Skip to main content

Product Reference Document

Status: Living Product Reference (v2)

Purpose: This document defines the core product vision for Gantral, the open-source AI Execution Control Plane.

1. The Core Problem​

Large organizations are quietling adopting AI, but lacking execution, control, and accountability.

  • Teams invent their own approval/logging mechanisms.
  • No standard for Human-in-the-Loop (HITL).
  • No clear record of "Who authorized this AI action?".

Enterprises do not lack AI tools. They lack an AI execution control plane.

2. What Gantral Is​

Gantral is an AI Execution Control Plane.

  • Infrastructure for running AI workflows with control.
  • System of Record for AI decisions.
  • Kubernetes for AI Execution (standardizes how it runs).

Gantral is NOT:

  • An agent builder (like LangChain/Vellum).
  • An observability tool (like LangFuse).
  • A "magic" autonomous dev platform.

3. Core Capabilities​

3.1 Execution Plane​

Standardizes how AI workflows run across teams.

  • Processes defined once (templates).
  • Instances provide isolated, auditable execution.
  • Configuration adapts processes per team.

3.2 Human-in-the-Loop (HITL)​

HITL is a first-class state transition, not a UI feature.

  • RUNNING → WAITING_FOR_HUMAN → APPROVED / REJECTED.
  • Captures Who, Why, When, and Context.

3.3 Materiality Assessment​

Governance framework to determine risk:

  • High Materiality: Mandatory HITL (e.g., financial decisions, prod code commits).
  • Low Materiality: Auto-approval OK (e.g., docs generation).

4. Initial Wedge: HITL for SDLC​

Focus on Code Review & Incident Management.

  1. AI Code Review: High-risk changes require human sign-off.
  2. Incident Response: AI suggests remediation; human approves execution.
  3. PR Risk Assessment: AI scores risk; high score triggers blocking approval.

5. Personas​

PersonaRoleNeeds
Platform EngineerBuyer/OwnerControl, standardization, auditability.
Engineering ManagerUserSafety, visibility into team's AI risks.
On-Call EngineerUserFaster incident resolution with safety rails.
Compliance/GRCInfluencer"Who authorized this?" audit trails.

6. Design Principles​

  1. Boring UX by Design: Clarity > Delight.
  2. Governance First: Auditability is non-negotiable.
  3. Process-Level: Orchestrate workflows, not just agents.
  4. No Magic: Humans remain accountable.