeddings endpoint (omitted from /v1/models, so this catalog is
+  // curated rather than live-discovered). Ids verified against the API
+  // reference (not the display-name collections page) at refresh time:
+  // https://openrouter.ai/docs/api/reference/embeddings and
+  // https://openrouter.ai/collections/embedding-models
   openrouter: {
     id: "openrouter",
     baseUrl: "https://openrouter.ai/api/v1/embeddings",
@@ -204,9 +210,29 @@ export const EMBEDDING_PROVIDERS: Record<string, EmbeddingProvider> = {
         dimensions: 3072,
       },
       {
-        id: "openai/text-embedding-ada-002",
-        name: "Text Embedding Ada 002 (OpenRouter)",
-        dimensions: 1536,
+        id: "qwen/qwen3-embedding-8b",
+        name: "Qwen3 Embedding 8B (OpenRouter)",
+        dimensions: 4096,
+      },
+      {
+        id: "qwen/qwen3-embedding-4b",
+        name: "Qwen3 Embedding 4B (OpenRouter)",
+        dimensions: 2560,
+      },
+      {
+        id: "baai/bge-m3",
+        name: "BGE-M3 (OpenRouter)",
+        dimensions: 1024,
+      },
+      {
+        id: "mistralai/mistral-embed-2312",
+        name: "Mistral Embed (OpenRouter)",
+        dimensions: 1024,
+      },
+      {
+        id: "google/gemini-embedding-001",
+        name: "Gemini Embedding 001 (OpenRouter)",
+        dimensions: 768,
       },
     ],
   },

src/app/api/providers/[id]/models/discovery/helpers.ts

@@ -1,5 +1,5 @@
 import { isSelfHostedChatProvider } from "@/shared/constants/providers";
-import type { LocalCatalogModel } from "@/lib/providers/staticModels";
+import { getStaticModelsForProvider, type LocalCatalogModel } from "@/lib/providers/staticModels";
 
 export type JsonRecord = Record<string, unknown>;
 
@@ -51,6 +51,36 @@ export function mergeLocalCatalogModels<T extends LocalCatalogModel, U extends L
   ];
 }
 
+// #6976 — providers whose live /v1/models endpoint is known to serve ONLY
+// chat models (verified: OpenRouter's catalog is chat-only — embeddings live
+// on a separate /api/v1/embeddings endpoint per
+// https://openrouter.ai/docs/api/reference/embeddings) never surface their
+// embedding/rerank specialty catalog once they have a live discovery config,
+// because the specialty catalog is otherwise only folded in on the no-config
+// local_catalog fallback. Deliberately an allowlist, not every provider with
+// an embeddingRegistry/rerankRegistry entry: some providers' live /v1/models
+// response legitimately DOES include embedding ids already (e.g. Gemini's
+// models.list mixes generateContent and embedContent models in one response —
+// see provider-models-route.test.ts pagination coverage), so blind-merging
+// the curated catalog there would risk stale-duplicate/conflicting entries.
+const LIVE_DISCOVERY_SPECIALTY_MERGE_PROVIDERS = new Set<string>(["openrouter"]);
+
+// Fold the embeddings/rerank subset of the static catalog into a successful
+// live-discovery response, additively and deduped by id, without touching
+// chat/image/video/audio entries — scoped to
+// LIVE_DISCOVERY_SPECIALTY_MERGE_PROVIDERS above.
+export function mergeSpecialtyCatalogIntoLiveModels<T extends { id: string }>(
+  liveModels: T[],
+  provider: string
+): Array<T | LocalCatalogModel> {
+  if (!LIVE_DISCOVERY_SPECIALTY_MERGE_PROVIDERS.has(provider)) return liveModels;
+  const specialty = (getStaticModelsForProvider(provider) || []).filter(
+    (model) => model.apiFormat === "embeddings" || model.apiFormat === "rerank"
+  );
+  if (specialty.length === 0) return liveModels;
+  return mergeLocalCatalogModels(liveModels, specialty);
+}
+
 export function buildOptionalBearerHeaders(
   token: string | null | undefined
 ): Record<string, string> {

