config.ts•6.61 kB
/**
* Embedding Configuration Schema and Utilities
*
* This module provides Zod schemas for validating embedding configurations
* and utilities for parsing configurations from environment variables.
*/
import { z } from 'zod';
/**
* OpenAI embedding configuration schema
*/
export const OpenAIEmbeddingConfigSchema = z.object({
type: z.literal('openai'),
apiKey: z.string().optional(),
model: z
.enum(['text-embedding-3-small', 'text-embedding-3-large', 'text-embedding-ada-002'])
.default('text-embedding-3-small'),
baseUrl: z.string().optional(),
organization: z.string().optional(),
dimensions: z.number().optional(),
timeout: z.number().default(30000),
maxRetries: z.number().default(3),
});
/**
* Gemini embedding configuration schema
*/
export const GeminiEmbeddingConfigSchema = z.object({
type: z.literal('gemini'),
apiKey: z.string().optional(),
model: z.enum(['gemini-embedding-001', 'text-embedding-004']).default('gemini-embedding-001'),
baseUrl: z.string().optional(),
timeout: z.number().default(30000),
maxRetries: z.number().default(3),
});
/**
* Ollama embedding configuration schema
*/
export const OllamaEmbeddingConfigSchema = z.object({
type: z.literal('ollama'),
baseUrl: z.string().default('http://localhost:11434'),
model: z.string().default('nomic-embed-text'),
timeout: z.number().default(30000),
maxRetries: z.number().default(3),
});
/**
* Voyage embedding configuration schema (for Claude/Anthropic fallback)
*/
export const VoyageEmbeddingConfigSchema = z.object({
type: z.literal('voyage'),
apiKey: z.string().optional(),
model: z.enum(['voyage-3-large', 'voyage-3', 'voyage-2']).default('voyage-3-large'),
baseUrl: z.string().optional(),
timeout: z.number().default(30000),
maxRetries: z.number().default(3),
});
/**
* Qwen embedding configuration schema
*/
export const QwenEmbeddingConfigSchema = z.object({
type: z.literal('qwen'),
apiKey: z.string().optional(),
model: z.enum(['text-embedding-v3']).default('text-embedding-v3'),
baseUrl: z.string().optional(),
dimensions: z
.number()
.refine(val => [1024, 768, 512].includes(val))
.default(1024),
timeout: z.number().default(30000),
maxRetries: z.number().default(3),
});
/**
* AWS Bedrock embedding configuration schema
*/
export const AWSBedrockEmbeddingConfigSchema = z.object({
type: z.literal('aws-bedrock'),
model: z
.enum(['amazon.titan-embed-text-v2:0', 'cohere.embed-english-v3'])
.default('amazon.titan-embed-text-v2:0'),
region: z.string().optional(),
accessKeyId: z.string().optional(),
secretAccessKey: z.string().optional(),
sessionToken: z.string().optional(),
dimensions: z
.number()
.refine(val => [1024, 512, 256].includes(val))
.default(1024),
timeout: z.number().default(30000),
maxRetries: z.number().default(3),
});
/**
* LM Studio embedding configuration schema
*/
export const LMStudioEmbeddingConfigSchema = z.object({
type: z.literal('lmstudio'),
baseUrl: z.string().default('http://localhost:1234/v1'),
model: z.string().default('nomic-embed-text-v1.5'),
dimensions: z.number().optional(),
timeout: z.number().default(30000),
maxRetries: z.number().default(3),
});
/**
* Main embedding configuration schema
*/
export const EmbeddingConfigSchema = z.union([
