refactor(knowledge): lock embedding model to KB_EMBEDDING_MODEL env var

Remove the user-facing model picker from the KB create modal and the
embeddingModel field from the create/update API schemas. The active model
is now selected server-side via KB_EMBEDDING_MODEL, which collapses Azure
routing to a single deployment (KB_OPENAI_MODEL_NAME) and drops the
per-model AZURE_OPENAI_DEPLOYMENT_TEXT_EMBEDDING_3_* env vars and
SUPPORTED_EMBEDDING_MODEL_IDS / UI-only label+description registry fields.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: waleedlatif1

apps/sim/app/api/knowledge/[id]/route.ts

@@ -27,8 +27,6 @@ const logger = createLogger('KnowledgeBaseByIdAPI')
 const UpdateKnowledgeBaseSchema = z.object({
   name: z.string().min(1, 'Name is required').optional(),
   description: z.string().optional(),
-  embeddingModel: z.literal('text-embedding-3-small').optional(),
-  embeddingDimension: z.literal(1536).optional(),
   workspaceId: z.string().nullable().optional(),
   chunkingConfig: z
     .object({

apps/sim/app/api/knowledge/route.ts

@@ -6,11 +6,7 @@ import { getSession } from '@/lib/auth'
 import { PlatformEvents } from '@/lib/core/telemetry'
 import { generateRequestId } from '@/lib/core/utils/request'
 import { withRouteHandler } from '@/lib/core/utils/with-route-handler'
-import {
-  DEFAULT_EMBEDDING_MODEL,
-  EMBEDDING_DIMENSIONS,
-  SUPPORTED_EMBEDDING_MODEL_IDS,
-} from '@/lib/knowledge/embeddings'
+import { EMBEDDING_DIMENSIONS, getConfiguredEmbeddingModel } from '@/lib/knowledge/embeddings'
 import {
   createKnowledgeBase,
   getKnowledgeBases,
@@ -25,10 +21,6 @@ const CreateKnowledgeBaseSchema = z.object({
   name: z.string().min(1, 'Name is required'),
   description: z.string().optional(),
   workspaceId: z.string().min(1, 'Workspace ID is required'),
-  embeddingModel: z
-    .enum(SUPPORTED_EMBEDDING_MODEL_IDS as [string, ...string[]])
-    .default(DEFAULT_EMBEDDING_MODEL),
-  embeddingDimension: z.literal(EMBEDDING_DIMENSIONS).default(EMBEDDING_DIMENSIONS),
   chunkingConfig: z
     .object({
       maxSize: z.number().min(100).max(4000).default(1024),
@@ -125,9 +117,13 @@ export const POST = withRouteHandler(async (req: NextRequest) => {
     try {
       const validatedData = CreateKnowledgeBaseSchema.parse(body)
 
+      const embeddingModel = getConfiguredEmbeddingModel()
+
       const createData = {
         ...validatedData,
         userId: session.user.id,
+        embeddingModel,
+        embeddingDimension: EMBEDDING_DIMENSIONS,
       }
 
       const newKnowledgeBase = await createKnowledgeBase(createData, requestId)
@@ -173,8 +169,8 @@ export const POST = withRouteHandler(async (req: NextRequest) => {
         metadata: {
           name: validatedData.name,
           description: validatedData.description,
-          embeddingModel: validatedData.embeddingModel,
-          embeddingDimension: validatedData.embeddingDimension,
+          embeddingModel,
+          embeddingDimension: EMBEDDING_DIMENSIONS,
           chunkingStrategy: validatedData.chunkingConfig.strategy,
           chunkMaxSize: validatedData.chunkingConfig.maxSize,
           chunkMinSize: validatedData.chunkingConfig.minSize,

