@ragpipe/plugin-openai

@ragpipe/plugin-openai

OpenAI embedding and generation plugin for ragpipe.

Install

pnpm add ragpipe @ragpipe/plugin-openai

Usage

import { defineConfig } from "ragpipe";
import { openaiEmbedding, openaiGeneration } from "@ragpipe/plugin-openai";

export default defineConfig({
  embedding: openaiEmbedding({
    apiKey: process.env.OPENAI_API_KEY ?? "",
    model: "text-embedding-3-small",
  }),
  generation: openaiGeneration({
    apiKey: process.env.OPENAI_API_KEY ?? "",
    model: "gpt-4o-mini",
    systemPrompt: "Answer based on the provided context.",
  }),
  // ... vectorStore
});

API

openaiEmbedding(options)

Returns an EmbeddingPlugin that calls the OpenAI Embeddings API.

Option	Type	Default	Description
apiKey	string	—	OpenAI API key (required)
model	string	"text-embedding-3-small"	Embedding model name
dimensions	number	model default (1536 / 3072)	Output dimensions (text-embedding-3-* only)
baseUrl	string	"https://api.openai.com/v1"	API base URL

• Rate limit: 200ms between calls (built-in)
• Batch support: embedMany() sends array input in a single API call
• Dimension reduction: text-embedding-3-small/large support custom dimensions to reduce output size
• Compatible APIs: Set baseUrl to use OpenRouter, Azure OpenAI, or other OpenAI-compatible endpoints

openaiGeneration(options)

Returns a GenerationPlugin that calls the OpenAI Chat Completions API.

Option	Type	Default	Description
apiKey	string	—	OpenAI API key (required)
model	string	"gpt-4o-mini"	Chat model name
systemPrompt	string	"Answer based on the provided context."	Default system instruction
baseUrl	string	"https://api.openai.com/v1"	API base URL

• Streaming: generateStream() returns an AsyncIterable<string> via SSE
• History: Pass { history } to include conversation context
• Per-call override: Pass { systemPrompt } at call time to override the default
• Compatible APIs: Set baseUrl to use OpenRouter, Azure OpenAI, or other OpenAI-compatible endpoints

Supported Models

Embedding

Model	Dimensions	Notes
text-embedding-3-small	1536	Default, cost-efficient
text-embedding-3-large	3072	High precision
text-embedding-ada-002	1536	Legacy

Generation

Model	Notes
gpt-4o-mini	Default, cost-efficient
gpt-4o	Flagship
o1	Reasoning
o3-mini	Reasoning (lightweight)

Get an API Key

• Go to OpenAI Platform
• Create an API key
• Set it as OPENAI_API_KEY in your environment

License

MIT