Agent-User
Security model

Security model

/docs/security

Renderer is designed to avoid direct access to raw Node API, raw fs, model keys, native API, raw knowledge data, and long-lived tokens.

01

Principle 1

nodeIntegration: false / contextIsolation: true / preload contextBridge

02

Principle 2

Strict script CSP (self only)

03

Principle 3

External file selection goes through Electron-mediated user choice + task workspace import

04

Principle 4

Model keys, high-value prompts, and sensitive routing policy are not hardcoded in the client

05

Principle 5

Medium/high-risk Skills require explicit confirmation; current-task Full Access grants still constrained by Runtime policy

06

Principle 6

Enterprise knowledge / RAGFlow access governed by identity, tenant policy, short-lived token, audit, and Runtime-only data keys