apps/sim/app/api/v1/knowledge/route.ts

@@ -2,11 +2,7 @@ import { AuditAction, AuditResourceType, recordAudit } from '@sim/audit'
 import { type NextRequest, NextResponse } from 'next/server'
 import { z } from 'zod'
 import { withRouteHandler } from '@/lib/core/utils/with-route-handler'
-import {
-  DEFAULT_EMBEDDING_MODEL,
-  EMBEDDING_DIMENSIONS,
-  SUPPORTED_EMBEDDING_MODEL_IDS,
-} from '@/lib/knowledge/embeddings'
+import { EMBEDDING_DIMENSIONS, getConfiguredEmbeddingModel } from '@/lib/knowledge/embeddings'
 import { createKnowledgeBase, getKnowledgeBases } from '@/lib/knowledge/service'
 import {
   authenticateRequest,
@@ -34,9 +30,6 @@ const CreateKBSchema = z.object({
   workspaceId: z.string().min(1, 'Workspace ID is required'),
   name: z.string().min(1, 'Name is required').max(255, 'Name must be 255 characters or less'),
   description: z.string().max(1000, 'Description must be 1000 characters or less').optional(),
-  embeddingModel: z
-    .enum(SUPPORTED_EMBEDDING_MODEL_IDS as [string, ...string[]])
-    .default(DEFAULT_EMBEDDING_MODEL),
   chunkingConfig: ChunkingConfigSchema.optional().default({
     maxSize: 1024,
     minSize: 100,
@@ -89,7 +82,7 @@ export const POST = withRouteHandler(async (request: NextRequest) => {
     const validation = validateSchema(CreateKBSchema, body.data)
     if (!validation.success) return validation.response
 
-    const { workspaceId, name, description, embeddingModel, chunkingConfig } = validation.data
+    const { workspaceId, name, description, chunkingConfig } = validation.data
 
     const accessError = await validateWorkspaceAccess(rateLimit, userId, workspaceId, 'write')
     if (accessError) return accessError
@@ -100,7 +93,7 @@ export const POST = withRouteHandler(async (request: NextRequest) => {
         description,
         workspaceId,
         userId,
-        embeddingModel,
+        embeddingModel: getConfiguredEmbeddingModel(),
         embeddingDimension: EMBEDDING_DIMENSIONS,
         chunkingConfig: chunkingConfig ?? { maxSize: 1024, minSize: 100, overlap: 200 },
       },

apps/sim/app/workspace/[workspaceId]/knowledge/components/create-base-modal/create-base-modal.tsx

@@ -22,11 +22,6 @@ import {
 } from '@/components/emcn'
 import type { StrategyOptions } from '@/lib/chunkers/types'
 import { cn } from '@/lib/core/utils/cn'
-import {
-  DEFAULT_EMBEDDING_MODEL,
-  SUPPORTED_EMBEDDING_MODEL_IDS,
-  SUPPORTED_EMBEDDING_MODELS,
-} from '@/lib/knowledge/embedding-models'
 import { formatFileSize, validateKnowledgeBaseFile } from '@/lib/uploads/utils/file-utils'
 import { ACCEPT_ATTRIBUTE } from '@/lib/uploads/utils/validation'
 import { useKnowledgeUpload } from '@/app/workspace/[workspaceId]/knowledge/hooks/use-knowledge-upload'
@@ -57,13 +52,6 @@ const STRATEGY_COMBOBOX_OPTIONS: ComboboxOption[] = STRATEGY_OPTIONS.map((o) =>
   value: o.value,
 }))
 
-const EMBEDDING_MODEL_OPTIONS: ComboboxOption[] = SUPPORTED_EMBEDDING_MODEL_IDS.map((id) => ({
-  label: SUPPORTED_EMBEDDING_MODELS[id].label,
-  value: id,
-}))
-
-const EMBEDDING_MODEL_VALUES = SUPPORTED_EMBEDDING_MODEL_IDS as [string, ...string[]]
-
 const FormSchema = z
   .object({
     name: z
@@ -87,7 +75,6 @@ const FormSchema = z
     strategy: z.enum(['auto', 'text', 'regex', 'recursive', 'sentence', 'token']).default('auto'),
     regexPattern: z.string().optional(),
     customSeparators: z.string().optional(),
-    embeddingModel: z.enum(EMBEDDING_MODEL_VALUES).default(DEFAULT_EMBEDDING_MODEL),
   })
   .refine(
     (data) => {
@@ -187,14 +174,12 @@ export const CreateBaseModal = memo(function CreateBaseModal({
       strategy: 'auto',
       regexPattern: '',
       customSeparators: '',
-      embeddingModel: DEFAULT_EMBEDDING_MODEL,
     },
     mode: 'onSubmit',
   })
 
   const nameValue = watch('name')
   const strategyValue = watch('strategy')
-  const embeddingModelValue = watch('embeddingModel')
 
   useEffect(() => {
     if (open) {
@@ -213,7 +198,6 @@ export const CreateBaseModal = memo(function CreateBaseModal({
         strategy: 'auto',
         regexPattern: '',
         customSeparators: '',
-        embeddingModel: DEFAULT_EMBEDDING_MODEL,
       })
     }
   }, [open, reset])
@@ -331,7 +315,6 @@ export const CreateBaseModal = memo(function CreateBaseModal({
         name: data.name,
         description: data.description || undefined,
         workspaceId: workspaceId,
-        embeddingModel: data.embeddingModel,
         chunkingConfig: {
           maxSize: data.maxChunkSize,
           minSize: data.minChunkSize,
@@ -477,23 +460,6 @@ export const CreateBaseModal = memo(function CreateBaseModal({
                   </p>
                 </div>
 
