> ## Documentation Index
> Fetch the complete documentation index at: https://docs.transitionzero.org/llms.txt
> Use this file to discover all available pages before exploring further.

# AMP AI Assistant

> AMP is an AI assistant designed to help you build, debug, and analyse energy system models directly within Scenario Builder.

# Changelog

<Update label="v1.2.0" description="26 May 2026">
  AMP can now:

  * Build and edit scenarios for you, without leaving the chat
  * Save and load your pending and past conversations, allowing you to resume old conversations or to multitask with AMP
  * Reason about and create constraints over [node and technology groups](https://docs.transitionzero.org/platform/editing-scenarios#group-constraints)
  * Interact with large scenarios, such as our Asean Power Grid scenario
</Update>

## Overview

AMP is an AI assistant embedded into Scenario Builder. It has access to specialised tools that allow it to interact with your scenarios: analysing inputs, debugging infeasible runs, drafting base scenarios from research questions, and applying scenario edits for your review.

## Key Features

<CardGroup cols={2}>
  <Card title="Contextual Analysis" icon="brain">
    AMP has direct access to your scenario's configuration. It can sanity-check your data and provide reasoned analysis of your results.
    <br />**Try asking:** *"Why is renewable curtailment high in the South?"*
  </Card>

  <Card title="Infeasibility Analysis" icon="wrench">
    When a model run fails or is deemed infeasible, AMP analyses your inputs for logical conflicts and suggests specific solutions.
    <br />**Try asking:** *"Why is this scenario infeasible?"*
  </Card>

  <Card title="Scenario Creation" icon="file-plus">
    AMP can create draft base scenarios from your research question or modelling brief, helping you move from idea to testable setup faster.
    <br />**Try asking:** *"Draft a long-term capacity expansion scenario for APG."*
  </Card>

  <Card title="Scenario Editing" icon="pen-to-square">
    AMP can modify scenario inputs and generate sensitivity variants, then present each edited scenario for verification or further refinement.
    <br />**Try asking:** *"Create three PV learning-rate sensitivity variants for this scenario."*
  </Card>
</CardGroup>

## User Guide

### Setting Context

To get the best results, ensure AMP is looking at the right data. Load a scenario into context in one of two ways:

<Steps>
  <Step title="Load your scenario">
    Navigate to the specific Scenario page or Results page within your workspace.
  </Step>

  <Step title="Use @ mentions">
    From anywhere in the app, click the `@` button in the chat bar to select a specific scenario from your project.
  </Step>
</Steps>

### Creating and editing scenarios

You can ask AMP a research question, describe a target outcome, or specify a modelling change, and AMP will convert that intent into draft scenarios or scenario edits.

When needed, AMP can create a new draft base scenario and/or apply updates to an existing scenario, including multiple sensitivity variants in one request.

After creating or editing scenarios, AMP always presents the outputs back to you for verification before you proceed. You can then request additional refinements in chat.

<Tip>
  **Best Practice:** Be explicit about scope (for example technologies, regions, or parameters) to get faster and more accurate scenario drafts and edits.
</Tip>

### Chat history

AMP conversations are saved automatically as you work.

You can reopen a previous chat at any time to continue a modelling thread.

### Managing Context

AMP's performance may degrade if a single conversation becomes too long or if you switch between multiple different scenarios within one chat session.

<Tip>
  **Best Practice:** Start a new chat session whenever you change topic or switch to a different scenario.
</Tip>

## Limitations & Constraints

It is important to understand how AMP interacts with your data to interpret its answers correctly.

<AccordionGroup>
  <Accordion title="Knowledge cut-off">
    AMP's general world knowledge is cut off as of **January 2025**. However, it can retrieve current data from your specific scenario context or the up-to-date Scenario Builder documentation.
  </Accordion>

  <Accordion title="No web search">
    AMP cannot browse the live internet. It relies strictly on its training data, the provided documentation, and the scenario data you have loaded.
  </Accordion>

  <Accordion title="Math capabilities">
    Like all Large Language Models (LLMs), AMP can make mistakes with complex mental arithmetic.

    * **Recommendation:** Use AMP to *set up* the model inputs, and let Scenario Builder's engine perform the calculation. Always verify precise numerical figures in the results table.
  </Accordion>

  <Accordion title="Numerical precision">
    AMP sees your scenario data at reduced numerical precision (fewer decimal places) to save context space. For most analysis this is sufficient, but you may notice small rounding discrepancies compared to your raw data exports.
  </Accordion>

  <Accordion title="Result subsets">
    For performance reasons, AMP does not read every data point in hourly model results. Instead, it analyses aggregated annual statistics and a sample of representative days.
  </Accordion>
</AccordionGroup>

<Warning>
  **AI Reliability**

  We use advanced prompting and retrieval-augmented generation (RAG) to ground AMP in your data, but AI models can occasionally hallucinate.

  * **Verification:** Always verify critical model inputs and results in the Scenario Builder tables and charts.
  * **Edits:** Review all drafted and edited scenarios to ensure they match your intent.
  * **Advice:** AMP provides modelling assistance, not financial or investment advice.
</Warning>

## Usage Limits

AMP is subject to daily token usage limits. Users running heavy queries against very large scenarios may reach these limits.

Please contact [support@transitionzero.org](mailto:support@transitionzero.org) if you require increased usage limits for enterprise projects.

## Data Privacy & Security

<Info>
  We take the security of your proprietary modelling data seriously.

  * **Data Isolation:** Your scenario data is sent to the LLM (Google Vertex AI) only when you actively use the assistant with a scenario loaded.
  * **Chat History Storage:** Saved AMP chat history is stored securely within your TransitionZero workspace.
  * **No Training on User Data:** Neither TransitionZero nor our LLM providers use your private scenario data or chat history to train public AI models.
  * **Authentication:** All AI requests are protected by your standard TransitionZero user session.
  * **Cloud Security:** The backend operates within a secure, enterprise-grade Google Cloud environment.

  For more details, see the [Data Privacy FAQs](/platform/data-privacy-faqs).
</Info>