@lota-sdk/core 0.4.38 → 0.4.40

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lota-sdk/core",
-  "version": "0.4.38",
+  "version": "0.4.40",
   "files": [
     "src",
     "infrastructure/schema"
@@ -32,7 +32,7 @@
     "@ai-sdk/provider": "^3.0.9",
     "@chat-adapter/slack": "^4.26.0",
     "@chat-adapter/state-ioredis": "^4.26.0",
-    "@lota-sdk/shared": "0.4.38",
+    "@lota-sdk/shared": "0.4.40",
     "@mendable/firecrawl-js": "^4.20.0",
     "@surrealdb/node": "^3.0.3",
     "ai": "^6.0.170",

package/src/ai-gateway/ai-gateway.ts

@@ -1,13 +1,14 @@
 import { devToolsMiddleware } from '@ai-sdk/devtools'
 import { createOpenAI } from '@ai-sdk/openai'
 import type { JSONSchema7 } from '@ai-sdk/provider'
-import { wrapEmbeddingModel, wrapLanguageModel } from 'ai'
+import { wrapLanguageModel } from 'ai'
 import type { LanguageModelMiddleware } from 'ai'
 import { Cause, Clock, Context, Duration, Effect, Fiber, Layer, Semaphore } from 'effect'
 
 import { DEFAULT_AI_GATEWAY_URL } from '../config/constants'
 import { ERROR_TAGS, AiGenerationError, ConfigurationError } from '../effect/errors'
 import { RuntimeConfigServiceTag } from '../effect/services'
+import { openRouterEmbeddingModel } from '../embeddings/openrouter'
 import { isRecord, readString } from '../utils/string'
 import { buildAiGatewayCacheHeaders } from './cache-headers'
 
@@ -15,7 +16,6 @@ type AiGatewayChatResponse = { body?: unknown }
 type AiGatewayTransformParamsOptions = Parameters<NonNullable<LanguageModelMiddleware['transformParams']>>[0]
 type WrapStreamOptions = Parameters<NonNullable<LanguageModelMiddleware['wrapStream']>>[0]
 type AiGatewayLanguageModel = Parameters<typeof wrapLanguageModel>[0]['model']
-type AiGatewayEmbeddingModel = Parameters<typeof wrapEmbeddingModel>[0]['model']
 type AiGatewayCallOptions = WrapStreamOptions['params']
 type AiGatewayFunctionTool = Extract<NonNullable<AiGatewayCallOptions['tools']>[number], { type: 'function' }>
 type AiGatewayGenerateResult = Awaited<ReturnType<WrapStreamOptions['doGenerate']>>
@@ -44,8 +44,6 @@ const AI_GATEWAY_MAX_RETRIES = 4
 const AI_GATEWAY_MAX_RETRY_DELAY_MS = 15_000
 const OPENAI_RESPONSES_PROVIDER_ID = 'openai.responses'
 const OPENAI_CHAT_PROVIDER_ID = 'openai.chat'
-const OPENAI_EMBEDDING_PROVIDER_ID = 'openai.embedding'
-const OPENAI_EMBEDDING_MAX_PER_CALL = 2_048
 const RETRYABLE_NETWORK_ERROR_CODES = new Set([
   'ECONNABORTED',
   'ECONNREFUSED',
@@ -1243,23 +1241,6 @@ function createAiGatewayLanguageModelPlaceholder(modelId: string, providerId: st
   }
 }
 
-function createAiGatewayEmbeddingModelPlaceholder(modelId: string): AiGatewayEmbeddingModel {
-  return {
-    specificationVersion: 'v3',
-    provider: OPENAI_EMBEDDING_PROVIDER_ID,
-    modelId,
-    maxEmbeddingsPerCall: OPENAI_EMBEDDING_MAX_PER_CALL,
-    supportsParallelCalls: true,
-    doEmbed: () =>
-      Promise.reject(
-        new Error(
-          `[ai-gateway] AiGateway embedding model ${modelId}.doEmbed was invoked without the gateway middleware; ` +
-            'this call path should be fully handled by aiGatewayEmbeddingModel middleware.',
-        ),
-      ),
-  }
-}
-
 export function aiGatewayModel(modelId: string, deps?: AiGatewayDeps) {
   if (isOpenRouterModel(modelId)) {
     return aiGatewayChatModel(modelId, deps)
@@ -1286,28 +1267,8 @@ export function aiGatewayChatModel(modelId: string, deps?: AiGatewayDeps) {
   )
 }
 
