Over the last year, we watched a pattern repeat itself across large enterprises: AI initiatives that looked extraordinary in controlled demos struggled once they met the realities of production environments.
It wasn’t because the models lacked power. Rather, the surrounding systems weren’t designed for security, governance, compliance, and risk.
Nowhere was this gap more visible than in legal search.
Federation has always been foundational to how we think about enterprise AI. That hasn’t changed. What has become unmistakably clear is that federation alone isn’t enough especially in legal environments where trust, traceability, and control are non-negotiable.
Legal teams aren’t asking for experimentation. They’re asking for systems that fit the way their organizations are actually built and governed. Systems that respect permissions without exception. Systems that don’t introduce new data risk. Systems whose outputs can be verified instantly.
When we stepped back and examined why SWIRL continues to resonate so strongly in legal contexts, we realized something important:
It isn’t one feature.
It isn’t one architectural decision.
It’s the interaction of twelve very specific capabilities, each shaped by real enterprise constraints:
- Private by design – Runs on premises or in a private cloud. Not hosted. No third-party data exposure.
- Permission-faithful security – Integrates with SSO and enforces existing permissions exactly as defined.
- Compliance without compromise – Requires zero data movement. No ingestion. No indexing.
- IT-owned deployment – Installed, governed, and managed like any other enterprise system.
- Configuration over custom code – No development projects required to deliver value.
- Re-ranking first – Identify the best results from responding systems instead of blindly aggregating content.
- LLM-agnostic architecture – Support for OpenAI, Google, Anthropic, and others. No lock-in.
- Fine-grained LLM control – Precise governance over exactly which documents are sent for summarization.
- Deep-linked citations – Every insight is easily traceable back to its source for immediate verification.
- Auditable and manageable – Fully observable, with dashboards and reporting designed for oversight.
- Source-ready integration – Works out of the box with iManage, Microsoft 365, Google Workspace, Box, and other critical systems.
- SharePoint-native operation – Uses existing collections and organizational structures as-is.
None of these capabilities are flashy on their own.
Together, they create something unique: AI that legal professionals can actually trust because it aligns with how their environments must operate.
This is the distinction between “interesting AI” and operational AI.
Today, we’re formalizing this focus with the launch of swirllegal.com, along with a new Azure Marketplace offer designed specifically for legal search use cases. At the same time, we continue expanding content sources and front-end adapters, including support for ChatGPT, Microsoft Copilot, and other AI interfaces enterprises are actively evaluating and deploying.
Enterprise AI doesn’t struggle because the models aren’t intelligent enough. It struggles because enterprise environments impose constraints that most AI architectures were never designed to respect.
SWIRL was built for those constraints.
Legal search is where that design choice is clearly proving its value.
And that’s where we’re leaning in.