1. Overview

Quivr is an open-source RAG (Retrieval-Augmented Generation) platform — a "Second Brain" that turns personal or enterprise data into an intelligent AI assistant. Users can simply upload documents and ask questions in natural language to easily search and put vast amounts of information to use.

Quivr has drawn attention from developers worldwide with 38,000+ GitHub stars, and is used by 50,000+ users and 6,000+ companies. Backed by Y Combinator, the project was founded by three lifelong French friends.

2. Why Quivr? (Pain Point & Solution)

In enterprise environments, roughly 20% of work time is spent simply searching for information. Employees repeatedly run into difficulties like these:

  • Having to ask a colleague who's on vacation for urgent information
  • The inefficiency of asking and answering the same question over and over
  • Not knowing where the information they need is located, or even whether it exists

Quivr solves this by providing an open-source AI platform that connects all of a company's tools, documents, APIs, and databases so you can converse with them. Quivr automates tasks such as:

  • Summarizing lengthy documents down to their essentials
  • Extracting actionable information from databases
  • Drafting context-aware emails automatically

3. Core Features

3.1 Opinionated RAG (Optimized RAG Workflow)

Quivr provides a pre-designed, optimized RAG workflow, so developers don't need to build a RAG pipeline from scratch. Designed with speed and efficiency at its core, it's ready for immediate use in production.

3.2 Support for All File Formats

Quivr supports a wide range of file formats, with the option to add custom parsers as needed:

  • Text files (.txt)
  • PDF documents
  • Markdown (.md)
  • Presentations (.ppt, .pptx)
  • Spreadsheets (.csv, .xlsx)
  • Word documents
  • Audio and video files

3.3 Multi-LLM Support

Quivr supports a wide range of LLMs (Large Language Models) to avoid vendor lock-in:

  • OpenAI (GPT-4, GPT-3.5)
  • Anthropic (Claude)
  • Mistral
  • Google (Gemma)
  • Groq
  • Local models (Ollama) — guarantees full data privacy

3.4 Customizable RAG Workflow

You can fine-tune the following elements through YAML configuration files:

  • Reranker model and settings
  • History depth (how much conversation context is factored in)
  • LLM temperature and max input tokens
  • Retrieval chunk size and count

3.5 Tool Integration and Web Search

Beyond static document knowledge, Quivr can connect to web search and external tools/APIs to enable dynamic information gathering and real-time intelligence.

3.6 Megaparse Integration

Megaparse, developed by the same team at QuivrHQ, is a tool for efficiently parsing large-scale documents, preprocessing thousands of files and connecting them directly to a Quivr "Brain."

3.7 Privacy and Self-Hosting

For companies and developers where data privacy matters, Quivr supports local deployment and self-hosting. Data stays under the user's control and can run entirely on-premises without any external API calls.

3.8 Tech Stack

Layer Technology Characteristics
Frontend Next.js + Vercel SSR-based, automatic deployment
Backend API FastAPI Python-based high-performance API framework
Async jobs Celery + Queue Handles large-file embedding and indexing
Vector store PGVector / FAISS High-performance semantic search
Auth/DB Supabase Open-source Firebase alternative

4. Benefits for Developers

4.1 Quick Start (Under 30 Seconds)

pip install quivr-core
# A full RAG system built in just 5 lines of code

4.2 Rich API Support

Quivr provides a RESTful API that's easy to explore and test via Swagger docs. It supports API-key-based authentication, making it easy to integrate into applications.

4.3 Extensible Architecture

  • Custom file parsers can be added
  • RAG workflow nodes can be extended
  • Vector stores can be swapped out
  • Supports a variety of embedding models (via LangChain integration)

4.4 An Active Open-Source Community

  • 38k+ GitHub stars, active contributions
  • Regular updates and feature improvements
  • Active issue triage and PR review

4.5 Improved Development Productivity

  • Abstracts RAG's complex internals so you can focus on business logic
  • YAML-based configuration lets you experiment with RAG strategies without code changes
  • A variety of example code provided (Chainlit, Streamlit integrations)

5. Installation and Usage

5.1 Installing the Python Package (Quick Start)

For the fastest way to get started, install the quivr-core package.

# Step 1: Install the package
pip install quivr-core

# Verify installation
python -c "import quivr_core; print('Quivr installed!')"

5.2 Basic Usage Example

from quivr_core import Brain

# 1. Create a Brain from documents
brain = Brain.from_files(
    name="my_smart_brain",
    file_paths=["./my_document.pdf", "./my_notes.txt"]
)

# 2. Ask the Brain a question
answer = brain.ask("Summarize the key points of this document")
print(answer.answer)

# 3. Run an interactive interface
while True:
    question = input("Question: ")
    if question.lower() == "exit":
        break
    response = brain.ask(question)
    print(f"Answer: {response.answer}")

5.3 Docker-Based Local Deployment (Self-Hosted)

If data privacy matters, or you want to use the full feature set:

# Step 1: Clone the repository
git clone https://github.com/quivrhq/quivr.git && cd quivr

# Step 2: Set up the environment
cp .env.example .env
# Enter your OPENAI_API_KEY in the .env file

# Step 3: Run with Docker
docker compose pull
docker compose up

# Step 4: Access it
# Web UI: http://localhost:3000
# API docs: http://localhost:5050/docs

5.4 Setting Up a Custom RAG Workflow

You can customize your RAG strategy with a YAML file:

# custom_rag.yaml
workflow_config:
  name: "advanced_rag"
  max_history: 10
  reranker_config:
    supplier: "cohere"
    model: "rerank-multilingual-v3.0"
    top_n: 5
  llm_config:
    max_input_tokens: 4000
    temperature: 0.3

from quivr_core import Brain
from quivr_core.config import RetrievalConfig

brain = Brain.from_files(
    name="custom_brain",
    file_paths=["./data/*.pdf"]
)

config = RetrievalConfig.from_yaml("./custom_rag.yaml")
answer = brain.ask("question", retrieval_config=config)

5.5 Building a Chat UI With Chainlit

cd examples/chatbot
rye sync
rye run chainlit run chainlit.py

5.6 Issuing and Using an API Key

# 1. Log in to the Quivr web app
# 2. Generate an API key on the /user page
# 3. Use the Bearer token when calling the API

curl -X GET https://api.quivr.app/brains/ \
  -H "Authorization: Bearer YOUR_API_KEY"

6. Understanding the "Brain" Concept

Quivr's core concept is the "Brain." A Brain is the fundamental component that stores and processes a user's knowledge.

  • A single Brain can have multiple documents linked to it
  • Each Brain can have its own unique RAG configuration and LLM
  • Can be set to public or private (shared or not)
  • Other users' Brains can also be used via the Brain Marketplace

7. References

Official Documentation

GitHub Repositories

Quick Links

  • Quick Start: https://core.quivr.com/en/stable/
  • Brain API guide: Create a Brain via the POST /brains/ endpoint
  • Chat API: Look up conversation history via GET /chat/{chat_id}/history

Community

8. Conclusion

Quivr is designed to be more than just a RAG tool — it's a developer-friendly AI framework. Prioritizing simplicity and extensibility above all, it's helping boost productivity for everyone from individual developers to large AI teams.

Reasons to choose Quivr:

  • Fully transparent and freely customizable as open source
  • 30-second install, ready to use with 5 lines of code
  • Data privacy — supports on-premises deployment
  • No vendor lock-in — compatible with all major LLMs
  • An active community with continuous updates

As one description puts it, "the idea of Obsidian, but supercharged with AI capabilities" — Quivr offers a new paradigm for knowledge management. Go build your own "second brain" with Quivr today.