// the find
mastra-ai/mastra
From the team behind Gatsby, Mastra is a framework for building AI-powered applications and agents with a modern TypeScript stack.
Mastra is a TypeScript framework for building AI agents and workflows, from the team that built Gatsby. It wraps model routing, memory, RAG, evals, and MCP server authoring into a single opinionated package targeting Node/Next.js developers. Think LangChain but TypeScript-first with an actual dev experience.
1. The workflow engine's suspend/resume with persistent state is the genuinely useful part — pausing mid-execution and resuming indefinitely is hard to bolt on later, and Mastra makes it a first-class primitive. 2. Single model routing interface over 40+ providers means you're not rewriting glue code every time a new model drops. 3. The eval system is built-in rather than an afterthought — most frameworks ship observability but skip structured evals, so having both in the same package matters for teams that actually need to regress-test agent behavior. 4. The monorepo is active: the changeset volume and breadth of storage adapters (DuckDB, Spanner, Redis, LibSQL, CouchBase...) suggests real adoption driving real requirements.
1. The dual-license model is a trap: the `ee/` directories are source-available, not open source, and 'freely usable for development' means you might build on enterprise features locally and only discover the license wall when you try to ship. 2. The package count in the changeset directory is enormous — this is a large, fast-moving dependency surface in a space where breaking changes are constant; pinning will be painful. 3. Mastra sits on top of the Vercel AI SDK for model routing, which means you're inheriting its abstractions and its bugs, and the abstraction occasionally leaks in confusing ways. 4. Documentation and API stability are still clearly in flux; the `.changeset/pre.json` file indicates they're in a prerelease cycle, so the API you write against today may not be the API you have in six months.