fix(embeddings): model-aware OpenAI dimensions + safe default model

- text-embedding-3-large returns 3072 dims, not 1536; use a getter on
  OpenAIEmbeddingProvider so dimensions resolve after modelName is set
- getConfiguredDimensions checks model name for 'large' before returning
  the OpenAI dimension value
- mergeConfig now defaults to text-embedding-3-small when
  EMBEDDING_PROVIDER=openai and EMBEDDING_MODEL is unset, avoiding a
  400 from the OpenAI API caused by sending 'Xenova/bge-small-en-v1.5'
- add text-embedding-3-large test case; fix stale test description

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: PatrickSys

src/core/indexer.ts

@@ -248,6 +248,14 @@ export class CodebaseIndexer {
     const defaultEmbeddingProvider =
       parseEmbeddingProviderName(process.env.EMBEDDING_PROVIDER) ?? 'transformers';
 
+    // When provider=openai and EMBEDDING_MODEL is not set, DEFAULT_MODEL resolves to the
+    // transformers fallback (Xenova/bge-small-en-v1.5), which the OpenAI API rejects.
+    // Use a sane OpenAI default instead.
+    const defaultModel =
+      defaultEmbeddingProvider === 'openai' && !process.env.EMBEDDING_MODEL
+        ? 'text-embedding-3-small'
+        : DEFAULT_MODEL;
+
     const defaultConfig: CodebaseConfig = {
       analyzers: {
         angular: { enabled: true, priority: 100 },
@@ -287,7 +295,7 @@ export class CodebaseIndexer {
       },
       embedding: {
         provider: defaultEmbeddingProvider,
-        model: DEFAULT_MODEL,
+        model: defaultModel,
         batchSize: 32
       },
       skipEmbedding: false,

src/embeddings/index.ts

@@ -20,7 +20,7 @@ import { TransformersEmbeddingProvider, MODEL_CONFIGS } from './transformers.js'
 export function getConfiguredDimensions(config: Partial<EmbeddingConfig> = {}): number {
   const provider = config.provider ?? parseEmbeddingProviderName(process.env.EMBEDDING_PROVIDER) ?? 'transformers';
   const model = config.model ?? process.env.EMBEDDING_MODEL ?? DEFAULT_MODEL;
-  if (provider === 'openai') return 1536; // text-embedding-3-small / ada-002
+  if (provider === 'openai') return model.includes('large') ? 3072 : 1536; // text-embedding-3-large: 3072, all others: 1536
   // Look up from the same MODEL_CONFIGS the provider uses — avoids stale hardcoded guesses
   return MODEL_CONFIGS[model]?.dimensions ?? 384;
 }

src/embeddings/openai.ts

@@ -11,7 +11,9 @@ interface OpenAIEmbeddingResponse {
  */
 export class OpenAIEmbeddingProvider implements EmbeddingProvider {
   readonly name = 'openai';
-  readonly dimensions = 1536; // Default for text-embedding-3-small
+  get dimensions(): number {
+    return this.modelName.includes('large') ? 3072 : 1536;
+  }
 
   constructor(
     readonly modelName: string = 'text-embedding-3-small',

tests/embedding-mismatch.test.ts

@@ -84,10 +84,14 @@ describe('getConfiguredDimensions', () => {
     ).toBe(384);
   });
 
-  it('returns 1536 for openai provider regardless of model', () => {
+  it('returns 1536 for text-embedding-3-small', () => {
     expect(getConfiguredDimensions({ provider: 'openai', model: 'text-embedding-3-small' })).toBe(1536);
   });
 
+  it('returns 3072 for text-embedding-3-large', () => {
+    expect(getConfiguredDimensions({ provider: 'openai', model: 'text-embedding-3-large' })).toBe(3072);
+  });
+
   it('returns 384 as fallback for unknown transformers model', () => {
     expect(getConfiguredDimensions({ provider: 'transformers', model: 'some/unknown-model' })).toBe(384);
   });