Revornix

English | 中文文档 | 日本語ドキュメント

Reject FOMO! When facing the information stream, be lazy, leave the rest to AI!

Revornix is an open-source, local-first AI information workspace. It helps you collect fragmented inputs, turn them into structured knowledge, generate reports with images and podcast audio, and deliver the output through automated notifications.

Links

• Official site: https://revornix.com
• Environment docs: https://revornix.com/docs/environment
• Roadmap: RoadMap
• Community: Discord | WeChat | QQ

Why Revornix

• One pipeline for noisy information: from ingestion to summary, graph, podcast, and notification.
• Built for AI retrieval quality: chunking + vector storage + personalized GraphRAG.
• Open and controllable: self-host locally and keep your data under your own infra.
• Model-flexible: any provider compatible with the OpenAI API can be wired in.
• Collaboration-ready: share private/public knowledge sections and co-create with others.

How It Works

1. Collect: web pages, PDF, Word, Excel, PPT, text, APIs, library docs, and more.
2. Understand: parse and normalize with pluggable converters (MinerU, Jina, custom engines).
3. Organize: store vectors, build graph context, and keep content query-ready.
4. Deliver: generate rich documents, add illustrations/podcasts, and push notifications.

Project Structure

Revornix/
├── web/                  # Next.js frontend (user interaction + dashboard)
├── gateway/              # Go public-entry gateway (routing, anti-scraping, upstream failover)
├── api/                  # FastAPI core backend (auth, documents, sections, AI APIs)
├── celery-worker/        # Async workflows (embedding, summary, graph, podcast, notifications)
├── hot-news/             # Trending aggregation service (based on DailyHotApi)
└── docker-compose-local.yaml # Local dependency bootstrap

Core Capabilities

• Flexible ingestion: multi-format parsing with customizable engines.
• Advanced transformation: strong markdown/content conversion pipelines.
• Vector retrieval: chunk-to-vector storage for semantic search and AI context.
• Graph reasoning: personalized GraphRAG for better context precision.
• Built-in MCP: both MCP client and MCP server are supported.
• Auto podcast: generate and update podcast audio for documents/sections.
• Illustration generation: generate and embed AI images into content.
• Trending in one place: major platform hot lists via integrated DailyHotApi.
• Responsive and multilingual: available on mobile/desktop with multi-language support.
• Layered request protection: gateway-level anti-scraping and API-side rate limiting for high-risk public endpoints.

Some UI

Note: The trending headlines feature is based on DailyHotApi.

Quick Start

[!NOTE] We recommend creating isolated Python environments per service (for example with conda), because dependencies across services can conflict.

1) Clone repository

git clone git@github.com:Qingyon-AI/Revornix.git
cd Revornix

2) Start base dependencies

[!NOTE] If you already have postgres, redis, neo4j, minio, and milvus installed, you can reuse them. Otherwise use docker-compose-local.yaml with .env.local.example.

[!WARNING] If some dependencies are already running on your machine, disable the corresponding services in docker-compose-local.yaml to avoid conflicts.

cp .env.local.example .env.local
docker compose -f ./docker-compose-local.yaml --env-file .env.local up -d

3) Configure env files for microservices

cp ./web/.env.example ./web/.env
cp ./gateway/.env.example ./gateway/.env
cp ./api/.env.example ./api/.env
cp ./celery-worker/.env.example ./celery-worker/.env

Configure env values based on environment docs.

[!WARNING] For manual deployment, keep OAUTH_SECRET_KEY consistent across services, or cross-service authentication will fail.

4) Initialize required data

cd api
python -m data.milvus.create
python -m data.sql.create

5) Run API service

cd api
conda create -n api python=3.11 -y
pip install -r ./requirements.txt
fastapi run --port 8001

6) Run gateway service

cd gateway
go run ./cmd/gateway

The gateway is optional for local development, but recommended for production. It handles public routing, failover, and the first layer of anti-scraping protection before traffic reaches api/.

7) Run trending aggregation service

cd hot-news
pnpm build
pnpm start

8) Run Celery worker

cd celery-worker
conda create -n celery-worker python=3.11 -y
pip install -r ./requirements.txt
playwright install
./start-worker.sh

9) Run frontend

cd web
pnpm build
pnpm start

After all services are running, open http://localhost:3000.

Contributors