Merge pull request #55 from amihos/fix/embedding-config-mismatch-warning

fix: Add warning for embedding configuration mismatch

Author: nullure

backend/src/memory/embed.ts

@@ -153,11 +153,11 @@ async function emb_gemini(
 ): Promise<Record<string, number[]>> {
     if (!env.gemini_key) throw new Error("Gemini key missing");
     const prom = gem_q.then(async () => {
-        const url = `https://generativelanguage.googleapis.com/v1beta/models/embedding-001:batchEmbedContents?key=${env.gemini_key}`;
+        const url = `https://generativelanguage.googleapis.com/v1beta/models/text-embedding-004:batchEmbedContents?key=${env.gemini_key}`;
         for (let a = 0; a < 3; a++) {
             try {
                 const reqs = Object.entries(txts).map(([s, t]) => ({
-                    model: "models/embedding-001",
+                    model: "models/text-embedding-004",
                     content: { parts: [{ text: t }] },
                     taskType: task_map[s] || task_map.semantic,
                 }));

backend/src/server/index.ts

@@ -28,6 +28,16 @@ console.log(`[CONFIG] Vector Dimension: ${env.vec_dim}`);
 console.log(`[CONFIG] Cache Segments: ${env.cache_segments}`);
 console.log(`[CONFIG] Max Active Queries: ${env.max_active}`);
 
+// Warn about configuration mismatch that causes embedding incompatibility
+if (env.emb_kind !== "synthetic" && (tier === "hybrid" || tier === "fast")) {
+    console.warn(
+        `[CONFIG] ⚠️  WARNING: Embedding configuration mismatch detected!\n` +
+        `         OM_EMBEDDINGS=${env.emb_kind} but OM_TIER=${tier}\n` +
+        `         Storage will use ${env.emb_kind} embeddings, but queries will use synthetic embeddings.\n` +
+        `         This causes semantic search to fail. Set OM_TIER=deep to fix.`
+    );
+}
+
 app.use(req_tracker_mw());
 
 app.use((req: any, res: any, next: any) => {

models.yml

@@ -8,35 +8,35 @@
 episodic:
   ollama: nomic-embed-text
   openai: text-embedding-3-small
-  gemini: models/embedding-001
+  gemini: models/text-embedding-004
   aws: amazon.titan-embed-text-v2:0
   local: all-MiniLM-L6-v2
 
 semantic:
   ollama: nomic-embed-text
   openai: text-embedding-3-small
-  gemini: models/embedding-001
+  gemini: models/text-embedding-004
   aws: amazon.titan-embed-text-v2:0
   local: all-MiniLM-L6-v2
 
 procedural:
   ollama: nomic-embed-text
   openai: text-embedding-3-small
-  gemini: models/embedding-001
+  gemini: models/text-embedding-004
   aws: amazon.titan-embed-text-v2:0
   local: all-MiniLM-L6-v2
 
 emotional:
   ollama: nomic-embed-text
   openai: text-embedding-3-small
-  gemini: models/embedding-001
+  gemini: models/text-embedding-004
   aws: amazon.titan-embed-text-v2:0
   local: all-MiniLM-L6-v2
 
 reflective:
   ollama: nomic-embed-text
   openai: text-embedding-3-large
-  gemini: models/embedding-001
+  gemini: models/text-embedding-004
   aws: amazon.titan-embed-text-v2:0
   local: all-mpnet-base-v2
 # Available Ollama models (pull with: ollama pull <model>)
@@ -50,7 +50,8 @@ reflective:
 # - text-embedding-3-large (3072d)
 
 # Gemini models:
-# - models/embedding-001 (768d)
+# - models/text-embedding-004 (768d) - latest
+# - models/embedding-001 (768d) - deprecated
 
 #AWS models:
 # - amazon.titan-embed-text-v2:0 (1024d, 512d, 256)