Anthropic Claude Mastra TypeScript Azure Container Apps

What It Does

The AI Engine is the intelligence layer of the SC Platform. It takes a user's question (or in the future, a scheduled trigger), decides which data sources to query, retrieves the relevant information, and synthesizes a response โ€” all within the security boundaries of the caller's permissions.

It runs as a separate deployment from the Platform API but shares the same codebase, auth model, and integration catalog.

How It Works Today

Currently, the AI Engine powers Ask Ricky. When a mechanic asks a question:

โ†’ โ†’
AI Engine

The engine receives the question along with the caller's identity and permissions. It then autonomously decides which tools to invoke:

ToolWhen UsedData Source
Get Live Telemetry Questions about current equipment status M-Link / Enbase
Get Work Order History Questions about past maintenance or repairs NetSuite via Platform API
Search Knowledge Base Troubleshooting procedures, best practices Azure AI Search (curated via Ricky Curator)

Scope-Aware Authorization

The AI Engine doesn't have blanket access to all data. The caller's identity determines which tools are available and which data each tool can return โ€” this is the platform's scope-based tool gating in action. A mechanic asking about unit 4523 only gets data for unit 4523 โ€” and only if they're authorized to see that unit.

Technology Stack

Anthropic Claude

The large language model powering all AI interactions. Claude provides reasoning, tool-use decisions, and natural language generation. SC uses Claude's tool-use capability to let the model autonomously decide which data to retrieve.

Mastra Runtime

The agent orchestration framework. Mastra manages the conversation loop โ€” sending the user's message to Claude, executing tool calls, feeding results back, and returning the final response. It handles retries, context management, and tool registration.

On the Roadmap โ€” Proactive Agents

Today the AI Engine is reactive โ€” it waits for a user to ask a question. The next evolution is proactive agents that run on a schedule, monitoring operational and financial data, and surface findings to the right person before anyone has to ask.

The critical insight: the same tools, data connections, and security model that power Ask Ricky will power proactive agents. No new integrations required โ€” just a different trigger (scheduled instead of user-initiated).