OpenAIEmbeddingConfigSchema,
GeminiEmbeddingConfigSchema,
OllamaEmbeddingConfigSchema,
VoyageEmbeddingConfigSchema,
QwenEmbeddingConfigSchema,
AWSBedrockEmbeddingConfigSchema,
LMStudioEmbeddingConfigSchema,
]);
/**
* Environment configuration type
*/
export interface EmbeddingEnvConfig {
type: string;
apiKey?: string | undefined;
baseUrl?: string | undefined;
model?: string | undefined;
region?: string | undefined;
accessKeyId?: string | undefined;
secretAccessKey?: string | undefined;
sessionToken?: string | undefined;
}
/**
* Parse embedding configuration from environment variables
*/
export function parseEmbeddingConfigFromEnv(): EmbeddingEnvConfig | null {
// Priority order: OpenAI > Gemini > Ollama
if (process.env.OPENAI_API_KEY) {
return {
type: 'openai',
apiKey: process.env.OPENAI_API_KEY,
baseUrl: process.env.OPENAI_BASE_URL,
model: process.env.OPENAI_EMBEDDING_MODEL || 'text-embedding-3-small',
};
}
if (process.env.GEMINI_API_KEY) {
return {
type: 'gemini',
apiKey: process.env.GEMINI_API_KEY,
baseUrl: process.env.GEMINI_BASE_URL,
model: process.env.GEMINI_EMBEDDING_MODEL || 'gemini-embedding-001',
};
}
if (process.env.QWEN_API_KEY || process.env.DASHSCOPE_API_KEY) {
return {
type: 'qwen',
apiKey: process.env.QWEN_API_KEY || process.env.DASHSCOPE_API_KEY,
baseUrl: process.env.QWEN_BASE_URL,
model: process.env.QWEN_EMBEDDING_MODEL || 'text-embedding-v3',
};
}
if (process.env.AWS_ACCESS_KEY_ID && process.env.AWS_SECRET_ACCESS_KEY) {
return {
type: 'aws-bedrock',
region: process.env.AWS_DEFAULT_REGION || 'us-east-1',
accessKeyId: process.env.AWS_ACCESS_KEY_ID,
secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
sessionToken: process.env.AWS_SESSION_TOKEN,
model: process.env.AWS_BEDROCK_EMBEDDING_MODEL || 'amazon.titan-embed-text-v2:0',
};
}
if (process.env.OLLAMA_BASE_URL) {
return {
type: 'ollama',
baseUrl: process.env.OLLAMA_BASE_URL,
model: process.env.OLLAMA_EMBEDDING_MODEL || 'nomic-embed-text',
};
}
if (process.env.LMSTUDIO_BASE_URL) {
return {
type: 'lmstudio',
baseUrl: process.env.LMSTUDIO_BASE_URL,
model: process.env.LMSTUDIO_EMBEDDING_MODEL || 'nomic-embed-text-v1.5',
};
}
return null;
}
/**
* Parse and validate embedding configuration
*/
export function parseEmbeddingConfig(config: unknown): z.infer<typeof EmbeddingConfigSchema> {
return EmbeddingConfigSchema.parse(config);
}
/**
* Validate embedding configuration
*/
export function validateEmbeddingConfig(config: unknown): boolean {
try {
EmbeddingConfigSchema.parse(config);
return true;
} catch {
return false;
}
}
// Export types
export type OpenAIEmbeddingConfig = z.infer<typeof OpenAIEmbeddingConfigSchema>;
export type GeminiEmbeddingConfig = z.infer<typeof GeminiEmbeddingConfigSchema>;
export type OllamaEmbeddingConfig = z.infer<typeof OllamaEmbeddingConfigSchema>;
export type VoyageEmbeddingConfig = z.infer<typeof VoyageEmbeddingConfigSchema>;
export type QwenEmbeddingConfig = z.infer<typeof QwenEmbeddingConfigSchema>;
export type AWSBedrockEmbeddingConfig = z.infer<typeof AWSBedrockEmbeddingConfigSchema>;
export type LMStudioEmbeddingConfig = z.infer<typeof LMStudioEmbeddingConfigSchema>;
export type EmbeddingConfig = z.infer<typeof EmbeddingConfigSchema>;