SWIRL Docs
/
swirlaiconnect.com
Community Edition | Enterprise Edition

SWIRL Overview

SWIRL AI Search is a hybrid metasearch engine that connects almost any Generative AI or Large Language Model (LLM) to enterprise data platforms, applications, and information services—without copying, ingesting, or indexing any data.

SWIRL runs in your own environment, anywhere Docker or Kubernetes is supported. Users can generate personalized, secure AI-driven insights from the data they're already authorized to access—without requiring developers, lengthy legal agreements, or complex data transfer/ETL (extract, transform, load) projects.

SWIRL RAG AI Insight with results

What is SWIRL AI Search Assistant?

SWIRL AI Search Assistant is a chat bot that converses with users to identify the best data sources for an inquiry and the exact query to run. It provides in-line insights using Retrieval-Augmented Generation (RAG) on the most relevant search results.

As of SWIRL Enterprise 4.0, the AI Search Assistant can generate queries in SQL, SPARQL, and other query languages known by the underlying LLM.

SWIRL RAG AI Insight with results

How Does SWIRL Provide Insights Without Copying, Ingesting, or Indexing Data?

SWIRL AI Search asynchronously sends user queries to authorized APIs and other configured endpoints. Response time depends on the slowest source.

SWIRL then re-ranks results from all responding sources using embeddings from the configured LLM, so users don't have to.

The re-ranking process follows these steps:

  • Vectorize the user's query (or relevant parts of it).
  • Send the text of the user's query and/or the vector to each requested (or default) source.
  • Asynchronously gather results from each source.
  • Normalize results using JSONPath (or XPath).
  • Vectorize each result snippet (or relevant parts of it).
  • Re-rank results based on similarity, frequency, and position, while adjusting for factors like length variation, freshness, and more.

The Xethub study, demonstrated that re-ranking "naive" keyword search engines outperforms re-indexing data into a vector database for tasks like question answering.

SWIRL AI Search also includes cross-silo RAG to generate AI-powered insights such as summarization, question answering, and visualizations of relevant result sets.

SWIRL AI Search Insight Pipeline

When a user requests an AI-generated insight, SWIRL:

  • Sends the request to relevant sources.
  • Normalizes and unifies the retrieved results.
  • Re-ranks the unified results using a non-generative Reader LLM.
  • Optionally presents them to the user, allowing adjustments to the result set.
  • Fetches the full text of results in real time.
  • Identifies the most relevant portions of the documents and binds them into a prompt using real-time vector analysis similar to the re-ranking process above.
  • Sends the refined prompt to the approved generative AI for insight generation.
  • Returns a single set of AI-generated insights with citations.

SWIRL AI Search includes the Galaxy UI and documented Swagger APIs for integration with nearly any front-end or system.

SWIRL Enterprise supports flexible, generic OAuth2 and Single Sign-On (SSO), with auto-provisioning via OpenID Connect.

How Does SWIRL AI Search Assistant Work?

SWIRL AI Search Assistant ensures that the configured LLM only accesses data the user is authorized to see, through its integration with SWIRL AI Search.

SWIRL manages chat context and history, triggering RAG through AI Search based on user requests and conversation flow. Users can only receive insights from data they are already authorized to access. Assistant retains context only within each user's active session, keeping conversations private and contextualized per user. All access is managed through the existing SSO system.

For more information, please refer to the AI Search Assistant Guide.

What Systems Can SWIRL Integrate With?

SWIRL AI Search integrates with a wide range of systems, including enterprise applications, cloud platforms, databases, search engines, and AI services.

For a full list of supported integrations, see swirlaiconnect.com/connectors.

How Do I Connect SWIRL to a New Source?

To connect SWIRL AI Search to an internal data source, create a SearchProvider record, which defines how SWIRL interacts with that source. See the SearchProvider Guide.

To integrate SWIRL Enterprise with an LLM, create an AIProvider record, which configures SWIRL to communicate with the AI system. See the AI Search Guide.

What Is Included in SWIRL Enterprise Products?

SWIRL Enterprise includes:

  • SWIRL AI Search Assistant — an interactive AI-powered search assistant.
  • Extended set of AI providers (e.g., Anthropic, Cohere) and Assistant-specific LLM roles including chat, query rewriting, and direct answer retrieval. (As of 4.5, SWIRL Community also supports configurable AIProviders for RAG and embeddings via OpenAI, Azure/OpenAI, and any OpenAI-compatible endpoint — see the Admin Guide.)
  • SSO support with various Identity Providers (IDPs) such as Ping Federate, plus auto-provisioning via OpenID Connect. (SWIRL Community only supports M365.)
  • AI-powered insights from 1,500+ file formats, including structured data such as tables and charts, plus text in images.
  • Authentication support for PageFetcher, enabling secure retrieval of protected content.
  • Configurable prompts.

How Much Do SWIRL Enterprise Products Cost?

SWIRL Enterprise pricing varies based on deployment type, features, and support level. See swirlaiconnect.com/pricing.

When Should I Use SWIRL AI Search Community Edition?

Use SWIRL Community if you need to search across one or more repositories and apply RAG to full-text content, without requiring authentication, indexing the data into another repository, or writing additional code.

You may also freely redistribute solutions that incorporate SWIRL Community under the Apache 2.0 License.

When Should I Use SWIRL Enterprise Edition?

Use SWIRL Enterprise when you have:

  • Repositories that require authentication via SSO or OAuth2.
  • A need to extract text from documents and fetch authenticated pages securely.
  • A need to apply RAG to long documents, complex tables, or text extracted from images.
  • A need to use LLMs beyond OpenAI and Azure OpenAI.
  • A desire to interact conversationally with your data using SWIRL AI Search Assistant.

How Can I Obtain SWIRL Enterprise Edition?

How Can I Get Help With SWIRL?

Security and Compliance

For security administrators and CISOs evaluating SWIRL Enterprise, see the Security Guide for the deployment security model, identity and data protection controls, network hardening, logging, vulnerability management, and a production hardening checklist.

What Is the SWIRL Architecture and Technology Stack?

SWIRL products are built on a Python/Django/Celery/Redis stack:

  • Python & Django — core framework for application logic and API services.
  • Celery & Redis — handle asynchronous processing and task management.
  • PostgreSQL (recommended for production) — scalability and reliability.

For a deeper dive into SWIRL's architecture, see the Developer Guide.

How Is SWIRL Usually Deployed?

SWIRL is typically deployed using Docker, with Docker Compose for setup.

For SWIRL Enterprise, deployments are also available as Kubernetes images for scalable, containerized orchestration.

What is the history of SWIRL?

SWIRL was created by Sid Probstein in 2023. SWIRL was incorporated in 2024 to provide support, services and enterprise-grade versions of SWIRL under commercial license. Sid serves as SWIRL's CEO.

As of 2026, SWIRL is focused on AI Legal Search.

Ask SWIRL

Have questions about SWIRL? Just type below.