ragpipe

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

Install

pnpm add ragpipe

Quick Start

CLI

# Scaffold a ragpipe.config.ts
npx ragpipe init

# Ingest documents
npx ragpipe ingest ./docs

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

Programmatic

import { loadConfig, createPipeline } from "ragpipe";

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

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

const result = await rag.ask("How does authentication work?");
console.log(result.answer);
console.log(result.sources.map((s) => s.source));

Configuration

Create a ragpipe.config.ts at your project root:

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 ?? "",
  }),
  vectorStore: supabaseVectorStore({
    supabaseUrl: process.env.SUPABASE_URL ?? "",
    supabaseKey: process.env.SUPABASE_SERVICE_ROLE_KEY ?? "",
  }),
  generation: geminiGeneration({
    apiKey: process.env.GEMINI_API_KEY ?? "",
  }),
});

API

defineConfig(config)

Identity helper that provides type-safe autocompletion for your config file.

loadConfig(overrides?)

Loads ragpipe.config.ts from the project root using c12. Validates that embedding, vectorStore, and generation plugins are present.

createPipeline(config)

Returns a pipeline with three methods:

| Method | Description | |---|---| | ingest(text, source) | Chunk text, embed each chunk, and store vectors. Returns chunk count. | | search(query, topK?) | Embed the query and return the top-K matching documents. | | ask(query, topK?) | Search for context, then generate an answer. Returns { answer, sources }. |

defaultChunker(options?)

Built-in paragraph-based chunker. Options: chunkSize (default 500), overlap (default 50).

createRateLimitedEmbedder(plugin)

Wraps an EmbeddingPlugin with throttling based on its rateLimit.delayMs.

Plugin Interfaces

Implement any of these to create a custom plugin:

• EmbeddingPlugin — embed(text), optional embedMany(texts), rateLimit
• VectorStorePlugin — search(vector, topK), upsert(source, content, vector), optional clear(), disconnect()
• GenerationPlugin — generate(question, context, options?), optional generateStream()
• ChunkerPlugin — chunk(text, source)

Official Plugins

| Package | Description | |---|---| | @ragpipe/plugin-gemini | Google Gemini embedding + generation | | @ragpipe/plugin-supabase | Supabase pgvector store |

License

MIT