-export function aiGatewayEmbeddingModel(modelId: string, deps?: AiGatewayDeps) {
-  return wrapEmbeddingModel({
-    model: createAiGatewayEmbeddingModelPlaceholder(modelId),
-    middleware: {
-      specificationVersion: 'v3',
-      wrapEmbed: ({ params }) => {
-        const resolvedDeps = resolveAiGatewayDeps(deps)
-        const embeddingModel = resolvedDeps.gateway.provider.embeddingModel(modelId)
-        return resolvedDeps.runPromise(
-          withAiGatewayConcurrency(
-            withAiGatewayResilience(
-              'ai-gateway.embed',
-              Effect.tryPromise({
-                try: () => embeddingModel.doEmbed(params),
-                catch: (cause) => classifyAiGatewayError('ai-gateway.embed', cause),
-              }),
-            ).pipe(Effect.withSpan('AiGateway.embed'), Effect.annotateSpans({ modelId })),
-          ).pipe(Effect.provideService(AiGatewayTag, resolvedDeps.gateway)),
-        )
-      },
-    },
-  })
+export function aiGatewayEmbeddingModel(modelId: string, _deps?: AiGatewayDeps) {
+  return openRouterEmbeddingModel(modelId)
 }
 
 /**
@@ -1319,7 +1280,7 @@ export function aiGatewayEmbeddingModel(modelId: string, deps?: AiGatewayDeps) {
 export type AiGatewayModels = {
   model(modelId: string): ReturnType<typeof aiGatewayModel>
   chatModel(modelId: string): ReturnType<typeof aiGatewayChatModel>
-  embeddingModel(modelId: string): ReturnType<typeof aiGatewayEmbeddingModel>
+  embeddingModel(modelId: string): ReturnType<typeof openRouterEmbeddingModel>
   openRouterResponseHealingModel(modelId: string): ReturnType<typeof aiGatewayOpenRouterResponseHealingModel>
 }
 
@@ -1327,7 +1288,7 @@ export function createAiGatewayModels(deps: AiGatewayDeps): AiGatewayModels {
   return {
     model: (modelId: string) => aiGatewayModel(modelId, deps),
     chatModel: (modelId: string) => aiGatewayChatModel(modelId, deps),
-    embeddingModel: (modelId: string) => aiGatewayEmbeddingModel(modelId, deps),
+    embeddingModel: (modelId: string) => openRouterEmbeddingModel(modelId),
     openRouterResponseHealingModel: (modelId: string) => aiGatewayOpenRouterResponseHealingModel(modelId, deps),
   }
 }

