Revornix

Revornix is an open-source, local-first AI information workspace built by Qingyon Technology (Hangzhou). It ingests noisy streams of links, papers, audio, and screenshots, then converts them into structured knowledge, illustrated reports, and two-voice podcasts — all delivered through automated notifications. The entire stack (web client, gateway, API, async workers, trending feed, and docs site) is self-hostable, and a managed cloud workspace is available at app.revornix.com.

What It Is

Revornix is an AI-native knowledge management platform that covers the full pipeline from collection to delivery: ingestion → conversion → summarization → graph → podcast → notification. It targets researchers, analysts, content creators, and product teams who deal with high volumes of information and need a structured, searchable, and collaborative knowledge base. The platform is model-agnostic — any OpenAI-compatible provider can be wired in — and each processing stage (parsing, embedding, summarizing, podcasting, illustration) can use an independently configured engine.

Core Architecture and Capabilities

The platform is built as a microservices stack with distinct components:

• Next.js web client — workspace UI and SEO-friendly public pages
• FastAPI backend — auth, documents, sections, AI features
• Celery workers — async embedding, summarization, graph updates, podcast generation, notifications
• Go gateway — public-entry routing, anti-scraping, upstream failover
• Hot-news service — trending aggregation from major platforms via DailyHotApi

Key capabilities include multi-format ingestion (web pages, PDF, Word, Excel, PPT, plain text, audio), pluggable converters (MinerU, Jina, custom engines), vector retrieval via Milvus, and a personalized GraphRAG layer on Neo4j. The AI assistant (Revornix AI) grounds its answers on the user's documents and personal knowledge graph.

MCP and Ecosystem Integration

Revornix implements both sides of the Model Context Protocol: as an MCP client it can drive external MCP servers, and as an MCP server it exposes the user's library to MCP-aware tools through API routes grouped into common, document, section, and graph capabilities. The platform also supports the OpenClaw skill ecosystem via the revornix-publisher skill, which lets OpenClaw agents directly operate Revornix documents, sections, labels, and publishing flows. SDKs are published for Python (PIP) and JavaScript (NPM), and a Chrome extension handles browser-based collection.

Deployment Model

Revornix is designed for self-hosting with full data sovereignty. The local dependency stack (Postgres, Redis, Neo4j, MinIO, Milvus) can be bootstrapped with a single docker-compose-local.yaml. For users who prefer not to self-host, the managed workspace at app.revornix.com (with a mainland China mirror at app.revornix.cn) provides the same feature set. The license is a modified Apache 2.0 that permits commercial use with restrictions on multi-tenant SaaS deployments and logo removal.

Update: v1.0.0-beta1

The latest release is v1.0.0-beta1, published on June 11, 2026. The project has maintained a high release cadence since its April 2025 launch, with over 25 versioned releases documented in the blog. Recent versions have added MCP server capabilities, OpenClaw skill support, knowledge graph (GraphRAG) features, podcast generation, AI illustration embedding, and a trending headlines feed. The roadmap lists an iOS app (expected March 2026) and a WeChat Mini Program (expected May 2026) as upcoming additions. The GitHub repository shows active development with regular commits to the develop branch.