src/app/api/providers/[id]/models/route.ts

@@ -85,6 +85,7 @@ import {
   getAzureOpenAIApiVersion,
   isLocalOpenAIStyleProvider,
   mergeLocalCatalogModels,
+  mergeSpecialtyCatalogIntoLiveModels,
   buildOptionalBearerHeaders,
   buildNamedOpenAiStyleHeaders,
 } from "./discovery/helpers";
@@ -408,10 +409,15 @@ export async function GET(
     ) => {
       const discoveredModels = await persistDiscoveredModels(provider, connectionId, models);
       if (discoveredModels.length > 0) {
+        // #6976 — merge curated embedding/rerank specialty entries (e.g.
+        // OpenRouter's embeddingRegistry catalog) into the live-discovery
+        // response; the live /v1/models endpoint only lists chat models, and
+        // the specialty catalog otherwise only reached local_catalog fallback.
+        const mergedModels = mergeSpecialtyCatalogIntoLiveModels(models, provider);
         return buildResponse({
           provider,
           connectionId,
-          models,
+          models: mergedModels,
           source: "api",
           ...(warning ? { warning } : {}),
           ...extraPayload,

tests/unit/openrouter-embeddings-catalog-6976.test.ts

@@ -0,0 +1,127 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-openrouter-embeddings-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const core = await import("../../src/lib/db/core.ts");
+const providersDb = await import("../../src/lib/db/providers.ts");
+const providerModelsRoute = await import("../../src/app/api/providers/[id]/models/route.ts");
+const embeddingRegistry = await import("../../open-sse/config/embeddingRegistry.ts");
+const staticModels = await import("../../src/lib/providers/staticModels.ts");
+
+const originalFetch = globalThis.fetch;
+
+/** Shape of the /api/providers/[id]/models discovery payload asserted below. */
+type DiscoveredModel = { id: string; name?: string };
+type ModelsResponseBody = { source: string; models: DiscoveredModel[] };
+
+async function resetStorage() {
+  globalThis.fetch = originalFetch;
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+  fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
+}
+
+async function seedConnection(provider: string, overrides: Record<string, unknown> = {}) {
+  return providersDb.createProviderConnection({
+    provider,
+    authType: "apikey",
+    name: `${provider}-${Math.random().toString(16).slice(2, 8)}`,
+    apiKey: "or-test-key",
+    isActive: true,
+    testStatus: "active",
+    providerSpecificData: {},
+    ...overrides,
+  });
+}
+
+async function callRoute(connectionId: string) {
+  return providerModelsRoute.GET(
+    new Request(`http://localhost/api/providers/${connectionId}/models`),
+    { params: { id: connectionId } }
+  );
+}
+
+test.beforeEach(async () => {
+  await resetStorage();
+});
+
+test.after(async () => {
+  globalThis.fetch = originalFetch;
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+test("embeddingRegistry curated openrouter catalog carries the refreshed lineup with dimensions (#6976)", () => {
+  const config = embeddingRegistry.getEmbeddingProvider("openrouter");
+  assert.ok(config, "openrouter embedding provider config must exist");
+  const ids = config!.models.map((m) => m.id);
+  for (const expected of [
+    "openai/text-embedding-3-small",
+    "openai/text-embedding-3-large",
+    "qwen/qwen3-embedding-8b",
+    "qwen/qwen3-embedding-4b",
+    "baai/bge-m3",
+    "mistralai/mistral-embed-2312",
+    "google/gemini-embedding-001",
+  ]) {
+    assert.ok(ids.includes(expected), `expected curated id ${expected}; got ${ids.join(", ")}`);
+    const dim = config!.models.find((m) => m.id === expected)?.dimensions;
+    assert.equal(typeof dim, "number", `${expected} must carry a dimensions value`);
+  }
+});
+
+test("getStaticModelsForProvider(openrouter) folds the curated embeddings into the specialty catalog (#6976)", () => {
+  const specialty = staticModels.getStaticModelsForProvider("openrouter");
+  assert.ok(specialty && specialty.length > 0);
+  const embeddingEntry = specialty!.find((m) => m.id === "baai/bge-m3");
+  assert.ok(embeddingEntry, "curated bge-m3 entry must be present in the static catalog");
+  assert.equal(embeddingEntry!.apiFormat, "embeddings");
+});
+
+test("live discovery merges curated embeddings into the response even when /v1/models returns none (#6976)", async () => {
+  const connection = await seedConnection("openrouter");
+  globalThis.fetch = async () =>
+    Response.json({
+      data: [{ id: "anthropic/claude-sonnet-5", name: "Claude Sonnet 5" }],
+    });
+
+  const response = await callRoute(connection.id);
+  const body = (await response.json()) as ModelsResponseBody;
+
+  assert.equal(response.status, 200);
+  assert.equal(body.source, "api");
+  const ids = body.models.map((m) => m.id);
+  // Chat model from the live /v1/models fetch is preserved.
+  assert.ok(ids.includes("anthropic/claude-sonnet-5"));
+  // RED before the Step 2 merge: the live discovery success path (buildApiDiscoveryResponse)
+  // returned `models` verbatim, so curated embeddings never appeared here — only on the
+  // no-config local_catalog fallback. GREEN after: curated embeddings/rerank entries from
+  // getStaticModelsForProvider() are folded in additively.
+  assert.ok(
+    ids.includes("baai/bge-m3"),
+    `curated embedding baai/bge-m3 should be merged into live discovery; got: ${ids.join(", ")}`
+  );
+  assert.ok(ids.includes("openai/text-embedding-3-small"));
+});
+
+test("live discovery dedups: a model already present in the live catalog is not duplicated (#6976)", async () => {
+  const connection = await seedConnection("openrouter");
+  globalThis.fetch = async () =>
+    Response.json({
+      // OpenRouter's live /v1/models never actually lists embedding ids today, but
+      // this proves the merge is a dedup-by-id union, not a blind concat.
+      data: [{ id: "baai/bge-m3", name: "BGE-M3 (live)" }],
+    });
+
+  const response = await callRoute(connection.id);
+  const body = (await response.json()) as ModelsResponseBody;
+
+  const bgeEntries = body.models.filter((m) => m.id === "baai/bge-m3");
+  assert.equal(bgeEntries.length, 1, "baai/bge-m3 must appear exactly once");
+  assert.equal(bgeEntries[0].name, "BGE-M3 (live)", "live entry wins over the curated duplicate");
+});

