Mastra is an opinionated TypeScript framework for building AI applications, offering primitives for agents, workflows, RAG, and evaluations. It targets developers seeking to rapidly develop and deploy AI-powered features, supporting any LLM provider through a unified interface.

How It Works

Mastra leverages the Vercel AI SDK for LLM interoperability, enabling seamless integration with providers like OpenAI, Anthropic, and Google Gemini. Its core components include: Agents that orchestrate LLM actions using tools and workflows; Tools as typed, executable functions with schema validation and integration access; Workflows as durable, graph-based state machines supporting complex logic like loops and error handling; RAG for building knowledge bases via ETL pipelines; and Integrations for auto-generated, type-safe API clients.

Quick Start & Requirements

• Install via npx create-mastra@latest.
• Run with npm run dev.
• Requires Node.js (v20.0+).
• LLM provider API keys (OpenAI, Anthropic, Google Gemini, Groq, Cerebras) are necessary.
• Environment variables ANTHROPIC_API_KEY or GOOGLE_GENERATIVE_AI_API_KEY may be required.
• Official docs: https://docs.mastra.ai/

Highlighted Details

• Unified interface for any LLM provider via Vercel AI SDK.
• Agents can call custom functions, third-party APIs, and access knowledge bases.
• Workflows feature built-in OpenTelemetry tracing.
• Evals provide automated, model-graded testing for LLM outputs.

Maintenance & Community

• Active community support via Discord: https://discord.gg/mastra

Licensing & Compatibility

• License: MIT.
• Compatible with commercial and closed-source applications.

Limitations & Caveats

The framework is opinionated, which may require adaptation for highly custom architectures. Contribution guidelines suggest discussing code contributions via issues before submitting pull requests.