Table of Contents

SWIRL Overview

What is SWIRL AI Connect?

SWIRL AI Connect is infrastructure software that connects most any Generative AI/LLM to enterprise data platforms, applications and information services without copying, indexing and/or ingesting data.

Installed in your own environment, or optionally hosted by SWIRL, our no-code, no-copying approach requires minimal IT involvement, generating personalized, AI insight from existing systems and applications.

How does SWIRL AI Connect Provide Insight without Copying and Ingesting and Indexing Data?

SWIRL AI Connect is an AI-powered metasearch engine that sends user requests to all the endpoints, APIs and other interfaces they're authorized to see - asynchronously, in a few seconds. SWIRL's unique Reader LLM re-ranks results from responding sources so the user doesn't have to. The Reader LLM can use the embeddings from most any model.

The re-ranking process is roughly the following:

  • Vectorize the user's query (or parts of it)
  • Send the text of the user's query and/or the vector, to each source requested (or default)
  • Asynchronously gather the results from each source
  • Normalize the results from each source using jsonpath (or xpath)
  • Vectorize each result snippet (or parts of it)
  • Re-rank the results by aggregating the similarity, frequency and position, and adjusting for other factors like length variation, freshness, etc

The Xethub study as explained by Simson Garfinkel showed that re-ranking so-called "naive" search engines like those that use the BM25 algorithm for retrieval, outperforms moving the data into a vector database for many common NLP tasks such as question answering.

SWIRL AI Connect also includes state-of-the-art cross-silo Retrieval Augmented Generation (RAG) for generating AI insights like summarization, question answering and visualization of relevant result sets.

SWIRL AI Connect Insight Pipeline

When a user requests an AI insight, SWIRL:

  • Sends the insight out to relevant sources
  • Normalizes and unifies the results
  • Re-ranks the united results using non-generative Reader LLM
  • Optionally, presents them to the user and allows them to adjust the result set
  • Fetches the full-text of the results, in real time
  • Identifies the most relevant portions of the documents and binding them to a prompt using real-time vector analysis similar to the re-ranking described above
  • Sends the prompt to the approved generative AI for insight generation
  • Returns a single set of insights with citations

SWIRL AI Connect includes the Galaxy UI, but includes fully Swagger'd APIs and is easy to integrate with most any front-end or system.

SWIRL AI Connect ENTERPRISE includes flexible, generic OAUTH2 and SSO, with auto-provisioning via OpenID Connect.

What do SWIRL AI Connect insights look like?

Here is an example:

SWIRL RAG AI Insight with results

For more information please refer to the AI Connect Guide.

What systems can SWIRL AI Connect integrate with?

The full list is here: https://swirlaiconnect.com/connectors

How do I connect SWIRL AI Connect to some new source?

To connect SWIRL with an internal data source, you create a SearchProvider record.

To integrate SWIRL Enterprise with a generative AI, you create an AIProvider record, as described in the AI Connect Guide.

What is SWIRL AI Co-Pilot?

SWIRL AI Co-Pilot turns any sufficiently capable Generative AI/LLM into an AI assistant that converses with users to determine what they are looking for and where they are most likely to find it. The Co-Pilot can search on the user's behalf and provide in-line RAG results whenever the conversation demands it.

How does SWIRL AI Co-Pilot work?

SWIRL AI Co-Pilot educates the designated GAI/LLM about the user and what they have access to via the integration with AI Connect. SWIRL manages the context and history for each chat, initiating RAG through AI Connect as directed by the user and Co-Pilot. The user can only see insight from data they are already authorized to see, and the Co-Pilot is privy only to each user's conversations and history when conversing with them. All access is controlled and provisioned via the existing sign on (SSO) system.

What does a conversation with your data via SWIRL AI Co-Pilot look like?

Here is an example:

SWIRL Co-Pilot image

For more information please refer to the AI Co-Pilot Guide.

What is included in SWIRL Enterprise Products?

SWIRL AI Connect Enterprise includes:

  • Configurable support for many enterprise AI providers (e.g. Anthropic and Cohere), including support for multiple GAI/LLMs in different roles - chat, query rewriting, direct answer, RAG and embeddings (for re-ranking/passage detection by the Reader LLM)

  • Support for Single Sign On (SSO) with various IDPs (e.g. Ping Federate) and autoprovisioning via OpenID Connect. (The Community version only supports M365.)

  • Support for generating AI insights from 1,500 different file formats, including tables and txt in images

  • Authentication support for the PageFetcher

  • Configurable prompts, including role, user, group and on-the-fly selection

SWIRL AI Co-Pilot is only available in Enterprise edition, and is not open source. Co-Pilot requires SWIRL AI Connect.

How much do SWIRL Enterprise Products cost?

Pricing for SWIRL Enterprise is here: https://swirlaiconnect.com/pricing

When should I use SWIRL AI Connect, Community Edition?

Use SWIRL AI Connect, Community Edition, if you have one or more repositories that you want to search and RAG against the full text without authenticating and/or indexing it into yet-another repository and/or writing more code.

Note that you may freely re-distribute solutions that incorporate SWIRL AI Connect, Community Edition, under the Apache 2.0 License.

When should I use SWIRL AI Connect, Enterprise Edition?

Use the Enterprise Edition of SWIRL AI Connect when you have:

  • Repositories that require Single Sign On (SSO) and/or OAUTH2
  • Content that requires text extraction, and/or authenticated page fetching
  • The need to RAG from long documents, complex tables, or text from images
  • Need to use GAI/LLMs other than OpenAI/Azure OpenAI

Why don't you use GitHub Issues?

We prefer to use our free Slack channel for support.

What is the SWIRL Architecture & Technology Stack

SWIRL products use the Python/Django/Celery/Redis stack, with PostgreSQL recommended for production deployments.

SWIRL AI Connect Architecture diagram

How is SWIRL usually deployed?

SWIRL is usually deployed via Docker. SWIRL Enterprise products are delivered as Kubernetes images.

Does SWIRL offer hosting? How can I learn more?

Please contact SWIRL for information about hosted SWIRL.