normalize LiteLLM embedding requests

src/routes/adminConfig.js

@@ -9,7 +9,8 @@ var { authMiddleware, adminMiddleware } = require('../middleware/auth');
 var PROMPTS = require('../utils/prompts');
 var logger = require('../utils/logger');
 var { gatewayUrl } = require('../utils/errors');
-var { GOOGLE_TTS_VOICES, OPENAI_TTS_VOICES, getTTSEnvProvider, getLiteLLMHeaders, getLiteLLMTTSModels, getTTSProvider, getTTSVoiceLists } = require('../utils/ttsProvider');
+var { GOOGLE_TTS_VOICES, OPENAI_TTS_VOICES, getTTSEnvProvider, getLiteLLMTTSModels, getTTSProvider, getTTSVoiceLists } = require('../utils/ttsProvider');
+var { getLiteLLMHeaders } = require('../utils/litellm');
 var { getSTTDependencies, getSTTModelLists, getSTTProvider } = require('../utils/sttProvider');
 
 router.use(authMiddleware);

src/utils/embeddings.js

@@ -4,6 +4,8 @@
 // ============================================================
 
 var axios = require('axios');
+var { gatewayUrl } = require('./errors');
+var { getLiteLLMHeaders } = require('./litellm');
 
 // Vertex AI embedding models (via LiteLLM or direct)
 // gemini-embedding-001: 768 dims, multilingual + code, best quality
@@ -61,12 +63,6 @@ async function generateEmbedding(text, opts) {
  */
 async function generateEmbeddingLiteLLM(text, model, dimensions) {
   try {
-    var base = process.env.LITELLM_API_BASE.replace(/\/+$/, '');
-    var headers = { 'Content-Type': 'application/json' };
-    if (process.env.LITELLM_API_KEY) {
-      headers['Authorization'] = 'Bearer ' + process.env.LITELLM_API_KEY;
-    }
-
     var payload = {
       model: model,
       input: text
@@ -77,8 +73,8 @@ async function generateEmbeddingLiteLLM(text, model, dimensions) {
       payload.dimensions = dimensions;
     }
 
-    var response = await axios.post(base + '/embeddings', payload, {
+    var response = await axios.post(gatewayUrl('/embeddings'), payload, {
-      headers: headers,
+      headers: getLiteLLMHeaders('application/json'),
       timeout: 30000
     });
 

src/utils/litellm.js

@@ -0,0 +1,10 @@
+function getLiteLLMHeaders(contentType) {
+  var headers = {};
+  if (contentType) headers['Content-Type'] = contentType;
+  if (process.env.LITELLM_API_KEY) headers.Authorization = 'Bearer ' + process.env.LITELLM_API_KEY;
+  return headers;
+}
+
+module.exports = {
+  getLiteLLMHeaders
+};

src/utils/ttsProvider.js

@@ -1,4 +1,5 @@
 const { isGoogleTTSConfigured } = require('./ttsGoogle');
+const { getLiteLLMHeaders } = require('./litellm');
 
 const GOOGLE_TTS_VOICES = ['en-US-Journey-F','en-US-Journey-D','en-US-Studio-O','en-US-Neural2-A','en-US-Neural2-C','en-US-Neural2-D','en-US-Neural2-E','en-US-Neural2-F','en-US-Neural2-G','en-US-Neural2-H','en-US-Neural2-I','en-US-Neural2-J'];
 const OPENAI_TTS_VOICES = ['alloy','echo','fable','onyx','nova','shimmer','ash','sage','coral'];
@@ -37,13 +38,6 @@ function getLiteLLMTTSModels(models) {
     .filter(isOpenAITTSModel);
 }
 
-function getLiteLLMHeaders(contentType) {
-  var headers = {};
-  if (contentType) headers['Content-Type'] = contentType;
-  if (process.env.LITELLM_API_KEY) headers.Authorization = 'Bearer ' + process.env.LITELLM_API_KEY;
-  return headers;
-}
-
 module.exports = {
   ELEVENLABS_DEFAULT_VOICES,
   GOOGLE_TTS_VOICES,