Skip to main content

Agent Framework

Overview

The Mastra Agent Framework is a transitional TypeScript/Python monorepo that composes shared platform capabilities with K12, SDAC, TAP, and Handbook domain packages. It includes the Mastra runtime, operator and product applications, ingestion and health services, SQL projects, Azure infrastructure, API contracts, tests, and GitHub Actions delivery automation.

Source ownership is intentionally split across packages/**, apps/**, src/mastra/**, and runtime-entry/**. Use the architecture pages before assuming that a legacy-looking path is inactive or that a package is the only implementation owner.

What It Does

The platform serves multiple domain workloads and shared operator surfaces:

  • K12 Safety Documentation — AI-powered editing and generation of Emergency Operation Plans (EOP) functional annexes and Quick Reference Guides (QRG) for K12 educational institutions, helping schools create and maintain critical safety documentation.

  • Content Moderation (TAP) — Analyzes text and image content for safety concerns using Azure AI Content Safety, performing automated batch tagging and content safety analysis for educational social media platforms.

  • School District Administrative Claims (SDAC) — Reviews MAA cost reports with validation tools, review coordinator agents, widget chat, feedback capture, and sendback preparation.

  • MSBA Handbook — Prepares and indexes handbook/policy PDFs, runs Handbook RAG and generation workflows, and reviews handbooks against authoritative MSBA policy plus the Student Handbook Checklist.

  • Reasoning Engine — A shared, domain-agnostic analysis pipeline that accepts data from any source format (JSON, CSV, API payloads) and produces insights in any target format through pluggable intake/output adapters.

Key Capabilities

  • Multi-Agent Orchestration — Specialized AI agents collaborate through defined workflows, each contributing domain expertise (temporal analysis, pattern recognition, correlation detection, explanation generation, validation).

  • Production Data Integration — SQL Server persistence for platform and domain state, plus Azure Storage and AI Search where the owning workload requires files, queues, artifacts, or retrieval indexes.

  • Format-Agnostic Design — Core reasoning logic remains unchanged while adapters handle source-specific data normalization and target-specific output formatting.

  • Azure Delivery System — App Service and Functions workloads, managed identity, Key Vault, SQL/DACPAC, APIM, Azure AI services, deployment slots, diagnostics, and contract-aware GitHub Actions promotion.

Getting started

Core areas

Common entry points

Browse all docs (with filters)

Use these filters to quickly narrow docs by domain and audience.

Domain
Audience