feat: add OpenAI embedding/reranking/query expansion provider

Port PR tobi#116 (tobi/qmd) to current main, adapting to the refactored
codebase. Adds OpenAI as an alternative to local GGUF models, fixing
the ARM64 segfault during hybrid search (issue tobi#68).

Changes:
- New src/openai-llm.ts: OpenAI API client (embed, embedBatch, rerank,
  expandQuery) with exponential backoff and rate limiting
- llm.ts: setEmbeddingConfig(), getDefaultEmbeddingLLM(), isUsingOpenAI()
- collections.ts: EmbeddingProviderConfig type, getEmbeddingConfig()
- store.ts: Provider-aware embedding, chunking (tiktoken), expand, rerank
- qmd.ts: Startup config loading, provider-aware embed command
- package.json: openai + tiktoken dependencies

Config via ~/.config/qmd/index.yml:
  embedding:
    provider: openai
    openai:
      model: text-embedding-3-small

Or env: QMD_OPENAI=1 + OPENAI_API_KEY

Author: oscar1byte

bun.lock

@@ -20,7 +20,9 @@
   "dependencies": {
     "@modelcontextprotocol/sdk": "^1.25.1",
     "node-llama-cpp": "^3.14.5",
+    "openai": "^4.77.0",
     "sqlite-vec": "^0.1.7-alpha.2",
+    "tiktoken": "^1.0.22",
     "yaml": "^2.8.2",
     "zod": "^4.2.1"
   },

package.json

@@ -31,12 +31,24 @@ export interface Collection {
   update?: string;        // Optional bash command to run during qmd update
 }
 
+/**
+ * Embedding provider configuration (optional in config file)
+ */
+export interface EmbeddingProviderConfig {
+  provider?: "local" | "openai";  // Default: 'local'
+  openai?: {
+    api_key?: string;              // Falls back to OPENAI_API_KEY env var
+    model?: string;                // Default: 'text-embedding-3-small'
+  };
+}
+
 /**
  * The complete configuration file structure
  */
 export interface CollectionConfig {
   global_context?: string;                    // Context applied to all collections
   collections: Record<string, Collection>;    // Collection name -> config
+  embedding?: EmbeddingProviderConfig;        // Optional embedding provider settings
 }
 
 /**
@@ -222,6 +234,14 @@ export function getGlobalContext(): string | undefined {
   return config.global_context;
 }
 
+/**
+ * Get embedding provider configuration
+ */
+export function getEmbeddingConfig(): EmbeddingProviderConfig {
+  const config = loadConfig();
+  return config.embedding || {};
+}
+
 /**
  * Set global context
  */

src/collections.ts

@@ -17,6 +17,7 @@ import {
 import { homedir } from "os";
 import { join } from "path";
 import { existsSync, mkdirSync, statSync, unlinkSync, readdirSync, readFileSync, writeFileSync } from "fs";
+import { OpenAIEmbedding, type OpenAIConfig } from "./openai-llm.js";
 
 // =============================================================================
 // Embedding Formatting Functions
@@ -291,6 +292,11 @@ export interface LLM {
    */
   embed(text: string, options?: EmbedOptions): Promise<EmbeddingResult | null>;
 
+  /**
+   * Get embeddings for multiple texts in a batch
+   */
+  embedBatch(texts: string[]): Promise<(EmbeddingResult | null)[]>;
+
   /**
    * Generate text completion
    */
@@ -1206,3 +1212,47 @@ export async function disposeDefaultLlamaCpp(): Promise<void> {
     defaultLlamaCpp = null;
   }
 }
+
+// =============================================================================
+// OpenAI Embedding Support
+// =============================================================================
+
+export type EmbeddingProvider = "local" | "openai";
+
+export type EmbeddingConfig = {
+  provider: EmbeddingProvider;
+  openai?: OpenAIConfig;
+};
+
+// Default embedding config: use local llama-cpp
+let embeddingConfig: EmbeddingConfig = { provider: "local" };
+let openAIEmbedding: OpenAIEmbedding | null = null;
+
+/**
+ * Set the embedding configuration. Call before using embeddings.
+ */
+export function setEmbeddingConfig(config: EmbeddingConfig): void {
+  embeddingConfig = config;
+  // Reset OpenAI instance if config changes
+  openAIEmbedding = null;
+}
+
+/**
+ * Check if using OpenAI for embeddings/query expansion/reranking.
+ */
+export function isUsingOpenAI(): boolean {
+  return embeddingConfig.provider === "openai";
+}
+
+/**
+ * Get the provider-aware embedding/query/rerank client.
+ */
+export function getDefaultEmbeddingLLM(): LLM {
+  if (embeddingConfig.provider === "openai") {
+    if (!openAIEmbedding) {
+      openAIEmbedding = new OpenAIEmbedding(embeddingConfig.openai);
+    }
+    return openAIEmbedding;
+  }
+  return getDefaultLlamaCpp();
+}