ragpipe

Pluggable TypeScript RAG toolkit — defineConfig() one file, embed → search → generate.

RAG is a composition of 3 functions: embed → search → generate. Everything else is just a plugin choice.

Features

• 4 functions to learn — ingest, search, ask, generate. That's it.
• defineConfig() driven — One config file to wire up your entire RAG pipeline.
• Plugin architecture — Mix and match embedding, vector store, and generation providers.
• CLI included — npx ragpipe ingest ./docs/ and you're done.
• TypeScript-first — Strict types from day one. No loose any.
• Tiny core — <50KB. 10x smaller than LangChain.js.

Quick Start

# Scaffold a config file interactively
npx ragpipe init

# Set up vector store schema (creates tables, indexes, etc.)
npx ragpipe setup

# Ingest your documents
npx ragpipe ingest ./docs/

# Ask questions
npx ragpipe ask "What is the refund policy?"

Configuration

Create a ragpipe.config.ts and pick your providers:

import { defineConfig } from "ragpipe";
import { geminiEmbedding, geminiGeneration } from "@ragpipe/plugin-gemini";
import { supabaseVectorStore } from "@ragpipe/plugin-supabase";

export default defineConfig({
  embedding: geminiEmbedding({
    apiKey: process.env.GEMINI_API_KEY!,
    model: "gemini-embedding-001",
  }),
  vectorStore: supabaseVectorStore({
    supabaseUrl: process.env.SUPABASE_URL!,
    supabaseKey: process.env.SUPABASE_SERVICE_ROLE_KEY!,
    tableName: "documents",
  }),
  generation: geminiGeneration({
    apiKey: process.env.GEMINI_API_KEY!,
    model: "gemini-2.5-flash",
  }),
});

Other Combinations

Bedrock Claude + Voyage + Supabase

import { defineConfig } from "ragpipe";
import { voyageEmbedding } from "@ragpipe/plugin-voyage";
import { supabaseVectorStore } from "@ragpipe/plugin-supabase";
import { bedrockGeneration } from "@ragpipe/plugin-bedrock";

export default defineConfig({
  embedding: voyageEmbedding({
    apiKey: process.env.VOYAGE_API_KEY!,
    model: "voyage-3-lite",
  }),
  vectorStore: supabaseVectorStore({
    supabaseUrl: process.env.SUPABASE_URL!,
    supabaseKey: process.env.SUPABASE_SERVICE_ROLE_KEY!,
  }),
  generation: bedrockGeneration({
    region: "us-east-1",
    model: "anthropic.claude-3-haiku-20240307-v1:0",
  }),
});

Fully Local — Ollama + SQLite (no API keys needed)

import { defineConfig } from "ragpipe";
import { ollamaEmbedding, ollamaGeneration } from "@ragpipe/plugin-ollama";
import { sqliteVectorStore } from "@ragpipe/plugin-sqlite-vec";

export default defineConfig({
  embedding: ollamaEmbedding({ model: "bge-m3" }),
  vectorStore: sqliteVectorStore({ path: "./rag.db" }),
  generation: ollamaGeneration({ model: "llama3" }),
});

Programmatic Usage

import { loadConfig, createPipeline } from "ragpipe";

const config = await loadConfig();
const rag = createPipeline(config);

// Ingest documents
await rag.ingest(markdownText, "docs/guide.md");

// Ask questions with source attribution
const { answer, sources } = await rag.ask("How does authentication work?");

Plugins

Embedding

Provider	Package	Status
Gemini	@ragpipe/plugin-gemini	stable
OpenAI	@ragpipe/plugin-openai	stable
Voyage AI	@ragpipe/plugin-voyage	stable
Ollama	@ragpipe/plugin-ollama	stable
Cloudflare Workers AI	@ragpipe/plugin-cloudflare	stable
AWS Bedrock	@ragpipe/plugin-bedrock	stable

Vector Store

Provider	Package	Status
Supabase (pgvector)	@ragpipe/plugin-supabase	stable
PostgreSQL (pgvector)	@ragpipe/plugin-pgvector	stable
SQLite (sqlite-vec)	@ragpipe/plugin-sqlite-vec	stable

Generation

Provider	Package	Status
Gemini	@ragpipe/plugin-gemini	stable
OpenAI	@ragpipe/plugin-openai	stable
AWS Bedrock (Claude)	@ragpipe/plugin-bedrock	stable
Ollama	@ragpipe/plugin-ollama	stable

How It Works

ragpipe has 3 plugin axes — just like hot-updater composes build × storage × database, ragpipe composes embedding × vectorStore × generation.

ragpipe
├── embedding     ← text → vector
├── vectorStore   ← vector storage / similarity search
├── generation    ← question + context → answer
└── chunker       ← document → chunks (built-in, replaceable)

The core pipeline is simple:

1. Setup — create vector store tables and indexes based on your config
2. Ingest — chunk documents → embed each chunk → store vectors
3. Search — embed query → find similar chunks
4. Ask — search → build context → generate answer with LLM

CLI Commands

Command	Description
npx ragpipe init	Scaffold a ragpipe.config.ts interactively
npx ragpipe setup	Set up vector store schema (tables, indexes). Use --force to recreate from scratch
npx ragpipe ingest <dir>	Chunk, embed, and store documents
npx ragpipe ask "<question>"	Search context and generate an answer

Why ragpipe?

	LangChain.js	ragpipe
Bundle size	~2MB+	<50KB core
Learning curve	Chain, Agent, Tool, Memory...	4 functions
Config	Assemble in code	Declarative defineConfig()
CLI	None	npx ragpipe init/setup/ingest/ask
TypeScript	Loose types	Strict from day one
Scope	General-purpose AI framework	RAG only — does one thing well

Contributing

Contributions are welcome! See the spec document for architecture details and roadmap.

License

MIT