refactor(embedding): simplify to OpenAI + Ollama with O(1) character-based truncation

packages/embedding/generate.ts

@@ -12,6 +12,10 @@ interface EmbedOptions {
   abortSignal?: AbortSignal;
 }
 
+interface EmbedManyOptions extends EmbedOptions {
+  maxParallelCalls?: number;
+}
+
 function getEmbedOptions(config: EmbeddingConfig): EmbedOptions {
   const options: EmbedOptions = {};
 
@@ -26,6 +30,16 @@ function getEmbedOptions(config: EmbeddingConfig): EmbedOptions {
   return options;
 }
 
+function getEmbedManyOptions(config: EmbeddingConfig): EmbedManyOptions {
+  const options: EmbedManyOptions = getEmbedOptions(config);
+
+  if (config.options?.maxParallelCalls !== undefined) {
+    options.maxParallelCalls = config.options.maxParallelCalls;
+  }
+
+  return options;
+}
+
 // =============================================================================
 // Error Detection
 // =============================================================================
@@ -50,19 +64,6 @@ function isContextLengthError(error: unknown): boolean {
   return false;
 }
 
-// =============================================================================
-// Provider Options
-// =============================================================================
-
-function getProviderOptions(
-  config: EmbeddingConfig,
-): Record<string, Record<string, string>> | undefined {
-  if (config.provider === "cohere") {
-    return { cohere: { truncate: "END" } };
-  }
-  return undefined;
-}
-
 // =============================================================================
 // Single Embedding (internal, no retry)
 // =============================================================================
@@ -82,13 +83,11 @@ async function generateEmbeddingOnce(
   config: EmbeddingConfig,
 ): Promise<SingleEmbedResult> {
   const model = getEmbeddingModel(config);
-  const providerOptions = getProviderOptions(config);
 
   const result = await embed({
     model,
     value: text,
     ...getEmbedOptions(config),
-    ...(providerOptions && { providerOptions }),
   });
 
   if (result.embedding.length !== config.dimensions) {
@@ -124,8 +123,7 @@ export async function generateEmbedding(
 ): Promise<SingleEmbedResult> {
   const maxTokens = config.options?.maxTokens ?? MAX_OPENAI_TOKENS;
 
-  // Non-OpenAI providers: truncate defensively, single attempt
-  // (Cohere also uses API-side truncation as backup)
+  // Ollama: truncate defensively, single attempt
   if (config.provider !== "openai") {
     const { text: truncated } = truncateText(text, maxTokens);
     return generateEmbeddingOnce(truncated, config);
@@ -178,7 +176,6 @@ export async function generateEmbeddings(
 
   const model = getEmbeddingModel(config);
   const maxTokens = config.options?.maxTokens ?? MAX_OPENAI_TOKENS;
-  const providerOptions = getProviderOptions(config);
 
   // Pre-truncate all providers defensively
   const texts = rows.map((row) => truncateText(row.content, maxTokens).text);
@@ -190,8 +187,7 @@ export async function generateEmbeddings(
     const { embeddings, usage } = await embedMany({
       model,
       values: texts,
-      ...getEmbedOptions(config),
-      ...(providerOptions && { providerOptions }),
+      ...getEmbedManyOptions(config),
     });
 
     // embedMany returns aggregate token count — distribute evenly
@@ -250,7 +246,6 @@ export async function generateEmbeddings(
           model,
           value: text,
           ...getEmbedOptions(config),
-          ...(providerOptions && { providerOptions }),
         });
 
         if (result.embedding.length !== config.dimensions) {

packages/embedding/package.json

@@ -6,9 +6,6 @@
   "main": "index.ts",
   "dependencies": {
     "ai": "^6.0.0",
-    "@ai-sdk/openai": "^3.0.0",
-    "@ai-sdk/cohere": "^3.0.0",
-    "@ai-sdk/mistral": "^3.0.0",
-    "@ai-sdk/google": "^3.0.0"
+    "@ai-sdk/openai": "^3.0.0"
   }
 }