tests/unit/provider-models-route.test.ts

@@ -572,7 +572,18 @@ test("provider models route prefers the remote OpenRouter /models API over stati
   assert.equal(response.status, 200);
   assert.equal(body.source, "api");
   assert.deepEqual(seenUrls, ["https://openrouter.ai/api/v1/models"]);
-  assert.deepEqual(body.models, [{ id: "openai/gpt-4.1", name: "GPT-4.1 via OpenRouter" }]);
+  // #6976 — OpenRouter's live /v1/models never lists embeddings/rerank (they live
+  // on dedicated endpoints), so the curated specialty catalog is folded into the
+  // live-discovery response additively; static IMAGE models stay excluded
+  // (hasChatRegistry is true for openrouter — see staticModels.ts).
+  const ids = body.models.map((m: { id: string }) => m.id);
+  assert.ok(ids.includes("openai/gpt-4.1"), "live-fetched chat model is preserved");
+  assert.ok(ids.includes("baai/bge-m3"), "curated embedding is merged in");
+  assert.ok(ids.includes("cohere/rerank-v3.5"), "curated rerank is merged in");
+  assert.ok(
+    !ids.some((id: string) => id.includes("gpt-5.4-image")),
+    "static image models stay excluded from the chat+specialty catalog"
+  );
 });
 
 test("provider models route returns the local catalog for embedding and rerank providers", async () => {