# Interface Concepts
The Interface Layer defines how humans interact with AI Fabrix–powered systems.
It is intentionally decoupled from orchestration logic and the Dataplane to ensure security, scalability, and interface flexibility.
The Interface Layer is not where integrations run, data is fetched, or policies are enforced.
Its responsibility is interaction, context capture, and identity propagation.
Interface vs Orchestration vs Dataplane
AI Fabrix enforces a strict separation of responsibilities across three layers:
| Layer | Responsibility | What it does not do |
|---|---|---|
| Interface | Human interaction, context input, response presentation | No integration logic, no data access |
| Orchestration | Agent logic, reasoning, tool selection | No policy enforcement, no raw system access |
| Dataplane | Data retrieval, normalization, RBAC/ABAC enforcement | No UI, no conversational state |
Key rule
Interfaces only call orchestration endpoints or MCP tools — never Dataplane APIs directly.
Identity and Context Propagation
Every interface interaction carries user identity and execution context into the system.
- User authenticates in the interface (e.g. Entra ID, Slack, Teams)
- Identity is validated by Miso
- A delegated execution token is issued
- Orchestration and Dataplane receive user identity, groups, roles, and environment context
This enables:
- RBAC enforcement at operation level
- ABAC filtering at data level
- Full auditability of user actions
- Consistent access behavior across interfaces
Human-in-the-Loop Patterns
Interfaces are the anchor point for human-in-the-loop workflows:
- Approval before write operations
- Review of AI-generated actions
- Validation of sensitive changes
- Case escalation to humans
Humans interact through interfaces.
Agents execute through orchestration.
Policies are enforced in the Dataplane.
Architectural Context
The diagram below illustrates the supported execution flow and responsibility boundaries described in this article.
