¶ Knowledge Access Control
The Knowledge Access Control framework in Fuse governs who can view, retrieve, or traverse knowledge — ensuring all usage is scoped, explainable, and policy-compliant.
¶ Overview
Fuse applies a multi-layered access model to all knowledge operations. This ensures that:
- Resources respect user, session, and assistant scopes
- Knowledge retrieval during inference honors access rights
- Traversals and embeddings are filtered by permission context
- Audit trails link every answer to its permitted sources
¶ Access Control Principles
| Principle | Description |
|---|---|
 Resource-Centric Enforcement |
Access to knowledge is governed by the source Resource’s policies |
 Scoped Filtering |
Results from search, graph traversal, or embeddings are permission-filtered |
 Assistant Context |
AI agents are treated as users with assigned roles and scopes |
 Trust & Governance Layers |
Permissions, trust scores, and provenance work together to enforce usage boundaries |
 Audit & Explainability |
All access and injection events are logged and traceable to their source and authorizer |
¶ Enforcement Points
Access control is evaluated:
- When viewing or editing Resources
- During KnowledgeBase and page indexing
- Before traversing or querying graph entities
- When retrieving or scoring embeddings
- At the point of prompt composition
- In any usage by assistants, workflows, or external systems
¶ Supported Mechanisms
ClientMethodroles and permissionsfnGetOperationsand operation-level filtersResourceAccessRuleexpressions- Token-based and domain-scoped access layers
- Assistant-bound identity evaluation (per session or task)
¶ Access Dimensions
Fuse supports filtering by:
- User Identity
- Assistant Role or Task
- Tenant or Org Scope
- Token Claims and Expiration
- Trust Score Thresholds
- KnowledgeBase Visibility
- Page Sensitivity or Classification
These dimensions allow flexible and layered control, even in multi-tenant or regulated deployments.
¶ Runtime Enforcement
- Access is checked before a Resource or node is injected into a prompt
- Traversals stop when links or nodes are not visible to the user
- Embedding-based retrieval filters out vectors tied to inaccessible content
- Prompt composers automatically apply all relevant constraints
¶ Future Enhancements
- Visual access policy simulator
- Resource trust inheritance graph
- Scoped override tools for assistant use
- Policy-aware summarization based on user role
Next: Runtime Usage & Injection — Explore how Fuse injects relevant knowledge into workflows, prompts, and runtime operations based on access, context, and orchestration policies.