Swirl Search

Swirl Search is an open-source, self-hosted web app and server. It provides federated search and retrieval-augmented generation (RAG) by querying data from external applications in real time rather than migrating it into a central vector database. The software is deployed as a server via Docker and includes a web interface for executing searches and viewing results.

Users can connect the platform to various enterprise tools to ask questions and receive answers with citations. Because it queries sources live, the system relies on the existing permissions of the user at the source level. This architecture removes the need for ETL pipelines or the maintenance of duplicate data copies for search indexing, allowing for a deployment process that takes a few minutes via a compose file.

Key features

• Federated search and RAG across over 100 connectors
• Real-time answer generation with citations using OpenAI keys
• Re-ranking of results using cosine vector similarity with spaCy and NLTK
• Microsoft 365 integration via OAuth2
• Pipelined Processor architecture for transforming queries and responses
• Synchronous and asynchronous federation through a REST API
• Result storage in SQLite or Postgres for analytics and post-processing
• Extensible Connector and Mixer objects for adding new data sources

Built with Python and Django, the system is designed for teams creating knowledge base searches across SharePoint, Confluence, and Google Drive, or building customer support assistants that draft responses grounded in documentation and tickets. It is also used for developer assistants that index GitHub and Jira. The community version provides the Galaxy UI and basic relevancy ranking, while the enterprise version adds a three-pass reranker and an MCP server for agents. The software comes with out-of-the-box support for Arxiv, European PMC, and Google News to facilitate immediate testing of the search capabilities.

Swirl Search prioritizes live data access over data ingestion and avoids the governance overhead of secondary data copies.