# SWIRL AI Connect > SWIRL is zero-copy AI search. It searches every system a company already runs - Microsoft 365, SharePoint, Google Workspace, Salesforce, ServiceNow, Confluence, Jira, Snowflake, GitHub, iManage, Box, existing search indexes, databases, internal APIs and the web - live, in parallel, as the signed-in user. Nothing is copied. There is no index of the customer's content and no mirror of the customer's permissions: each source enforces its own access control at the moment of the query. SWIRL then re-scores the results on one scale, groups the versions it finds, elects the one the organization approved, and writes a cited answer from it. SWIRL is offered as open-source Community Edition (Apache 2.0), SWIRL Enterprise (self-hosted or private cloud), and OEM / source-code licensing for product builders. The commercial consequence of copying nothing: there is nothing new to bring into audit scope, no second copy of the data with its own residency, nothing extra to purge when a deletion or retention obligation lands, and no central store of everything the company knows for an attacker to take. The audience this matters most to is compliance-sensitive enterprises - legal, financial services, healthcare, government - that cannot centralize their data. SWIRL was created and open-sourced by Sid Probstein in 2023 and is developed by SWIRL Corporation, headquartered in Greater Boston. Retrieval is live rather than index-based: candidates come back from each source, then are re-ranked locally in three passes (keyword and BM25, E5-Large-V2 embeddings with hybrid RRF fusion, then an MS-MARCO cross-encoder), and an answer is generated with traceable citations using the LLM of the customer's choice, including a fully on-prem model. Because retrieval is live, standing up a vector database is not a precondition for search; the optional Semantic Cache module does use a vector store, for meaning-based recall of documents users have already retrieved with their own credentials. SWIRL 5 is API-first with a first-class MCP server, so any agent - Claude, Copilot, ChatGPT or a custom agent - calls SWIRL and gets the same ranked, permissioned, cited answers a person gets, under the same rules. SWIRL Enterprise also includes its own assistant, which answers questions over federated results and cites the governing document; customers can use SWIRL's assistant, the agent they already run, or both. The Galaxy UI ships with SWIRL and can be white-labeled. This site (swirlaiconnect.com) is the SWIRL marketing site. Product documentation lives at https://docs.swirlaiconnect.com (llms.txt: https://docs.swirlaiconnect.com/llms.txt). The legal-AI vertical site is https://swirllegal.com (llms.txt: https://swirllegal.com/llms.txt). ## What SWIRL 5 does - Searches every connected system at once, live and in place, with no ETL and no ingestion pipeline. - Enforces permissions at the source: SSO over OAuth2 and OpenID Connect signs the user in, and per-source OAuth2 passes that user's own token to each system. There is no ACL mirror to keep in sync. - Re-scores results from incompatible systems on one scale, de-duplicates them, groups versions of the same document and elects the version that governs. - Writes the answer from that document and cites it, and flags any claim the cited sources do not support. - Lets a team pin and approve the result for a query, so later searches - by a person or by an agent over MCP - return the endorsed answer. - Opens on a dashboard of workspaces and topics, with recent files from every connected OAuth2 source. - Ships a business console with AI-Yield analytics, boost/block/spotlight controls, deduplication and the separately licensed Semantic Cache module. - Runs inside the customer's own boundary: on-prem, private VPC, their cloud tenant, Docker Compose, Kubernetes, a one-click private deployment, or the Azure Marketplace. ## Product - [Home - Search everything. Copy nothing.](https://swirlaiconnect.com/): Zero-copy AI search for enterprises that can't centralize their data. Live answers from every system you already run, as each signed-in user, with permissions enforced at the source. - [Under the Hood](https://swirlaiconnect.com/under-the-hood): The technical build, for CIOs, architects and security reviewers. What is dispatched rather than ingested, how permissions are decided by each source per user per query, where SWIRL runs, and what the vendor can see (nothing). - [Connectors](https://swirlaiconnect.com/connectors): 150+ SearchProviders that query the source live, as the signed-in user - enterprise platforms (M365, SharePoint, Google Workspace, Salesforce, ServiceNow, Confluence, Jira, Box, iManage, NetDocuments, M-Files), search engines, databases (Snowflake, PostgreSQL, Elasticsearch, OpenSearch, Solr, BigQuery), AI models and research