Autoflow
pingcap/autoflow is a Graph RAG based and conversational knowledge base tool built with TiDB Serverless Vector Storage. Demo: https://tidb.ai
Last verified:
What is Autoflow?
AutoFlow is an open-source Graph RAG (Knowledge Graph RAG) based conversational knowledge base tool built on top of TiDB Vector, LlamaIndex, and DSPy. It enables users to create AI-powered chatbots that answer questions based on their own documents and knowledge bases, using knowledge graph-enhanced retrieval-augmented generation for more accurate and context-aware responses.
Key features include a Perplexity-style conversational search page with an built-in website crawler that navigates official and documentation sites through sitemap URL scraping, an embeddable JavaScript snippet widget for integrating conversational search into websites, support for multiple data sources including local file uploads (PDF, Markdown, DOCX, PPTX, XLSX, TXT), website pages via sitemap.xml, and specific web pages by URL. The tool includes configurable LLMs, embedding models, and optional rerankers, along with out-of-the-box chat interface, programmatic API access, and admin dashboard for knowledge base management.
AutoFlow is designed for developers, DevOps teams, technical writers, and organizations building AI-powered documentation chatbots or customer support assistants. It is particularly suitable for teams working with TiDB who want a Graph RAG solution for their knowledge bases, documentation sites, or internal documentation systems. The tool is open-source under Apache License 2.0 and has a live demo at https://tidb.ai showcasing TiDB documentation Q&A.
Autoflow pricing
Pricing model: Freemium
AutoFlow is completely free and open-source under the Apache License Version 2.0. There are no paid plans or subscription tiers. Users can self-host the tool on their own servers using Docker Compose at no cost. The live demo at https://tidb.ai is available for testing without registration. All features including Graph RAG, knowledge base management, multiple data sources, embeddable widget, and API access are included in the free open-source version.
Autoflow pros
- Open-source under Apache License 2.0 with no licensing costs
- Graph RAG (Knowledge Graph RAG) for enhanced retrieval accuracy
- Built on TiDB Vector for scalable vector storage
- Perplexity-style conversational search interface out-of-the-box
- Built-in website crawler with sitemap URL scraping support
- Embeddable JavaScript widget for website integration
- Supports multiple file formats: PDF, MD, DOCX, PPTX, XLSX, TXT
- Website data sources via sitemap.xml and direct URL
- Configurable LLM selection (can change default model)
- Configurable embedding model selection
- Optional reranker configuration for optimized search results
- Docker Compose deployment for easy self-hosting
- Programmatic API for chat integration
- Admin dashboard for knowledge base management
- Chat history storage in TiDB database
- Asynchronous document indexing process
- Knowledge graph index for GraphRAG capabilities
- Next.js frontend with Tailwind CSS and shadcn/ui design
- Active development with 2.8k GitHub stars
- Discord community for support and collaboration
Autoflow cons
- Still in early stages of development according to official warning
- Requires 4 CPU cores and 8GB RAM for Docker Compose deployment
- Knowledge graph index is mandatory (cannot be disabled)
- Vector index is forced to be enabled (no option to disable)
- Self-hosted solution requires own infrastructure and maintenance
- No native Python package yet (planned for future: pip install autoflow-ai)
- Limited to admin account for knowledge base configuration
- Data source deletion cannot be undone
- Asynchronous document deletion may leave residual data temporarily
- Website crawler limited to official and documentation sites
Frequently asked questions about Autoflow
What is AutoFlow?
AutoFlow is an open-source Graph RAG (Knowledge Graph RAG) based conversational knowledge base tool built on top of TiDB Vector, LlamaIndex, and DSPy. It enables users to create AI-powered chatbots that answer questions based on their own documents and knowledge bases using knowledge graph-enhanced retrieval-augmented generation.
How do I deploy AutoFlow?
You can deploy AutoFlow self-hosted on your server using Docker Compose. The recommended system requirements are 4 CPU cores and 8GB RAM. After deployment, you need to login to the admin dashboard to configure LLM, embedding model, and optionally reranker settings.
What data sources does AutoFlow support?
AutoFlow supports three types of data sources: local file uploads (Markdown, PDF, Microsoft Word, PowerPoint, Excel, TXT files), website pages via sitemap.xml, and specific web pages by direct URL. Each knowledge base can contain multiple data sources.
Can I customize the LLM used by AutoFlow?
Yes, you can configure and change the LLM model. Go to the Models > LLMs page in the admin dashboard to configure the LLM. The LLM is used for extracting knowledge from docs and generating responses, and you can change the default LLM to another one.
What is Graph RAG and how does it work in AutoFlow?
Graph RAG (Knowledge Graph RAG) uses knowledge graph indexing in addition to vector indexing for enhanced retrieval. AutoFlow builds both vector index (forced enabled) and knowledge graph index to enable more accurate and context-aware responses by leveraging relationships between entities in the knowledge graph.
Can I embed AutoFlow chat on my website?
Yes, AutoFlow provides an embeddable JavaScript snippet that you can copy and paste into your website. This widget, typically placed at the bottom right corner, facilitates instant responses to product-related queries and provides a conversational search window.
What is the tech stack of AutoFlow?
AutoFlow uses TiDB as the database for storing chat history, vectors, JSON, and analytics; LlamaIndex as the RAG framework; DSPy for programming foundation models; Next.js as the frontend framework; Tailwind CSS for styling; and shadcn/ui for design components.
How do I add documents to a knowledge base?
Go to the Knowledge Base page, click New Knowledge Base, configure the name, description, LLM, embedding model, and indexing methods. Then click the Data Sources tab and add a new data source by uploading local files or entering website URLs. Documents will be asynchronously indexed after adding the data source.
Is there an API to use AutoFlow programmatically?
Yes, AutoFlow provides an API to chat with users programmatically. The API documentation is available and allows programmatic integration of the chat functionality into your applications beyond the built-in web interface.
Where can I get support for AutoFlow?
You can reach out to the AutoFlow team on Discord for support. The project also has a Discussion section on GitHub where you can ask questions or provide suggestions. The project has an active community with 2.8k GitHub stars and 18 contributors.