package/src/embeddings/openrouter.ts

@@ -0,0 +1,46 @@
+import { createOpenAI } from '@ai-sdk/openai'
+
+import { ConfigurationError } from '../effect/errors'
+
+const OPENROUTER_BASE_URL = 'https://openrouter.ai/api/v1'
+const OPENROUTER_MODEL_PREFIX = 'openrouter/' as const
+const OPENAI_TEXT_EMBEDDING_3_SMALL_MODEL_ID = 'openai/text-embedding-3-small'
+const OPENROUTER_API_KEY_ENV = 'OPENROUTER_API_KEY'
+
+type OpenRouterProvider = ReturnType<typeof createOpenAI>
+
+let cachedProvider: { apiKey: string; provider: OpenRouterProvider } | null = null
+
+function normalizeOpenRouterEmbeddingModelId(modelId: string): string {
+  const normalized = modelId.trim()
+  if (normalized === 'text-embedding-3-small') return OPENAI_TEXT_EMBEDDING_3_SMALL_MODEL_ID
+  return normalized.startsWith(OPENROUTER_MODEL_PREFIX) ? normalized.slice(OPENROUTER_MODEL_PREFIX.length) : normalized
+}
+
+function readOpenRouterApiKey(env: Record<string, string | undefined> = process.env): string {
+  const apiKey = env[OPENROUTER_API_KEY_ENV]?.trim()
+  if (!apiKey) {
+    throw new ConfigurationError({
+      message: `[embeddings-provider] ${OPENROUTER_API_KEY_ENV} is required for direct OpenRouter embeddings.`,
+      key: OPENROUTER_API_KEY_ENV,
+    })
+  }
+  return apiKey
+}
+
+function getOpenRouterProvider(apiKey: string): OpenRouterProvider {
+  if (cachedProvider?.apiKey === apiKey) return cachedProvider.provider
+
+  const provider = createOpenAI({ apiKey, baseURL: OPENROUTER_BASE_URL })
+  cachedProvider = { apiKey, provider }
+  return provider
+}
+
+export function openRouterEmbeddingModel(modelId: string) {
+  const normalizedModelId = normalizeOpenRouterEmbeddingModelId(modelId)
+  if (!normalizedModelId) {
+    throw new ConfigurationError({ message: '[embeddings-provider] Model id is required.', key: 'embeddingModelId' })
+  }
+
+  return getOpenRouterProvider(readOpenRouterApiKey()).embeddingModel(normalizedModelId)
+}

package/src/embeddings/provider.ts

@@ -1,10 +1,11 @@
 import { embed, embedMany } from 'ai'
 import { Schema, Effect } from 'effect'
 
-import { aiGatewayEmbeddingModel } from '../ai-gateway/ai-gateway'
 import { ERROR_TAGS, ConfigurationError } from '../effect/errors'
+import { openRouterEmbeddingModel } from './openrouter'
 
 const SUPPORTED_EMBEDDING_PREFIXES = ['openai/', 'openrouter/'] as const
+const SUPPORTED_BARE_EMBEDDING_MODEL_IDS = ['text-embedding-3-small'] as const
 
 type SharedEmbeddingCache = {
   get(model: string, text: string): Promise<number[] | null>
@@ -30,14 +31,18 @@ function resolveEmbeddingModel(modelId: string) {
     throw new ConfigurationError({ message: '[embeddings-provider] Model id is required.', key: 'embeddingModelId' })
   }
 
+  if (SUPPORTED_BARE_EMBEDDING_MODEL_IDS.includes(normalized as (typeof SUPPORTED_BARE_EMBEDDING_MODEL_IDS)[number])) {
+    return openRouterEmbeddingModel(`openai/${normalized}`)
+  }
+
   if (!SUPPORTED_EMBEDDING_PREFIXES.some((prefix) => normalized.startsWith(prefix))) {
     throw new ConfigurationError({
-      message: `[embeddings-provider] Unsupported model id "${modelId}". Use one of: ${SUPPORTED_EMBEDDING_PREFIXES.join(', ')}*.`,
+      message: `[embeddings-provider] Unsupported model id "${modelId}". Use one of: ${SUPPORTED_EMBEDDING_PREFIXES.join(', ')}* or ${SUPPORTED_BARE_EMBEDDING_MODEL_IDS.join(', ')}.`,
       key: 'embeddingModelId',
     })
   }
 
-  return aiGatewayEmbeddingModel(normalized)
+  return openRouterEmbeddingModel(normalized)
 }
 
 function normalizeEmbedding(embedding: readonly number[]): number[] {

package/src/index.ts

@@ -4,6 +4,7 @@ export * from './ai-gateway'
 export * from './config'
 export * from './db'
 export * from './document'
+export * from './embeddings/openrouter'
 export * from './queues'
 export * from './redis'
 export * from './runtime'

package/src/runtime/runtime-config.ts

@@ -291,6 +291,7 @@ export const LOTA_RUNTIME_ENV_KEYS = Object.freeze([
   'REDIS_URL',
   'AI_GATEWAY_URL',
   'AI_GATEWAY_KEY',
+  'OPENROUTER_API_KEY',
   'AI_EMBEDDING_MODEL',
   'AI_GATEWAY_MAX_CONCURRENCY',
   'S3_ENDPOINT',