Pranor Decision — AI Governance Engine

Version: 2.0.0-dev
Module Path: github.com/vyuvaraj/pranor/decision
License: AGPL-3.0 (OSS) / EE

Overview

Pranor Decision provides a Governed AI execution decision layer with a 6-level veto ladder. It ensures safe and predictable AI operations.

Key Features

6-Level Priority Veto Ladder
SIMULATION Mode: Counterfactual evaluation without committing state
Fault Contracts

6-Level Priority Veto Ladder

Level	Name	Module	Hard/Soft	Effect
1	Auth	decision	Hard	DENY blocks all subsequent levels
2	Budget	decision	Hard	DENY on cost/token overflow
3	Risk	decision	Soft	APPROVE/DENY from risk signals
4	Rules	decision	Soft	APPROVE/DENY/TRANSFORM policy rules
5	Learn	learn	Soft	Advisory from ML predictor (skip on timeout)
6	Default	decision	Hard	Final ALLOW fallback

Fault Contract

Returns DENY if graph context is unavailable.
Learn level is skipped on ErrSidecarTimeout.

Types

DecisionRequest Input parameters containing context, agent info, and action intent.

DecisionResult Output containing the veto outcome, priority level hit, and metadata.

Quick Start

engine := decision.NewEngine(cfg)
ctx := context.Background()

// Standard execution
res, err := engine.Evaluate(ctx, decision.DecisionRequest{
    AgentID: "agent_88",
    Action:  "transfer_funds",
})

// Simulation mode
simRes, err := engine.Evaluate(ctx, decision.DecisionRequest{
    AgentID:   "agent_88",
    Action:    "transfer_funds",
    Simulate:  true, // Do not commit state
})

Enterprise Edition

Feature	OSS	EE
Basic 6-level ladder	✓	✓
Simulation mode	✓	✓
Advanced Risk Models	—	✓
Custom Rules Engine UI	—	✓

Pranor Documentation