fix(embeddings): batch OpenRouter requests

Author: TheodoreSpeaks

apps/sim/app/api/providers/openrouter/embeddings/models/route.test.ts

@@ -38,9 +38,9 @@ describe('GET /api/providers/openrouter/embeddings/models', () => {
       statusText: 'OK',
       json: async () => ({
         data: [
-          { id: 'qwen/qwen3-embedding-8b' },
-          { id: 'openai/text-embedding-3-small' },
-          { id: 'qwen/qwen3-embedding-8b' },
+          { id: 'qwen/qwen3-embedding-8b', context_length: 32768 },
+          { id: 'openai/text-embedding-3-small', context_length: 8192 },
+          { id: 'qwen/qwen3-embedding-8b', context_length: 32768 },
         ],
       }),
     })

apps/sim/app/api/providers/openrouter/embeddings/models/route.ts

@@ -1,11 +1,8 @@
 import { createLogger } from '@sim/logger'
 import { type NextRequest, NextResponse } from 'next/server'
-import {
-  openRouterEmbeddingModelsUpstreamResponseSchema,
-  providerModelsResponseSchema,
-} from '@/lib/api/contracts/providers'
+import { providerModelsResponseSchema } from '@/lib/api/contracts/providers'
 import { withRouteHandler } from '@/lib/core/utils/with-route-handler'
-import { toOpenRouterEmbeddingModelId } from '@/lib/embeddings/openrouter-models'
+import { fetchOpenRouterEmbeddingModelCatalog } from '@/lib/embeddings/openrouter-model-catalog.server'
 import { filterBlacklistedModels, isProviderBlacklisted } from '@/providers/utils'
 
 const logger = createLogger('OpenRouterEmbeddingModelsAPI')
@@ -16,20 +13,7 @@ export const GET = withRouteHandler(async (_request: NextRequest) => {
     return NextResponse.json({ models: [] })
   }
 
-  const response = await fetch('https://openrouter.ai/api/v1/embeddings/models', {
-    headers: { 'Content-Type': 'application/json' },
-    next: { revalidate: 300 },
-  })
-  if (!response.ok) {
-    throw new Error(
-      `Failed to fetch OpenRouter embedding models: ${response.status} ${response.statusText}`
-    )
-  }
-
-  const data = openRouterEmbeddingModelsUpstreamResponseSchema.parse(await response.json())
-  const uniqueModels = Array.from(
-    new Set(data.data.map((model) => toOpenRouterEmbeddingModelId(model.id)))
-  )
+  const uniqueModels = (await fetchOpenRouterEmbeddingModelCatalog()).map((model) => model.id)
   const models = filterBlacklistedModels(uniqueModels)
 
   logger.info('Successfully fetched OpenRouter embedding models', {

apps/sim/app/api/tools/embeddings/route.test.ts

@@ -4,9 +4,22 @@
 import { createMockRequest, hybridAuthMockFns } from '@sim/testing'
 import { beforeEach, describe, expect, it, vi } from 'vitest'
 
-const { mockEmbed, mockEmbedOpenRouter } = vi.hoisted(() => ({
-  mockEmbed: vi.fn(),
-  mockEmbedOpenRouter: vi.fn(),
+const { mockEmbed, mockEmbedOpenRouter, mockGetOpenRouterEmbeddingModelMetadata } = vi.hoisted(
+  () => ({
+    mockEmbed: vi.fn(),
+    mockEmbedOpenRouter: vi.fn(),
+    mockGetOpenRouterEmbeddingModelMetadata: vi.fn(),
+  })
+)
+
+vi.mock('@/lib/embeddings/openrouter-model-catalog.server', () => ({
+  getOpenRouterEmbeddingModelMetadata: mockGetOpenRouterEmbeddingModelMetadata,
+  OpenRouterEmbeddingModelNotFoundError: class OpenRouterEmbeddingModelNotFoundError extends Error {
+    constructor(model: string) {
+      super(`Unsupported OpenRouter embedding model: ${model}`)
+      this.name = 'OpenRouterEmbeddingModelNotFoundError'
+    }
+  },
 }))
 
 vi.mock('@/lib/embeddings', async () => {
@@ -21,6 +34,7 @@ vi.mock('@/lib/embeddings', async () => {
   }
 })
 
