Mastra

Mastra is a TypeScript framework for building AI agents and agent-powered applications. Its bet is that agent development belongs in the same language and the same repository as the product it ships inside, rather than in a separate Python service — so it runs on Node.js and drops into React, Next.js, Hono, Express or Fastify projects directly.

What the framework gives you

• Agents that reason about a goal and call tools to reach it
• Workflows as explicit graphs, composed with .then(), .branch() and .parallel(), for the paths you want deterministic rather than model-decided
• Memory covering conversation history, semantic recall and observational memory
• Tools and RAG for grounding agents in your own data and APIs
• Observability built in — evals, metrics, traces, datasets and scoring, so agent quality is measured rather than eyeballed

Model access is routed through a unified layer spanning 90+ providers, OpenAI, Anthropic, Google and xAI among them, so swapping a model is a configuration change.

Studio

Mastra Studio is a local interface at localhost:4111 for building, testing and inspecting agents — you can watch a workflow graph execute step by step and see what each agent actually did, which is the part that is painful to reconstruct from logs alone.

Factory

Mastra Factory, the newest piece, turns the framework on software development itself: specialised agents triage issues, plan and write code, validate it, open pull requests, update documentation and monitor production. It connects to a GitHub organisation and repository, optionally a Linear project, and a model provider, then runs work across a shared board with explicit gates — intake, triage, planning, building, review, completion — where humans answer triage questions, approve or edit plans, and review the pull requests. Factory is open source.

Licensing and cost

The core framework is Apache 2.0 open source and free to start, with no seats or usage tiers; enterprise features sit under a source-available commercial licence. Mastra Cloud is an optional managed deployment, and agents can equally be deployed as standalone servers, embedded in a Next.js app, or shipped to Vercel, Netlify or Cloudflare.

Best suited to TypeScript teams who want agents as a normal part of their application codebase — versioned, tested and observable — rather than as a separate system to operate.