Skip to content
Provara

Introduction

What is Provara?

Section titled “What is Provara?”

Provara is a secure command orchestration system that places a human-in-the-loop between AI agents and system operations. Every command an AI agent wants to execute must be queued, reviewed, and explicitly approved by a human operator before it touches your system.

AI Agent → /plan API → Pending Queue → Human Approval → Policy Check → Execute → Audit Log

The Problem

Section titled “The Problem”

AI agents are increasingly capable of performing system operations — running shell commands, modifying files, managing services. But giving an AI agent direct access to your system creates serious risks:

  • Unintended destructive operations — an agent might Remove-Item critical files
  • Privilege escalation — agents executing commands with inherited permissions
  • No accountability — no audit trail of what was executed and why
  • No guardrails — nothing preventing dangerous commands from running

How Provara Solves This

Section titled “How Provara Solves This”

Provara enforces a strict approval pipeline:

  1. Queue — AI agents submit commands via the /plan API endpoint
  2. Review — Commands appear in the pending queue for human review
  3. Policy Check — A deny-first policy engine validates commands against 26 deny rules and allow patterns
  4. Execute — Approved commands run in a restricted environment with timeout enforcement
  5. Audit — Every operation is logged with timestamp, result, duration, and exit code

Key Features

Section titled “Key Features”
FeatureDescription
Human ApprovalEvery command requires explicit human approval
Policy Engine26 deny patterns + category-based allow rules
Audit TrailJSON-line logging of all operations
Token AuthAPI authentication via bearer tokens
Rate LimitingConfigurable per-client rate limits
File JailPath validation restricts file access to project root
Real-Time EventsSSE endpoint for live command status
Agent SDKPython and PowerShell integration libraries

Architecture

Section titled “Architecture”
┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│  AI Agent   │────▶│  /plan API  │────▶│   Pending   │
│ (AutoGen,   │     │  (FastAPI)  │     │   Queue     │
│  Claude,    │     └─────────────┘     └──────┬──────┘
│  Custom)    │                                │
└─────────────┘                                ▼
┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   Audit     │◀────│   Policy    │◀────│  Human UI   │
│    Log      │     │   Engine    │     │  (Approve)  │
└─────────────┘     └─────────────┘     └─────────────┘

Tech Stack

Section titled “Tech Stack”
  • Python 3.12+ with type hints and Pydantic validation
  • FastAPI for the REST API server
  • Pydantic Settings for configuration management
  • PowerShell as the command execution backend
  • httpx for the Python client SDK

Next Steps

Section titled “Next Steps”