Enversal: Universal Environments for AI Agents

Enversal is a system that helps you build secure, highly controlled environments for both real and simulated AI agents. It provides an isolated, collaborative "universe" where agents can operate with physical, OS-level constraints to ensure host safety.

🚀 Key Concepts

Enversal is built around three core environmental abstractions:

Concept	Description
Isolone	A strictly isolated sandbox for a single agent. No network or external access by default. Ideal for untrusted tool execution.
Commune	A shared, collaborative environment where multiple agents work toward a common goal, featuring leader election and shared context.
Wormhole	An inter-environment gateway allowing temporary data trading and resource bidding between distinct Communes or Isolones.

🛡️ Zero-Trust Security (The Seatbelt Engine)

Unlike traditional orchestrators, Enversal presumes that AI Agents are unpredictable. It utilizes native OS security primitives to enforce constraints:

• macOS (Seatbelt): Dynamically generated Scheme profiles for kernel-level process jails.
• Linux (Landlock & seccomp): Granular file-system and syscall filtering.
• Resource Quotas: Hard limits on CPU, RAM, and network egress per agent.

🏗️ Architecture Overview

graph TD
    CLI[Enversal CLI/TUI] -->|gRPC| Daemon[Control Plane Daemon]
    Daemon -->|Orchestrates| Env[Autonomous Environments]
    
    subgraph "Daemon Layer"
        Daemon --> Registry[Runtime Registry]
        Daemon --> Cognitive[Cognitive Engine - Gemini/Ollama]
        Daemon --> Sandbox[Sandbox Executor - Seatbelt]
    end

    subgraph "Environment Layer"
        Env --> Agent1[Agent Alpha - Leader]
        Env --> Agent2[Agent Beta - Worker]
        Agent1 <-->|mTLS JSON-RPC| Agent2
        Agent1 -.->|Shared Context| VectorDB[(Vector DB)]
    end

Loading

📂 Project Structure

Enversal is a high-performance Rust workspace:

• core: Pure data models and manifest parsing (enversal.yaml).
• brain: The AI reasoning interface (mapping to Gemini/local LLMs).
• sandbox: Kernel-level execution abstractions and security policy engines.
• daemon: The tokio-based gRPC Control Plane and lifecycle orchestrator.
• cli: Terminal gateway for deploying and inspecting universes.

🛠️ Quick Start

1. Prerequisites

• Rust (Stable)
• macOS (for Seatbelt support) or Linux (for Landlock support)

2. Launch the Control Plane

Start the background daemon to listen for gRPC requests:

cargo run --bin enversal-daemon

3. Deploy an Environment

Use the CLI to spin up a universe from a blueprint:

cargo run --bin enversal-cli -- run blueprints/example_commune.yaml

📖 Documentation

For deep technical details, refer to the docs/ folder:

• Technical Architecture
• Project Specification

---

Developed by Kenmburu | Built with Rust 🦀