hyparProduction RAG you can read like a tutorial.

Open-source Nuxt 3 + pgvector reference for teams shipping RAG — and a learning track for devs starting from zero.

Click to copy

What is hypar?

RAG (Retrieval-Augmented Generation) lets an AI answer questions from your own documents — but production RAG is hard: chunking strategy, hybrid retrieval, durable ingestion, citations, evals. hypar is a fully-working TypeScript reference app. Read the code like a tutorial, run it locally in minutes, adapt the patterns to your stack.

Devs learning RAGTeams shipping RAGResearchers benchmarking

Try these in the demo:

What is HyDE and when is it useful?Compare BM25 and vector search Explain the ingestion workflow

Pre-loaded with the hypar docs — ask anything.

Run locally

git clone https://github.com/albegosu/hypar.git
cd hypar
cp .env.example .env
# Edit .env — set GOOGLE_API_KEY at minimum
docker compose --profile full up -d --build
open http://localhost:3000

git clone https://github.com/albegosu/hypar.git
cd hypar
pnpm install
cp .env.example .env
# Edit .env — set DATABASE_URL + GOOGLE_API_KEY
pnpm db:migrate
pnpm dev

Full setup guide →

Nuxt 3 · port 3000

Vue 3 — pages + components

Nitro — API routes (h3)

Workflow SDK — durable ingestupload / reprocess

PostgreSQL + pgvectorvector(768) · HNSW · BM25

Gemini · OpenAI · Ollamaembed · generate

Library

Drop-in patterns — hybrid search + MMR + HyDE, durable ingest, streaming citations. Production-ready today.

Tutorial

VitePress docs as a full RAG-from-zero curriculum. Small named files, every step documented. EN + ES.

Portfolio

Built in the open — 10-stage roadmap, admin dashboard, runtime settings, Vitest coverage. v0.4 live today.

What's inside

Hybrid retrieval

pgvector cosine similarity combined with BM25, then MMR diversification.

Multi-provider embeddings

Gemini, OpenAI or Ollama. Switch at runtime without re-ingesting.

Durable ingestion

Workflow SDK with per-step retries; status polled, never lost on restart.

Vitest suite

Chunking, text helpers, agent commands, and search utilities covered in CI.

Source citations

Inline [1], [2] markers persisted on every assistant message and audited.

Rate limits and admin APIs

30/min chat, 10/min upload. /api/admin/* accepts a signed-in session or ADMIN_API_KEY.

Roadmap — 10 stages

View full roadmap →

01 · Now

Traceability

02 · Next

Measurable quality

SOTA retrieval

Robustness

Pluggable

Beyond text

Real memory

Knowledge graph

Self-hosted

Live product

Learning RAG from zero?

Start with the getting-started guide and feature chapters.

Open the guide →

Want to contribute?

Browse open issues on GitHub — good first issues are tagged and ready to pick up.

See open issues →