-                <div className='flex flex-col gap-2'>
-                  <Label>Embedding Model</Label>
-                  <Combobox
-                    options={EMBEDDING_MODEL_OPTIONS}
-                    value={embeddingModelValue}
-                    onChange={(value) =>
-                      setValue('embeddingModel', value as FormValues['embeddingModel'])
-                    }
-                    dropdownWidth='trigger'
-                    align='start'
-                  />
-                  <p className='text-[var(--text-muted)] text-xs'>
-                    {SUPPORTED_EMBEDDING_MODELS[embeddingModelValue]?.description ??
-                      'Choose how documents are vectorized. Cannot be changed after creation.'}
-                  </p>
-                </div>
-
                 <div className='flex flex-col gap-2'>
                   <Label>Chunking Strategy</Label>
                   <Combobox

apps/sim/hooks/queries/kb/knowledge.ts

@@ -697,7 +697,6 @@ export interface CreateKnowledgeBaseParams {
   name: string
   description?: string
   workspaceId: string
-  embeddingModel?: string
   chunkingConfig: {
     maxSize: number
     minSize: number

apps/sim/lib/core/config/env.ts

@@ -122,9 +122,8 @@ export const env = createEnv({
     AZURE_ANTHROPIC_ENDPOINT:              z.string().url().optional(),            // Azure Anthropic service endpoint
     AZURE_ANTHROPIC_API_KEY:               z.string().min(1).optional(),           // Azure Anthropic API key
     AZURE_ANTHROPIC_API_VERSION:           z.string().min(1).optional(),           // Azure Anthropic API version (e.g. 2023-06-01)
-    KB_OPENAI_MODEL_NAME:                  z.string().optional(),                  // Knowledge base OpenAI model name (works with both regular OpenAI and Azure OpenAI). Used as the Azure deployment for text-embedding-3-small (legacy/default).
-    AZURE_OPENAI_DEPLOYMENT_TEXT_EMBEDDING_3_SMALL: z.string().optional(),          // Azure deployment name serving text-embedding-3-small. If unset, falls back to KB_OPENAI_MODEL_NAME.
-    AZURE_OPENAI_DEPLOYMENT_TEXT_EMBEDDING_3_LARGE: z.string().optional(),          // Azure deployment name serving text-embedding-3-large. Required to use 3-large via Azure.
+    KB_OPENAI_MODEL_NAME:                  z.string().optional(),                  // Azure deployment name serving the configured KB embedding model (used only when AZURE_OPENAI_* credentials are set).
+    KB_EMBEDDING_MODEL:                    z.string().optional(),                  // Embedding model used for all new knowledge bases. Must be one of the supported model ids; defaults to text-embedding-3-small.
     WAND_OPENAI_MODEL_NAME:                z.string().optional(),                  // Wand generation OpenAI model name (works with both regular OpenAI and Azure OpenAI)
     OCR_AZURE_ENDPOINT:                    z.string().url().optional(),            // Azure Mistral OCR service endpoint
     OCR_AZURE_MODEL_NAME:                  z.string().optional(),                  // Azure Mistral OCR model name for document processing

apps/sim/lib/knowledge/embedding-models.ts

@@ -1,6 +1,8 @@
 /**
- * Client-safe registry of embedding models supported by the platform.
- * Kept free of server imports so it can be imported into UI code.
+ * Registry of embedding models supported by the platform.
+ * Selection happens server-side via the `KB_EMBEDDING_MODEL` env var; this
+ * registry exists to resolve provider, tokenizer, and pricing metadata at
+ * runtime for any model recorded on a knowledge base row.
  */
 
 export const EMBEDDING_DIMENSIONS = 1536
@@ -19,9 +21,6 @@ export interface EmbeddingModelInfo {
   pricingId: string
   /** Provider id for `estimateTokenCount` so token counts match the embedding provider's tokenization. */
   tokenizerProvider: TokenizerProviderId
-  label: string
-  /** Short user-facing description shown in the KB creation UI. */
-  description: string
 }
 
