AI with Enterprise Grade Security.

What Liminal Does

Liminal is an enterprise AI platform that sits between your people and the world’s best AI models. Every prompt passes through Liminal first, where sensitive data is detected and protected before it ever reaches a model provider — then restored in the response so the work stays useful.

The platform covers the full lifecycle of enterprise AI adoption

• Discover — Find the shadow AI already in use across your organization
• Enable — Give everyone unlimited access to leading models, connected to your own data
• Protect — Shield PHI, PII, PCI, and IP on every prompt and file, in real time
• Govern — Apply policies by team and user, with a complete auditable record
• Improve — Measure adoption and ROI, and see where AI can do more

Shadow AI Detection

You can’t govern what you can’t see. Liminal’s browser extension detects AI use across your organization — including personal accounts of popular assistants — and shows usage by tool, department, and user, with sensitive data exposure flagged. Then it gives employees a sanctioned alternative that’s better than what they were sneaking around with.

Unlimited Multi-Model Access

No single model is best at everything. Liminal includes unlimited use of the latest models from OpenAI, Anthropic, Google, Perplexity, and xAI, plus support for 400,000+ other models and your own fine-tuned or privately hosted ones (AWS Bedrock, SageMaker, Vertex AI, Azure OpenAI, self-hosted endpoints). Administrators decide which models each team can use and for which data.

Connected Data

Ground every answer in your own sources. Through MCP and Enterprise Search, Liminal connects to thousands of systems — Salesforce, ServiceNow, SharePoint, Google Drive, Box, Confluence, and virtually any system that exposes endpoints. Existing permissions are respected, every answer cites its source, and your search index stays inside your single-tenant environment.

Data Protection

Liminal detects PHI, PII, PCI, and intellectual property across more than two dozen built-in data types (plus your own custom terms like client names or trade secrets) in real time. Sensitive values are redacted or masked before any prompt leaves your environment, then rehydrated in the response. Every detection and policy decision is logged.

Governance and Observability

Connect your identity provider via SAML 2.0 SSO (Microsoft Entra ID, Okta, Ping Identity), set role-based access controls with custom roles, and monitor activity in real time with streaming logs and custom alerts. Every interaction is recorded — who used which model, when, and what was protected — and logs export on demand or on schedule (CSV, JSON, or XML, or automatic delivery to AWS S3, Azure Blob, or Google Cloud Storage every 15 minutes) for your SIEM or BI tools.

Insights and ROI

The Insights Engine shows adoption, usage, and topic trends as they happen, while Smart Nudges surface high-value opportunities, workflow gaps, and training needs — turning a pilot into an organization-wide program with measurable impact.

Integrations

• MCP (Model Context Protocol): Connect AI to virtually any system that makes its endpoints available — business applications and internal tools alike
• Enterprise Search: Document repositories including SharePoint, Google Drive, Box, and Confluence, with source permissions preserved
• Identity: SAML 2.0 SSO with any compliant IdP
• Security stack: Scheduled log exports feed your SIEM; Liminal’s detected AI tool list works alongside Zscaler, Netskope, or DNS filtering for blocking
• Model providers: OpenAI, Anthropic, Google, Perplexity, xAI, plus bring-your-own models on Bedrock, SageMaker, Vertex AI, Azure OpenAI, or self-hosted endpoints

The Liminal Workspace: Browser Extension and SDK

• Browser extension rolls out through your existing browser management tools, detecting shadow AI use across your organization and redirecting employees to the governed workspace
• Web workspace gives every employee a fast, familiar chat interface with the model picker, custom assistants, and connected data — with security that stays out of the way
• Liminal SDK brings the same real-time data protection (detection, redaction, masking, rehydration) to the AI applications your developers build in-house

Custom Assistants

Turn repeatable work into shareable tools. Build custom assistants with their own instructions and preferred model, and share them with colleagues — so your organization’s best prompts become everyone’s starting point. A model-agnostic assistant can even pick the right model for each task automatically.

Security, Certifications, and Standards

Liminal is built to the standards of regulated industries:

• Single-tenant architecture — every customer runs in a dedicated AWS environment with isolated services, data, search indexes, logs, and encryption keys
• SOC 2 Type 2 certified (three consecutive independent audits) — ISO 27001 in progress
• HIPAA / HITRUST compliant with business associate agreements available (including BAAs with Anthropic, OpenAI, and Google)
• FERPA, COPPA, and CIPA compliant for education; built to support GLBA (banking), CJIS (government), GxP (life sciences), GDPR, and CCPA
• US, UK, and EU data residency
• No training on your data, ever — detection models are trained only on synthetic data
• Encryption in transit and at rest; AES-256 prompt logs; KMS-isolated keys per customer
• Annual third-party penetration testing, a formal incident response plan with 72-hour breach notification, and full data deletion within 30–90 days of offboarding

Predictable Cost

One platform replaces per-seat AI subscriptions — at 40 to 60% lower cost than ChatGPT Enterprise, Claude Team, or Microsoft Copilot, with no token limits or metering on included models.

Liminal Demo

See how Liminal works, how it protects, and keeps your organization productive and secure.