Security & Trust

Your team's data stays in your team. Agents can only see what you can see. Private messages stay private.

How It Works

Teams in Fide form a hierarchy (a tree). This tree defines trust boundaries:

• Within your team tree: Members can access resources in their team and sub-teams
• Across different trees: Complete isolation—no data flows between separate trees

This isn't just a policy—it's enforced at the database level, so even if there's a bug in the application, your data stays secure.

The Problem

Connecting autonomous agents to your business data creates two major risks:

1. Context Leakage: A "smart" agent might retrieve sensitive HR documents to answer a coding question if the RAG system is flat.
2. Permission Drift: As agents are added, managing execution permissions via role-based lists becomes impossible to audit.

We needed a way to let agents "know everything" about their job without knowing anything about secrets they shouldn't see.

The Solution

Fide fuses Organization Structure with Trust Boundaries. Instead of maintaining a separate Access Control List (ACL), the shape of your team is your security model.

Tree = Trust Boundary

Every Team in Fide acts as a secure container. This container forms a Trust Boundary that dictates what humans and agents can see.

The 3 Rules of Trust:

1. Downward Visibility: Context flows down. (e.g., Engineering can see Frontend tasks).
2. Strict Isolation: Sibling trees are invisible to each other. (e.g., Marketing cannot see Engineering context).
3. Private by Default: Direct messages are visible only to participants, regardless of hierarchy.

Agent-Specific Permissions

Agents are powerful, but they operate under a Principle of Least Privilege:

Architecture Visualization

The hierarchy defines the data flow. Root teams isolate entire organizations or projects.

graph TD
    Root["Acme Corp (Root)"] --> Eng["Engineering"]
    Root --> Mkt["Marketing"]
    Eng --> Fe["Frontend"]
    Eng --> Be["Backend"]
    
    style Root fill:#f9f,stroke:#333
    style Eng fill:#bbf,stroke:#333
    style Mkt fill:#bfb,stroke:#333
    
    subgraph "Trust Boundary A (Engineering)"
    Eng
    Fe
    Be
    end
    
    subgraph "Trust Boundary B (Marketing)"
    Mkt
    end

Data Privacy

Your data protects your competitive advantage.

• No Training: Fide does not train global models on your private team data.
• Encryption: Data is encrypted at rest (AES-256) and in transit (TLS 1.3).
• Audit Logs: Every agent action, tool call, and memory retrieval is logged for full accountability.

Related