Security-focused AI agent platform Frona unveiled

Frona is a self-hosted personal AI agent platform built on a single Rust engine and a policy language. Users can deploy it directly on their own infrastructure, connect the LLM of their choice, and run autonomous agents that perform web browsing, code execution, app building, and more.

Core Security Model:

• Per-principal sandboxing: Every agent and tool runs inside an isolated Linux process, with file system and network access strictly restricted according to policy.
• Unified policy engine: A Cedar-based language manages tool permissions and isolation policies from a single point.
• Credential protection: Vault features integrated with 1Password, Bitwarden, and others ensure credentials are never exposed to agent memory or LLM traffic.
• Dual LLM dispatch: Untrusted input is routed to an isolated LLM to prevent agent hijacking attacks.

Key Features:

• Agents: Supports default agents such as Assistant, Researcher, and Developer, along with collaboration between agents.
• Tool extensibility: Supports browser automation, web search (SearXNG, Tavily), in-sandbox code execution (Python, Node.js), and voice calls via Twilio.
• Connectivity: Users can interact with agents through messaging channels such as Telegram and SMS.