Zero-copy AI search

Federated AI answers
for regulated firms.

SWIRL searches your existing systems live and cites its sources. Nothing is copied.

Get a demo on your systems How federated AI works ↓
Running in production at
RJ Kelly eShare Innoflight
Federated AI, defined

Federated learning trains models.
SWIRL answers questions.

Both leave your data where it is. SWIRL sends each question to the systems that hold the answer.

Federated learning SWIRL
What it does Trains one model across sites Answers a question across systems
What travels Training tasks out, model updates back The question out, ranked results back
Who decides access Each site Each system, as the signed-in user
Your content Stays at each site No index, no copy

The one exception: documents you promote into the licensed Semantic Cache are stored inside your own environment.

SWIRL 5 - Galaxy
SWIRL searching Microsoft 365 Teams, SharePoint, iManage, Box, ServiceNow, GitHub and case law at once for data retention policy; each source shows its own result count
One question, sent to every connected system at once.
Two ways in

Use SWIRL,
or build on it.

Government · Pharma · Finance · Legal

Use SWIRL at your firm.

One search across every system you run. One cited answer.

See who runs it ↓ For law firms →
OEMs · SaaS · Developers

Build on SWIRL.

The same search inside your product or agent.

Start building ↓ SWIRL for Embedding →
Who runs it

In production where data
cannot be centralized.

Federal data organization on Azure Government

ChallengeHundreds of curated sources it was legally and contractually barred from centralizing.
ResultCross-source analysis went from days to minutes. No ingestion pipelines.

Federal integrator adds search to an air-gapped CMS

ChallengeAnalysts could not search other sources inside a classified compartment.
ResultLive across GCC High and Low networks in under a day, with one engineer.

AmLaw firm replaces aging federated search

ChallengeOne aging platform across M365, SharePoint, legal publishers and matter management.
ResultPlatform retired. Legal research now takes one query.
Read all case studies →
For your security review

No copy of your content.
No mirror of your permissions.

Copy-everything search SWIRL
Your content Copied into a central index Queried live, in place
Your permissions Copied, then kept in sync Enforced by each source, per user
Audit scope One more system in scope Nothing new holds your data
Data residency A second copy, elsewhere Stays where it is
Deletion Delete, then purge the copy Delete at the source. Done.
Breach surface Everything in one place No central store to take

The one exception: documents you promote into the licensed Semantic Cache are stored inside your own environment.

See it work

One question. Every system.
One cited answer.

Search
Every system, one query.
In parallel, back in seconds.
Approve
The approved version, on top.
Grouping and pinning are the Semantic Cache add-on, licensed separately.
Cite
Every fact traced to its source.
Each claim links back to the document.
Under the hood

Follow a search
through the system.

SWIRL 5 - Architecture
The SWIRL architecture map

Forty-two components, drawn from the code. The interactive map needs a wider screen.

Open the full-screen map
Read what's under the hood → Open the full-screen map →
Agents and MCP

Agents get the same search
under the same rules.

Any MCP client gets the same cited answers, under the same permissions.

Claude Copilot ChatGPT Your own agents Any MCP client REST API
150+ sources. Any LLM.
M365
Microsoft 365
SF
Salesforce
SN
ServiceNow
GitHubGitHub
ConfluenceConfluence
JiraJira
Google DriveGoogle Drive
BigQueryBigQuery
SnowflakeSnowflake
ElasticsearchElasticsearch
PostgreSQLPostgreSQL
MongoDBMongoDB
AtlassianAtlassian
HubSpotHubSpot
AsanaAsana
iM
iManage
OR
Oracle
BoxBox
OpenSearchOpenSearch
GWS
Google Workspace
M365
Microsoft 365
SF
Salesforce
SN
ServiceNow
GitHubGitHub
ConfluenceConfluence
JiraJira
Google DriveGoogle Drive
BigQueryBigQuery
SnowflakeSnowflake
ElasticsearchElasticsearch
PostgreSQLPostgreSQL
MongoDBMongoDB
AtlassianAtlassian
HubSpotHubSpot
AsanaAsana
iM
iManage
OR
Oracle
BoxBox
OpenSearchOpenSearch
GWS
Google Workspace
OAI
OpenAI
AnthropicAnthropic
GeminiGemini
Mistral AIMistral AI
HuggingFaceHuggingFace
OllamaOllama
DeepSeekDeepSeek
AZ
Azure OpenAI
Perplexity AIPerplexity AI
AWS
AWS Bedrock
Vertex AIVertex AI
Snowflake CortexSnowflake Cortex
CO
Cohere
AlgoliaAlgolia
YT
YouTrack
RedmineRedmine
CrunchbaseCrunchbase
Internet ArchiveInternet Archive
PC
Pinecone
TS
ThoughtSpot
OAI
OpenAI
AnthropicAnthropic
GeminiGemini
Mistral AIMistral AI
HuggingFaceHuggingFace
OllamaOllama
DeepSeekDeepSeek
AZ
Azure OpenAI
Perplexity AIPerplexity AI
AWS
AWS Bedrock
Vertex AIVertex AI
Snowflake CortexSnowflake Cortex
CO
Cohere
AlgoliaAlgolia
YT
YouTrack
RedmineRedmine
CrunchbaseCrunchbase
Internet ArchiveInternet Archive
PC
Pinecone
TS
ThoughtSpot
View all 150+ connectors →

See it run on your systems.
Bring the auditor's questions.

A 30-minute working session on your own data.

Talk to Us Deploy SWIRL Enterprise privately →
Build on SWIRL

Three ways
to start.

Free · Apache-2.0

Let Claude install SWIRL for you.

The Claude Code plugin installs SWIRL on your own data in about an hour.

claude plugin marketplace add swirlai/swirl-claude-plugin
Get SWIRL 5 for Claude
Free · Apache-2.0

Fix search inside Backstage.

Ranked results across the catalog, TechDocs, GitHub and Confluence, inside the portal.

docker pull swirlai/swirl-backstage
SWIRL for Backstage
OEM · White-label

Embed SWIRL in your product.

Search and cited answers inside your customer's tenant. From $60K per year plus revenue share.

SWIRL for Embedding