Skip to main content
Starfort
Starfort is an AI governance platform by AIM Intelligence — an intelligence control solution that lets an organization govern both how its people use AI and how the AI itself behaves, exactly as the organization intends. Security, regulatory compliance, and quality management all sit on top of this single control model. Control is carried out by modules, one per kind of inspection. The Guardian module inspects the text content flowing to and from AI services in real time and enforces your organization’s policies — masking sensitive data, blocking disallowed topics, and recording every decision for audit. The Stratum module, added in v1.4, classifies files (documents) into Confidential / Internal / General grades so that only documents cleared for external transfer leave for external generative AI. The two modules coexist in a project as peer Module Instances.

​
Why Starfort: the AI risk hierarchy

Starfort exists because AI risk is not one undifferentiated problem. It ranks into three tiers, and each tier is handled differently — this ordering is what justifies “enforce centrally, optimize locally.” This maps onto a two-team responsibility split: the Security team owns AI Security risk, the AI Governance team owns Service/Quality risk, and the two jointly own Regulatory risk. Starfort is built to support exactly this collaboration — the most critical policies are enforced from the center, while teams keep the freedom to tune for quality.

​
How Starfort is used

There are three entry points, each providing a different level of control. They share the same Guardian engine and policies.

API — service level

Internal/external services (backends, chatbots) embed guardrails by calling the Guard API before/after their model.

Desktop Agent — device level

Employees use everyday AI tools while the Desktop Agent (Windows) transparently enforces policy on the endpoint.

Proxy Server — infrastructure level

AI calls made at an external service’s server-side entry point are routed through Starfort.
Governance for all three is configured by an Account Admin in the Console (cloud.starfort.io): organizations, Guardians, Guard Policies, keys, and monitoring.

​
What Guardian does to a request

Every piece of content is evaluated against your Guard Policies and assigned an action. The action set depends on the policy type:

PASS

Content is allowed through unchanged.

MASK

Sensitive spans are replaced with tokens (e.g. [PHONE_NUMBER_1]) — PII policies.

BLOCK

The request is stopped before it reaches the AI service.
PII policies resolve to PASS / MASK / BLOCK; Topic verdicts are 2-state and resolve to PASS / BLOCK. See Actions: PASS / MASK / BLOCK for details. In a project that uses Stratum, files go to Stratum instead of Guardian. Rather than an action, Stratum assigns each document a grade (Confidential / Internal / General) and, based on the grade and its confidence, routes external transfer to auto-finalized / gray zone (human review) / blocked.

​
Core ideas

Guardian

The engine that inspects content and decides the action.

Stratum

The module that grades files and governs external transfer.

Guard Policy

The PII and Topic rules a Guardian enforces.

Module Instances

The two axes — Project Type × module — plus entitlements and the Kill Switch.

Organization hierarchy

Company › Organization › Project, and how Guardians attach to projects.

Glossary

Definitions for every Starfort term used in these docs.

​
What’s new in v1.4

  • Stratum — the file (document) grading module — classifies documents as Confidential / Internal / General, produces a masked copy and re-grades it, and routes external transfer to auto-finalized / gray zone / blocked based on grade and confidence. Gray-zone and Confidential verdicts go to human secondary review. See Stratum.
  • The module instance model — generalizes the “Project Guardians inside a project” resource model into module instances. Project Guardian and Project Stratum coexist as peer instances, module usage rights are governed by entitlements, and emergency stops work per module instance via the Kill Switch. See Module Instances.
  • Processed-content egress — the content that leaves for external AI services is now the processed content Guardian verified and rebuilt, never the caller’s original. Even on PASS, the original doesn’t go out directly. See Actions: PASS / MASK / BLOCK.
This documentation covers Starfort v1.4. Product names — Starfort, Guardian, Stratum, Guard Policy, Rule Package, Taxonomy, Guard API, Opticon, Control Profile — are kept in English across all languages.