- Home
- AI Enterprise Search Software
- SWIRL AI Search
SWIRL AI Search
Waltham, Massachusetts, United States
What it is
SWIRL AI Search provides federated search across enterprise systems, querying source systems as each signed-in user and applying permissions at the source. The maker says it does not copy, index, or move source content; retrieved text is held for the request. Its three-pass ranking combines keyword/BM25, E5-Large-V2 embedding re-ranking with hybrid fusion, and an MS-MARCO cross-encoder. SWIRL can produce cited answers and flag claims unsupported by retrieved material. The connector list includes more than 150 options, such as Microsoft 365, Google Workspace, Salesforce, ServiceNow, GitHub, Confluence, Jira, and Box. It supports more than 100 LLMs, including on-prem models through Ollama and compatible frameworks. Enterprise offers a REST API and MCP server for agent access to ranked, permissioned answers. Deployment options include on-prem, a private VPC, or a cloud tenant using Docker Compose or Kubernetes. Departmental licensing costs $24,000 per year; Business costs $72,000 per year and Enterprise costs $150,000 per year. The Semantic Cache is separately licensed and is an exception to the no-vector-store approach.
For and against
For
- Searches source systems as signed-in users.
- Preserves source permissions through user identity.
- Generates cited answers and flags unsupported claims.
- Offers more than 150 listed connectors.
- Can deploy on-prem, in a private VPC, or cloud tenant.
Against
- Departmental licensing costs $24,000 per year.
- Semantic Cache is separately licensed.
- Compliance depends on customer configuration and surrounding controls.
The verdict
SWIRL focuses on permission-aware search over enterprise sources and cited answers, with several deployment options. Its annual license costs and separately licensed Semantic Cache are important considerations for prospective organizations.
Who it suits
SWIRL may suit enterprises that need search across systems without centralizing source data. The maker identifies lawyers, analysts, and clinicians as examples of users who rely on grounded AI answers.
Plans and prices
Paid pilot
Paid evaluation pilot
Read 30 Sep 2026 · swirlaiconnect.com
Departmental
One business unit/workspace · core connector pack · SSO · MCP server + REST API
Read 30 Sep 2026 · swirlaiconnect.com
OEM & Source Code License
Annual platform fee · white-label · source-code option · on-prem deployment
Read 30 Sep 2026 · swirlaiconnect.com
Business
Multiple business units/workspaces · full 150+ connector library · business console · production agent/MCP access
Read 30 Sep 2026 · swirlaiconnect.com
Enterprise
Unlimited business units · FedRAMP/air-gapped/gov-cloud option · on-prem or private-cloud deployment · premium SLA + named technical account manager
Read 30 Sep 2026 · swirlaiconnect.com
On the record
- Agent access
- The Enterprise product exposes a REST API and first-class MCP server for agents to access the same ranked, permissioned, cited answers.swirlaiconnect.com · read 30 Sep 2026
- Answers
- SWIRL generates cited answers from retrieved sources and can flag claims that are not supported by the retrieved material.swirlaiconnect.com · read 30 Sep 2026
- Audience
- The maker positions SWIRL for enterprises that cannot centralize their data and identifies lawyers, analysts, and clinicians as examples of users relying on grounded AI answers.swirlaiconnect.com · read 30 Sep 2026
- Compliance
- The maker says customers configure SWIRL and their surrounding controls to meet their regulatory obligations; it offers configuration guidance and, under NDA, audit-support artifacts.swirlaiconnect.com · read 30 Sep 2026
- Connectors
- The maker lists more than 150 connectors, including Microsoft 365, Google Workspace, Salesforce, ServiceNow, GitHub, Confluence, Jira, Box, and databases.swirlaiconnect.com · read 30 Sep 2026
- Deployment
- SWIRL can run inside a customer's environment on-prem, in a private VPC, or in a cloud tenant, using Docker Compose or Kubernetes.swirlaiconnect.com · read 30 Sep 2026
- Limits
- The optional Semantic Cache is separately licensed and is the exception to the no-vector-store approach; its caching is based on documents users have already retrieved with their credentials.swirlaiconnect.com · read 30 Sep 2026
- LLMs
- SWIRL supports more than 100 LLMs, including OpenAI, Azure OpenAI, Anthropic Claude, Google Gemini, Mistral, and on-prem models through Ollama and compatible frameworks.swirlaiconnect.com · read 30 Sep 2026
- PII controls
- Enterprise can use Microsoft Presidio to detect and redact personal information, with redaction enabled per SearchProvider.swirlaiconnect.com · read 30 Sep 2026
- Product
- SWIRL provides federated AI search across enterprise systems, querying live as each signed-in user with permissions enforced at the source.swirlaiconnect.com · read 30 Sep 2026
- Ranking
- Its three-pass ranking uses keyword/BM25, E5-Large-V2 embedding re-ranking with hybrid fusion, and an MS-MARCO cross-encoder.swirlaiconnect.com · read 30 Sep 2026
- Security
- SWIRL Enterprise supports SSO through OAuth2/OpenID Connect, and the maker says source ACLs are preserved by passing through the user's identity.swirlaiconnect.com · read 30 Sep 2026
- Support
- Departmental includes community and email support, Business includes a standard SLA and priority support, and Enterprise includes a premium SLA and named technical account manager.swirlaiconnect.com · read 30 Sep 2026
- Zero-copy retrieval
- The maker says searches run against source systems without copying, indexing, or moving source content; retrieved text is held for the request duration.swirlaiconnect.com · read 30 Sep 2026
