Mastra
Open-source TypeScript framework for building agents and workflows with memory, evals and observability
Mastra is an open-source TypeScript framework for building, observing and improving AI agents, including agents that run for days. Developers define typed agents with instructions, models and tools in code, compose them into workflows and serve them from a built-in server.
Key features:
- Define typed agents with models, tools, memory and runtime behavior in one place.
- Compose durable multi-step workflows with typed steps, retries and control flow.
- Control agents through a harness with modes such as plan and build.
- Give agents message history, semantic recall and observational memory.
- Collect traces, metrics and logs, and run evals with scorers, datasets and experiments.
- Iterate in Studio and deploy agents into products or tools such as Slack.
The framework is free and Apache 2.0 licensed for self-hosting anywhere, with an enterprise license for RBAC, SSO and VPC-only data. The hosted Mastra Platform has a free Starter plan with 100K observability events and 24 CPU hours a month; Teams is $250/month, and Enterprise is custom.