+import { OpenRouterEmbeddingModelNotFoundError } from '@/lib/embeddings/openrouter-model-catalog.server'
 import { POST } from '@/app/api/tools/embeddings/route'
 
 const baseBody = {
@@ -37,6 +51,10 @@ function post(body: Record<string, unknown>) {
 describe('POST /api/tools/embeddings', () => {
   beforeEach(() => {
     vi.clearAllMocks()
+    mockGetOpenRouterEmbeddingModelMetadata.mockResolvedValue({
+      id: 'openrouter/qwen/qwen3-embedding-8b',
+      maxInputTokens: 32768,
+    })
     hybridAuthMockFns.mockCheckInternalAuth.mockResolvedValue({
       success: true,
       userId: 'user-1',
@@ -142,6 +160,7 @@ describe('POST /api/tools/embeddings', () => {
       ['hello world'],
       expect.objectContaining({
         apiKey: 'or-test',
+        maxInputTokens: 32768,
         model: 'openrouter/qwen/qwen3-embedding-8b',
       })
     )
@@ -175,6 +194,23 @@ describe('POST /api/tools/embeddings', () => {
     expect(mockEmbedOpenRouter).not.toHaveBeenCalled()
   })
 
+  it('rejects a qualified model that is absent from OpenRouter', async () => {
+    mockGetOpenRouterEmbeddingModelMetadata.mockRejectedValue(
+      new OpenRouterEmbeddingModelNotFoundError('openrouter/example/missing')
+    )
+
+    const response = await post({
+      provider: 'openrouter',
+      model: 'openrouter/example/missing',
+      input: 'hello world',
+      apiKey: 'or-test',
+    })
+
+    expect(response.status).toBe(400)
+    expect((await response.json()).error).toContain('Unsupported OpenRouter embedding model')
+    expect(mockEmbedOpenRouter).not.toHaveBeenCalled()
+  })
+
   it('keeps API keys required for non-OpenRouter providers', async () => {
     const response = await post({
       provider: 'openai',

apps/sim/app/api/tools/embeddings/route.ts

@@ -17,6 +17,11 @@ import {
   findEmbeddingModelInfo,
   resolveDimensions,
 } from '@/lib/embeddings'
+import {
+  getOpenRouterEmbeddingModelMetadata,
+  type OpenRouterEmbeddingModelMetadata,
+  OpenRouterEmbeddingModelNotFoundError,
+} from '@/lib/embeddings/openrouter-model-catalog.server'
 import { normalizeOpenRouterEmbeddingModelId } from '@/lib/embeddings/openrouter-models'
 
 const logger = createLogger('EmbeddingsToolAPI')
@@ -112,6 +117,7 @@ export const POST = withRouteHandler(async (request: NextRequest) => {
   }
 
   let resolvedModel: string
+  let openRouterModelMetadata: OpenRouterEmbeddingModelMetadata | undefined
   if (provider === 'openrouter') {
     try {
       resolvedModel = normalizeOpenRouterEmbeddingModelId(
@@ -123,6 +129,23 @@ export const POST = withRouteHandler(async (request: NextRequest) => {
         { status: 400 }
       )
     }
+    try {
+      openRouterModelMetadata = await getOpenRouterEmbeddingModelMetadata(resolvedModel)
+    } catch (error) {
+      const modelError = error instanceof OpenRouterEmbeddingModelNotFoundError
+      return NextResponse.json(
+        {
+          success: false,
+          error: getErrorMessage(
+            error,
+            modelError
+              ? 'Unsupported OpenRouter embedding model'
+              : 'Failed to load OpenRouter embedding model metadata'
+          ),
+        },
+        { status: modelError ? 400 : 502 }
+      )
+    }
   } else {
     resolvedModel = model || DEFAULT_MODEL_BY_PROVIDER[provider]
   }
@@ -163,26 +186,32 @@ export const POST = withRouteHandler(async (request: NextRequest) => {
   logger.info(`Embedding ${texts.length} input(s) with ${provider}/${resolvedModel}`)
 
