Add multi-provider embedding + separate [embedding] config

- Support Gemini (gemini-embedding-001), Azure OpenAI, OpenAI embeddings
- Add [embedding] section to kiwi.toml (separate from [llm] for different providers)
- Semantic search uses embedding config, falls back to llm config
- Fix Gemini model name: text-embedding-004 → gemini-embedding-001
- Add EmbeddingConfig interface to config.ts
- 125/416 pages embedded (Gemini free tier rate limit)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: jiunbae

src/config.ts

@@ -21,10 +21,16 @@ export interface Persona {
   content_style: string; // injected into content generation prompts
 }
 
+export interface EmbeddingConfig {
+  provider: string; // "gemini" | "openai" | "azure-openai"
+  api_key: string;
+}
+
 export interface KiwiConfig {
   project: { name: string; created: string };
   build: { output_dir: string };
   llm: LLMConfig;
+  embedding?: EmbeddingConfig; // separate config for embeddings (optional, falls back to llm)
   deploy: { target: string };
   personas?: Persona[];
   active_persona?: string; // name of the active persona

src/index.ts

@@ -270,14 +270,17 @@ program
       console.log(`\x1b[32m✅ ${count}개 페이지가 빌드되었습니다!\x1b[0m`);
       console.log(`  출력: ${join(root, config.build.output_dir)}/`);
 
-      // Generate embeddings (optional — requires API key)
+      // Generate embeddings (optional — uses [embedding] config or falls back to [llm])
       try {
-        if (config.llm.api_key && config.llm.endpoint) {
+        const embConfig = config.embedding
+          ? { ...config.llm, provider: config.embedding.provider, api_key: config.embedding.api_key }
+          : config.llm;
+        if (embConfig.api_key && embConfig.provider !== "demo") {
           const { generateMissingEmbeddings } = await import("./services/embedding");
-          await generateMissingEmbeddings(store, config.llm, (msg) => console.log(msg));
+          await generateMissingEmbeddings(store, embConfig, (msg) => console.log(msg));
         }
-      } catch {
-        // Embedding generation is optional
+      } catch (e: unknown) {
+        console.log(`  ⚠ 임베딩 생성 건너뜀: ${e instanceof Error ? e.message : String(e)}`);
       }
     } catch (e: unknown) {
       const message = e instanceof Error ? e.message : String(e);

src/server.ts

@@ -387,9 +387,13 @@ export function startServer(root: string, port: number, host: string): void {
         // Try semantic search first (if embeddings exist)
         try {
           const searchConfig = loadConfig(root);
-          if (searchConfig.llm.api_key && searchConfig.llm.endpoint) {
+          // Use embedding config if available, fall back to llm config
+          const embeddingLlmConfig = searchConfig.embedding
+            ? { ...searchConfig.llm, provider: searchConfig.embedding.provider, api_key: searchConfig.embedding.api_key }
+            : searchConfig.llm;
+          if (embeddingLlmConfig.api_key) {
             const { semanticSearch } = await import("./services/embedding");
-            const semanticResults = await semanticSearch(query, store, searchConfig.llm, 5);
+            const semanticResults = await semanticSearch(query, store, embeddingLlmConfig, 5);
             if (semanticResults.length > 0) {
               return Response.json({
                 results: semanticResults.map(r => ({

src/services/embedding.ts

@@ -12,24 +12,59 @@ function cosineSimilarity(a: Float32Array, b: Float32Array): number {
   return dot / (Math.sqrt(normA) * Math.sqrt(normB));
 }
 
-// Get embedding from Azure OpenAI
+// Get embedding — auto-detect provider
 async function getEmbedding(text: string, config: LLMConfig): Promise<Float32Array> {
-  const { default: OpenAI } = await import("openai");
+  const input = text.slice(0, 8000);
 
-  // Azure OpenAI embedding endpoint
-  const client = new OpenAI({
-    apiKey: config.api_key,
-    baseURL: `${config.endpoint}/openai/deployments/text-embedding-3-small`,
-    defaultQuery: { "api-version": "2024-06-01" },
-    defaultHeaders: { "api-key": config.api_key },
+  if (config.provider === "gemini") {
+    return await geminiEmbedding(input, config);
+  } else if (config.provider === "azure-openai") {
+    return await azureEmbedding(input, config);
+  } else if (config.provider === "openai") {
+    return await openaiEmbedding(input, config);
+  }
+  throw new Error(`Embedding not supported for provider: ${config.provider}`);
+}
+
+// Gemini Embedding API (free)
+async function geminiEmbedding(text: string, config: LLMConfig): Promise<Float32Array> {
+  const url = `https://generativelanguage.googleapis.com/v1beta/models/gemini-embedding-001:embedContent`;
+  const resp = await fetch(url, {
+    method: "POST",
+    headers: { "Content-Type": "application/json", "x-goog-api-key": config.api_key },
+    body: JSON.stringify({
+      model: "models/gemini-embedding-001",
+      content: { parts: [{ text }] }
+    })
   });
+  if (!resp.ok) throw new Error(`Gemini embedding error (${resp.status})`);
+  const data = await resp.json() as { embedding: { values: number[] } };
+  return new Float32Array(data.embedding.values);
+}
 
-  const response = await client.embeddings.create({
-    model: "text-embedding-3-small",
-    input: text.slice(0, 8000), // Limit input length
+// Azure OpenAI Embedding
+async function azureEmbedding(text: string, config: LLMConfig): Promise<Float32Array> {
+  const url = `${config.endpoint}/openai/deployments/text-embedding-3-small/embeddings?api-version=2024-06-01`;
+  const resp = await fetch(url, {
+    method: "POST",
+    headers: { "Content-Type": "application/json", "api-key": config.api_key },
+    body: JSON.stringify({ input: text, model: "text-embedding-3-small" })
   });
+  if (!resp.ok) throw new Error(`Azure embedding error (${resp.status})`);
+  const data = await resp.json() as { data: Array<{ embedding: number[] }> };
+  return new Float32Array(data.data[0].embedding);
+}
 
-  return new Float32Array(response.data[0].embedding);
+// OpenAI Embedding
+async function openaiEmbedding(text: string, config: LLMConfig): Promise<Float32Array> {
+  const resp = await fetch("https://api.openai.com/v1/embeddings", {
+    method: "POST",
+    headers: { "Content-Type": "application/json", "Authorization": `Bearer ${config.api_key}` },
+    body: JSON.stringify({ input: text, model: "text-embedding-3-small" })
+  });
+  if (!resp.ok) throw new Error(`OpenAI embedding error (${resp.status})`);
+  const data = await resp.json() as { data: Array<{ embedding: number[] }> };
+  return new Float32Array(data.data[0].embedding);
 }
 
 // Generate embeddings for all pages that don't have one