STG

Use of Generative AI

How Marula Landlord OS does — and does not — use generative AI, and who is responsible for AI agents that connect through our API.

Use of Generative AI

Last Updated: June 24, 2026

This page explains how Marula Landlord OS (“we”, “us”, or “our”) uses generative AI, and — just as importantly — where it does not. We believe a property management platform that handles your financial and tenant records should be clear and direct about this.

1. The Backend Is Deterministic — and AI Features Are Always Labeled

The core of Marula Landlord OS is deterministic software: the same input produces the same result every time. Your records, balances, reports, and the actions you take in the dashboard are computed by fixed rules — not generated, predicted, or invented by an AI model. Today, the entire backend is deterministic, and no generative AI is involved in any feature.

We may add optional, AI-assisted features in the future — for example, reading a scanned lease to pre-fill fields, or summarizing a document. If and when we do, we commit to the following:

  • Clear labeling. Any feature that uses generative AI will be explicitly marked as AI in the interface, at the point where you use it. You will always know when you are looking at AI-generated output.
  • Assistive, not authoritative. AI features will supplement your data, not silently replace it. AI-generated output is a suggestion to review — never a final source of truth — and it may be inaccurate.
  • Opt-in by nature. You choose to invoke an AI feature. We do not route your data through a generative model in the background of ordinary, deterministic operations.

We will update this page before any generative-AI feature goes live.

2. The MCP Interface Talks Only to Deterministic Systems

We offer a Model Context Protocol (MCP) interface so that you can connect your own AI agent to your account. It is important to understand what this is and what it is not.

The MCP interface is a set of deterministic tools. Each tool maps to the same API the web interface and scripts use. When a tool is called, it reads or writes your data exactly as instructed and returns a precise, predictable result. The MCP server itself contains no AI — it does not interpret intent, guess, or generate content. It validates inputs, performs the requested operation, and returns the outcome.

In other words: the “intelligence” in an MCP session lives entirely in your agent. Our side of the connection is plain, auditable software.

3. The Platform Is API-First — Agents Get the Same Behavior as Everyone Else

Marula Landlord OS is built API-first. Every operation available in the web interface is a first-class API endpoint. The UI, your own scripts, and an AI agent connected over MCP all go through the same API, with the same validation, the same permission checks, the same rate limits, and the same audit logging.

This means an AI agent has no special powers and no special exemptions. It cannot do anything you could not do yourself through the interface or a script, and it is held to the same rules. The platform behaves identically regardless of what is on the other end of the connection.

4. You Are Responsible for Your Own AI Agents

If you choose to connect an AI agent to your account through MCP, you are responsible for that agent’s inputs and outputs, and for any actions it takes on your behalf.

AI agents — including large language models — can behave in ways that are outside our control. An agent may:

  • Fabricate (“hallucinate”) information — inventing values, records, or results that were never in your data
  • Swallow or ignore errors — proceeding as if an operation succeeded when it did not, or hiding a failure from you
  • Misinterpret instructions — taking an action you did not intend, or acting on the wrong record
  • Act on untrustworthy input — following instructions embedded in data it reads rather than the instructions you actually gave

When an agent calls our API, we faithfully perform the operation it requests and return an accurate result. What the agent does with that result — how it reports it back to you, whether it acts on it correctly, and whether it tells you the truth about what happened — is entirely up to the agent and is outside our control.

Write operations require confirmation. When your agent calls a tool that creates, updates, or deletes data, the server does not execute it right away. Instead, it returns a confirmation prompt describing the proposed change, and a separate confirmation step is needed before anything in your account changes. This gives you a checkpoint to catch an unintended action — a delete aimed at the wrong record, a message to the wrong tenant — before it takes effect.

Because of this, when you use the MCP interface:

  • You are responsible for reviewing and verifying anything an agent reports, especially financial figures, balances, and changes to records
  • You are responsible for the actions your agent takes against your account
  • You should treat agent output as a starting point to confirm, not as a final source of truth, for any decision that matters

We provide accurate, deterministic tools. We cannot guarantee, and are not responsible for, the correctness of any AI agent you connect to them.

5. Questions

If you have questions about how generative AI relates to the Service:

Email: support@marulaos.com