Roycept AI Platform Guide

1. Supported Frameworks & Agent Roles

Roycept AI uses the provider configured by your administrator. Its conversational agent works through your request with project tools:

• PlannerInspects workspace files and creates a structured task plan and dependency list.
• BuilderCreates and edits files via write_file, edit_file, and install_dependencies.
• DebuggerCaptures compiler/build errors with run_build and inspect_logs for bounded automated repair.
• ReviewerVerifies preview health, syncs the PostgreSQL file index, and records a point-in-time recovery snapshot.

Supported runtimes: Vite + React, Node.js + Express, and Static HTML/CSS/JS.

2. Sandbox Isolation & Security

Every project executes in a dedicated non-root container (UID 1000) with dropped Linux capabilities (cap-drop=ALL), no-new-privileges, a read-only root filesystem, PID limits, and an isolated network namespace.

Outbound network traffic is routed exclusively through an SSRF-filtering egress proxy that blocks private IPv4/IPv6 ranges, loopback, and cloud metadata (169.254.169.254). Live previews are served on a separate subdomain origin so generated student apps can never access Roycept authentication cookies.

3. Student Free Tier Limits & Auto-Suspension