packages/embedding/provider.ts

@@ -1,6 +1,3 @@
-import { createCohere } from "@ai-sdk/cohere";
-import { createGoogleGenerativeAI } from "@ai-sdk/google";
-import { createMistral } from "@ai-sdk/mistral";
 import { createOpenAI } from "@ai-sdk/openai";
 import type { EmbeddingModel } from "ai";
 import type { EmbeddingConfig } from "./types";
@@ -12,11 +9,11 @@ import type { EmbeddingConfig } from "./types";
 /**
  * Get an embedding model for the configured provider.
  *
- * Supports: openai, ollama, cohere, mistral, google
+ * Supports: openai, ollama
  *
  * API key resolution:
  * 1. config.apiKey if provided
- * 2. Environment variable: {PROVIDER}_API_KEY (e.g., OPENAI_API_KEY)
+ * 2. Environment variable: OPENAI_API_KEY
  *
  * Ollama special handling:
  * - Auto-appends /v1 to baseUrl if missing
@@ -25,12 +22,9 @@ import type { EmbeddingConfig } from "./types";
 export function getEmbeddingModel(config: EmbeddingConfig): EmbeddingModel {
   const provider = config.provider.toLowerCase();
 
-  // Resolve API key from config or environment
-  const envKey = `${provider.toUpperCase()}_API_KEY`;
-  const apiKey = config.apiKey ?? process.env[envKey];
-
   switch (provider) {
     case "openai": {
+      const apiKey = config.apiKey ?? process.env.OPENAI_API_KEY;
       if (!apiKey) {
         throw new Error(
           `API key not found for OpenAI. Set apiKey in config or OPENAI_API_KEY environment variable.`,
@@ -57,40 +51,10 @@ export function getEmbeddingModel(config: EmbeddingConfig): EmbeddingModel {
       return ollama.embedding(config.model);
     }
 
-    case "cohere": {
-      if (!apiKey) {
-        throw new Error(
-          `API key not found for Cohere. Set apiKey in config or COHERE_API_KEY environment variable.`,
-        );
-      }
-      const cohere = createCohere({ apiKey });
-      return cohere.embedding(config.model);
-    }
-
-    case "mistral": {
-      if (!apiKey) {
-        throw new Error(
-          `API key not found for Mistral. Set apiKey in config or MISTRAL_API_KEY environment variable.`,
-        );
-      }
-      const mistral = createMistral({ apiKey });
-      return mistral.embedding(config.model);
-    }
-
-    case "google": {
-      if (!apiKey) {
-        throw new Error(
-          `API key not found for Google. Set apiKey in config or GOOGLE_API_KEY environment variable.`,
-        );
-      }
-      const google = createGoogleGenerativeAI({ apiKey });
-      return google.textEmbeddingModel(config.model);
-    }
-
     default:
       throw new Error(
         `Unsupported embedding provider: ${config.provider}. ` +
-          `Supported providers: openai, ollama, cohere, mistral, google`,
+          `Supported providers: openai, ollama`,
       );
   }
 }

packages/embedding/types.ts

@@ -2,12 +2,7 @@
 // Provider
 // =============================================================================
 
-export type EmbeddingProvider =
-  | "openai"
-  | "ollama"
-  | "cohere"
-  | "mistral"
-  | "google";
+export type EmbeddingProvider = "openai" | "ollama";
 
 // =============================================================================
 // Config
@@ -25,13 +20,11 @@ export interface EmbeddingConfig {
 export interface EmbeddingOptions {
   /** Max tokens per text (truncates longer inputs) */
   maxTokens?: number;
-  /** Number of texts per embedding API call */
-  batchSize?: number;
   /** Timeout per embedding API call in milliseconds */
   timeoutMs?: number;
   /** Max retries on transient failures */
   maxRetries?: number;
-  /** Max concurrent embedding API calls */
+  /** Max concurrent chunk requests when embedding many values (default: Infinity) */
   maxParallelCalls?: number;
 }