 export const SUPPORTED_EMBEDDING_MODELS: Record<string, EmbeddingModelInfo> = {
@@ -30,31 +29,21 @@ export const SUPPORTED_EMBEDDING_MODELS: Record<string, EmbeddingModelInfo> = {
     supportsCustomDimensions: true,
     pricingId: 'text-embedding-3-small',
     tokenizerProvider: 'openai',
-    label: 'OpenAI text-embedding-3-small',
-    description: 'Cheapest. Good for English-heavy retrieval at low cost.',
   },
   'text-embedding-3-large': {
     provider: 'openai',
     supportsCustomDimensions: true,
     pricingId: 'text-embedding-3-large',
     tokenizerProvider: 'openai',
-    label: 'OpenAI text-embedding-3-large',
-    description: 'Slightly better quality than 3-small at ~6.5× the cost.',
   },
   'gemini-embedding-001': {
     provider: 'gemini',
     supportsCustomDimensions: true,
     pricingId: 'gemini-embedding-001',
     tokenizerProvider: 'google',
-    label: 'Google gemini-embedding-001',
-    description: 'Strong multilingual retrieval. Good cost/quality balance.',
   },
 }
 
-export const SUPPORTED_EMBEDDING_MODEL_IDS = Object.keys(SUPPORTED_EMBEDDING_MODELS) as Array<
-  keyof typeof SUPPORTED_EMBEDDING_MODELS
->
-
 export function getEmbeddingModelInfo(model: string): EmbeddingModelInfo {
   const info = SUPPORTED_EMBEDDING_MODELS[model]
   if (!info) {

apps/sim/lib/knowledge/embeddings.ts

@@ -22,7 +22,6 @@ export {
   DEFAULT_EMBEDDING_MODEL,
   EMBEDDING_DIMENSIONS,
   getEmbeddingModelInfo,
-  SUPPORTED_EMBEDDING_MODEL_IDS,
   SUPPORTED_EMBEDDING_MODELS,
 } from '@/lib/knowledge/embedding-models'
 
@@ -174,19 +173,21 @@ function buildGeminiProvider(modelName: string, apiKey: string): ResolvedProvide
 }
 
 /**
- * Resolve the Azure deployment name for a given OpenAI embedding model.
- * Returns null if no deployment is configured for that model — caller falls
- * back to direct OpenAI rather than risk routing to a wrong-model deployment
- * (which would silently produce mismatched vectors).
+ * Returns the embedding model to use for new knowledge bases.
+ * Sourced from the `KB_EMBEDDING_MODEL` env var; falls back to the default if
+ * unset or set to an unsupported model.
  */
-function resolveAzureDeployment(embeddingModel: string): string | null {
-  if (embeddingModel === 'text-embedding-3-small') {
-    return env.AZURE_OPENAI_DEPLOYMENT_TEXT_EMBEDDING_3_SMALL || env.KB_OPENAI_MODEL_NAME || null
+export function getConfiguredEmbeddingModel(): string {
+  const configured = env.KB_EMBEDDING_MODEL
+  if (configured && SUPPORTED_EMBEDDING_MODELS[configured]) {
+    return configured
   }
-  if (embeddingModel === 'text-embedding-3-large') {
-    return env.AZURE_OPENAI_DEPLOYMENT_TEXT_EMBEDDING_3_LARGE || null
+  if (configured) {
+    logger.warn(
+      `KB_EMBEDDING_MODEL="${configured}" is not a supported embedding model — falling back to ${DEFAULT_EMBEDDING_MODEL}`
+    )
   }
-  return null
+  return DEFAULT_EMBEDDING_MODEL
 }
 
 async function resolveProvider(
@@ -198,7 +199,7 @@ async function resolveProvider(
   const azureApiVersion = env.AZURE_OPENAI_API_VERSION
   const isOpenAIModel = SUPPORTED_EMBEDDING_MODELS[embeddingModel]?.provider === 'openai'
   const azureDeployment =
-    isOpenAIModel && azureApiKey && azureEndpoint ? resolveAzureDeployment(embeddingModel) : null
+    isOpenAIModel && azureApiKey && azureEndpoint ? env.KB_OPENAI_MODEL_NAME || null : null
 
   if (azureDeployment) {
     return {