Mastra
TypeScript-first framework for AI agents, workflows, memory, and RAG — by the Gatsby team.
About
Mastra is an open-source TypeScript framework for building production AI agents, workflows, and retrieval pipelines, created by the team behind Gatsby. It crossed 22,000 GitHub stars within fifteen months of release and shipped v1.0 in January 2026, backed by a $13M seed round led by Lightspeed in October 2025.
Its philosophy is opinionated defaults with escape hatches: Agents, typed Tools with Zod validation, graph-based Workflows with suspend/resume and human-in-the-loop, and Observational Memory that compresses long conversations automatically. A unified model router covers thousands of models across dozens of providers, and mastra dev gives you a local playground with evals and trace viewer. Deploy to Mastra Cloud, Vercel, Cloudflare Workers, or plain Node.
Worth knowing: the core is open source but enterprise features (RBAC, auth integrations) live in a proprietary directory. A 2025 directory-traversal CVE in the MCP docs server is a reminder to keep dependencies current. For simple streaming chat UIs, the Vercel AI SDK is lighter.
What it can do
- Agents with reasoning loops, tools, and memory
- Durable workflows with branching and human-in-the-loop
- Observational memory with automatic compression
- Unified router across 3,300+ models and 94 providers
- Built-in evals, tracing, and MCP server authoring
Setup
Scaffold with npm create mastra@latest and follow the CLI wizard. Requires modern Node LTS.
Prerequisites
Node.js LTS; an LLM provider API key.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 45; Trending score 88/100 from researcher assessment.
★ 22,000 GitHub stars
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.