   try {
-    const result =
-      provider === 'openrouter'
-        ? await embedOpenRouter(texts, {
-            model: resolvedModel,
-            dimensions,
-            apiKey,
-            projectInputs: null,
-          })
-        : await embed(texts, {
-            model: resolvedModel,
-            taskType,
-            dimensions,
-            apiKey,
-            /**
-             * Callers reach this route through a tool whose `request.modelInput`
-             * already projected `input` at the HTTP hop, so projecting again here
-             * would run the substitution over already-projected content.
-             */
-            projectInputs: null,
-          })
+    let result
+    if (provider === 'openrouter') {
+      if (!openRouterModelMetadata) {
+        throw new Error('OpenRouter embedding model metadata was not resolved')
+      }
+      result = await embedOpenRouter(texts, {
+        model: resolvedModel,
+        dimensions,
+        apiKey,
+        maxInputTokens: openRouterModelMetadata.maxInputTokens,
+        projectInputs: null,
+      })
+    } else {
+      result = await embed(texts, {
+        model: resolvedModel,
+        taskType,
+        dimensions,
+        apiKey,
+        /**
+         * Callers reach this route through a tool whose `request.modelInput`
+         * already projected `input` at the HTTP hop, so projecting again here
+         * would run the substitution over already-projected content.
+         */
+        projectInputs: null,
+      })
+    }
 
     return NextResponse.json({
       success: true,

apps/sim/lib/api/contracts/providers.ts

@@ -64,6 +64,10 @@ export const openRouterEmbeddingModelsUpstreamResponseSchema = z.object({
     z
       .object({
         id: z.string().min(1, 'OpenRouter embedding model id cannot be empty'),
+        context_length: z
+          .number()
+          .int('OpenRouter embedding context length must be an integer')
+          .positive('OpenRouter embedding context length must be positive'),
       })
       .passthrough()
   ),

apps/sim/lib/embeddings/client.test.ts

@@ -451,6 +451,7 @@ describe('embedOpenRouter', () => {
     const result = await embedOpenRouter(['alpha', 'beta'], {
       model: 'openrouter/qwen/qwen3-embedding-8b',
       apiKey: 'or-test',
+      maxInputTokens: 32768,
       projectInputs: null,
     })
 
@@ -480,6 +481,7 @@ describe('embedOpenRouter', () => {
       embedOpenRouter(['alpha', 'beta'], {
         model: 'openrouter/qwen/qwen3-embedding-8b',
         apiKey: 'or-test',
+        maxInputTokens: 32768,
         projectInputs: null,
       })
     ).rejects.toThrow('returned 1 embeddings for 2 inputs')
@@ -492,10 +494,58 @@ describe('embedOpenRouter', () => {
       embedOpenRouter(['alpha', 'beta'], {
         model: 'openrouter/qwen/qwen3-embedding-8b',
         apiKey: 'or-test',
+        maxInputTokens: 32768,
         projectInputs: null,
       })
     ).rejects.toThrow('inconsistent dimensions')
   })
+
+  it('truncates inputs to the selected model context length', async () => {
+    fetchMock.mockResolvedValue(jsonResponse(openAIBody([[1, 2]])))
+
+    await embedOpenRouter(['alpha beta gamma'], {
+      model: 'openrouter/thenlper/gte-base',
+      apiKey: 'or-test',
+      maxInputTokens: 1,
+      projectInputs: null,
+    })
+
+    const [, init] = fetchMock.mock.calls[0]
+    const body = JSON.parse((init as RequestInit).body as string)
+    expect(body.input).toHaveLength(1)
+    expect(body.input[0]).not.toBe('alpha beta gamma')
+  })
+
+  it('splits dynamic models at the provider item limit and recombines in order', async () => {
+    fetchMock.mockImplementation(async (_url, init) => {
+      const body = JSON.parse((init as RequestInit).body as string)
+      const inputs = body.input as string[]
+      return jsonResponse(
+        openAIBody(
+          inputs.map((input) => [Number(input.slice(1))]),
+          inputs.length
+        )
+      )
+    })
+    const inputs = Array.from({ length: 2049 }, (_, index) => `i${index}`)
+
+    const result = await embedOpenRouter(inputs, {
+      model: 'openrouter/qwen/qwen3-embedding-8b',
+      apiKey: 'or-test',
+      maxInputTokens: 32768,
+      projectInputs: null,
+    })
+
+    expect(fetchMock).toHaveBeenCalledTimes(2)
+    expect(
+      fetchMock.mock.calls
+        .map(([, init]) => JSON.parse((init as RequestInit).body as string).input.length)
+        .sort((a, b) => a - b)
+    ).toEqual([1, 2048])
+    expect(result.embeddings).toHaveLength(2049)
+    expect(result.embeddings[0]).toEqual([0])
+    expect(result.embeddings[2048]).toEqual([2048])
+  })
 })
 
 describe('knowledge embedding transport fallback', () => {