sources. Nothing is crawled, copied or indexed. - [Token Efficiency](https://swirlaiconnect.com/token-efficiency): Why a cleaner prompt produces a better answer. SWIRL sends the version the organization approved instead of five near-identical copies, with a source-checked study of what each stack actually hands the model. - [Pricing](https://swirlaiconnect.com/pricing): An annual platform license in three editions (Departmental $24K, Business $72K, Enterprise $150K+), a creditable paid pilot, add-on modules including Semantic Cache, and OEM / embed licensing from $60K per year. Priced by scope, not per seat. - [SWIRL for Embedding](https://swirlaiconnect.com/swirl-for-embedding): Embed federated search and grounded RAG in your own product, across your customers' silos, inside the tenant where their data already lives. REST API, MCP server, or the white-label Galaxy UI. - [SWIRL for Backstage](https://swirlaiconnect.com/backstage): A search engine module for Backstage. A Tantivy index inside SWIRL gives the catalog and TechDocs real BM25 relevance, n-gram title matching, typo tolerance and attribute filters, while GitHub and Confluence are queried live and ranked in the same list. One container, no cluster, no JVM. Apache 2.0. - [SWIRL for Claude](https://swirlaiconnect.com/claude/): A SWIRL 5 license by email, then the free Apache-2.0 SWIRL plugin for Claude Code installs SWIRL, connects your sources, stands up cited answers and wires the MCP server, in about an hour. - [Deploy SWIRL Enterprise](https://swirlaiconnect.com/deploy/): One click deploys the complete stack into your own cloud workspace on your own subscription. Your sources, your keys, your data. - [SWIRL 5.1 Sneak Preview](https://swirlaiconnect.com/5-1-preview/): Auto-pinning with guardrails, an improvement pipeline that turns curator feedback into one-click tuning, and Quick and Thoughtful answer modes. - [Releases](https://swirlaiconnect.com/releases): Release notes for SWIRL Community and SWIRL Enterprise. - [FAQ](https://swirlaiconnect.com/faq): What zero-copy means in practice, plus editions, deployment, connectors, RAG, MCP, security and compliance, OEM embedding and support. ## Use cases and proof - [Case Studies](https://swirlaiconnect.com/case-studies): Production deployments in government, legal, pharma and higher education, and who the product is built for - legal, financial services, healthcare and public sector teams that cannot copy their data. Named production customers are RJ Kelly, eShare and Innoflight. ## Company - [History](https://swirlaiconnect.com/history): From the first Apache-2.0 release in 2023 to the zero-copy AI search running today in law firms, federal agencies and SaaS platforms. - [Team](https://swirlaiconnect.com/team): The people who build SWIRL, headquartered in Greater Boston. - [News and Announcements](https://swirlaiconnect.com/news): Company news, partnerships and press coverage. - [Contact](https://swirlaiconnect.com/contact-us/): Book a working session against a slice of your own systems, or ask about OEM licensing and deployment. ## Content - [Blog](https://swirlaiconnect.com/blog): Roughly 200 posts on federated AI search, zero-copy retrieval, agentic systems and MCP, RAG, version election, knowledge management and SWIRL product updates. Older posts still use SWIRL's earlier positioning language. - [Resources](https://swirlaiconnect.com/resources): Documentation, source on GitHub, the connector library, release notes, the resource library and support. - [Summer 2026 Newsletter](https://swirlaiconnect.com/newsletter/summer-2026): SWIRL 5, version election, MCP, the Claude Code plugin, Slack integration and one-click private deployment. ## Documentation - [SWIRL Docs](https://docs.swirlaiconnect.com/): Quick Start, User Guide, Admin Guide, Developer Guide and Reference, Installation, Kubernetes, Security, RAG, MCP, and the Microsoft 365, Google Workspace and SharePoint guides, for both Community and Enterprise. - [Docs llms.txt](https://docs.swirlaiconnect.com/llms.txt): Machine-readable index of the documentation site. ## Localization - [Japanese](https://swirlaiconnect.com/jp): Japanese-language overview of SWIRL. This page still carries SWIRL's earlier positioning and is awaiting translation of the current one. ## Optional - [Sitemap (XML)](https://swirlaiconnect.com/sitemap.xml): Full URL list for swirlaiconnect.com. - [robots.txt](https://swirlaiconnect.com/robots.txt): Crawl policy. - [Privacy Policy](https://swirlaiconnect.com/privacy-policy/) - [Terms of Service](https://swirlaiconnect.com/terms-of-service/) - [SWIRL on GitHub](https://github.com/swirlai/swirl-search): Apache 2.0 source for SWIRL Community Edition. - [SWIRL Legal](https://swirllegal.com/): Vertical site for legal AI, with traceable citations and Claude-for-Legal integration.