Most AI frameworks were born in Python. Teams living in JavaScript and TypeScript were left out. Mastra fixes that: it is a TS-native framework to build AI agents, workflows and RAG in production.
What is Mastra?
It brings agents, deterministic workflows, RAG, memory and observability into a single package, is model-agnostic (built on the Vercel AI SDK) and includes a local playground to test agents. It reached version 1.0 in early 2026 with hundreds of thousands of weekly downloads.
Key features
- Agents, workflows, RAG, memory and observability in one TS package
- Model-agnostic, on the Vercel AI SDK (GPT, Claude, Gemini, Llama)
- Local playground to develop and test agents
- Tool support via MCP
- Version 1.0 in Jan 2026, built by an ex-Gatsby team
How Reche uses it
Reche's default stack is TypeScript end to end, so frameworks like Mastra fit straight into the flow. Reche builds TS-native AI agents and products, with RecheOS quality gates ensuring speed does not cost quality.