Everything included
Stop spending weeks configuring boilerplate. FastRAG handles the infrastructure so you can focus on what makes your product unique.
Paste any URL — docs site, blog, or landing page. Cheerio targets semantic content zones like <main> and <article>, strips boilerplate, and vectorizes clean text in seconds. Fully serverless, zero ESM conflicts on Vercel.
Drag & drop any PDF up to 20MB. Multipart upload via Formidable, full text extraction with pdf-parse, automatic chunking with configurable overlap — from dense legal contracts to research papers.
Anthropic's recommended embedding model — voyage-3.5 via LangChain. Separate document and query embedders optimise the vector space for asymmetric retrieval. First 200M tokens are completely free.
claude-haiku-4-5 streams answers token-by-token via Server-Sent Events. The first SSE event carries source citations and usage data — no loading spinners, no separate status calls.
Each user's documents live in a private Pinecone namespace. One index, unlimited tenants, zero cross-contamination. Swap "demo" for a userId and you have enterprise-grade multi-tenancy from day one.
Built-in demo protection: 5 messages and 3 ingestions per visitor per 24 hours. File-based session store — no Redis needed. Returns resetAt timestamps so the UI shows exact countdowns.
Full access to every API route under pages/api. Swap the vector DB, change the LLM, tweak prompts — everything is your code. MIT licensed, no vendor lock-in, no magic abstractions.
RecursiveCharacterTextSplitter chunking, cosine similarity threshold filtering (0.3), context injected into Claude system prompt via <context> tags, graceful fallback when retrieval finds nothing relevant.
Buy once, get every future update. Private GitHub repo access means you pull improvements the day they ship. No new version to purchase, no subscription.
Simple setup
No ML expertise required. No infrastructure headaches. Just clone, configure, and ship.
Drop in a PDF or paste any public URL. The ingestion pipeline scrapes with Cheerio, parses with pdf-parse, and chunks everything using LangChain's RecursiveCharacterTextSplitter — 1000-token chunks with 200-token overlap by default.
// pages/api/ingest-url.js// → fetch URL with 15s timeout// → load HTML into Cheerio// → remove script/style/nav/footer// → targetor first // → chunk text (1000 tokens, 200 overlap)// → embed + upsert to Pinecone namespace// pages/api/ingest-pdf.js// → parse multipart upload (max 20MB)// → extract text with pdf-parse// → chunk → embed → upsert// → return { chunksIngested, totalCharacters }
Text chunks are embedded using Voyage AI's voyage-3.5 — Anthropic's recommended model, with 200M free tokens. Vectors are stored in a Pinecone namespace scoped to the user. One index, unlimited tenants, zero cross-contamination.
// lib/embeddings.js// Separate instances for ingestion vs retrieval// inputType: 'document' → for chunks// inputType: 'query' → for user questions// lib/vector-store.js// upsertVectors(vectors, namespace)// → pinecone.index().namespace(userId).upsert()// Each tenant's vectors are fully isolated.// Swap 'demo' for userId for multi-tenancy.
When a user asks a question, FastRAG embeds it with Voyage AI, retrieves the top-5 most relevant chunks from Pinecone (above a 0.3 cosine threshold), injects context into Claude's system prompt, and streams the response token-by-token via SSE.
// pages/api/chat.js// 1. embedQuery(query) → Voyage AI query embedding// 2. querySimilar(embedding, topK=5, namespace)// 3. filter matches by score >= 0.3// 4. inject chunks into Claude system prompt// 5. stream claude-haiku-4-5 via SSE// First SSE event carries:// → sources[], hasContext, usage{}// Subsequent events carry text deltas.// Final event: { done: true, usage }
Push to Vercel with one command. Add three environment variables and you're live. The IP-based rate limiter protects your demo with zero infrastructure — no Redis, no database, just a lightweight JSON session store.
# .env.local — only 3 keys requiredANTHROPIC_API_KEY=sk-ant-...PINECONE_API_KEY=pcsk_...PINECONE_INDEX=fastragVOYAGEAI_API_KEY=pa-...# One-command deploy$ vercel --prod# ✅ https://yourapp.vercel.app# Demo rate limiting: zero extra infra# 5 messages + 3 ingestions per visitor/24h
Real-world applications
FastRAG is a foundation, not a template. Here's what you can build with it this weekend.
Let employees query your entire docs, PDFs, and internal pages in plain English. Ingest URLs with one click — no Notion API, no Confluence plugin. Just paste and ask.
Train on your help docs, FAQs, and product manuals. Claude Haiku answers tier-1 tickets instantly — 24/7, grounded in your actual content, with source citations on every response.
Upload lecture notes, textbooks, and syllabi. Students get instant, cited answers from course material instead of waiting for office hours or searching through PDFs manually.
Upload NDAs, contracts, and compliance PDFs. Lawyers and ops teams query them conversationally — FastRAG cites the exact clause it drew from, so answers are always verifiable.
Point FastRAG at your API reference, README files, and changelogs. Scrape any docs URL in seconds with Cheerio — developers find what they need without leaving their editor.
Ingest your product catalog, specs, and reviews. Build a conversational shopping assistant powered by Claude that upsells intelligently and handles product questions at scale.
Why trust this
I built FastRAG because I was tired of re-wiring the same Pinecone + LangChain plumbing for every AI project. It's the kit I wish existed — and I stand behind it.
If product is not delivered under 48 hours, you get a full refund. No questions asked.
The live demo at fastrag.live/demo runs real Claude Haiku streaming and real Pinecone retrieval — no signup required.
Use it in unlimited client and personal projects. Charge for what you build, keep all the revenue.
Simple pricing
No subscriptions. No seat fees. No surprise bills. You only pay for your own infrastructure — Anthropic, Voyage AI, Pinecone, Vercel.
Everything you need to ship your first production RAG app this weekend.
Launch price · ends soon · deploy to unlimited projects
🔒 Secure checkout via Gumroad · Instant delivery
Advanced features for teams building multi-tenant SaaS products at scale.
Launch price · ends soon · deploy to unlimited projects
🔒 Secure checkout via Gumroad · Instant delivery
Questions about which plan is right for you? Talk to us →
Common questions
Still have a question after reading? Reach out — we reply within 24 hours.
Still have a question?
Contact support