Mastra

Mastra is a TypeScript-native framework for building AI agents, designed to take teams from npm create mastra@latest to production without leaving the TypeScript ecosystem. According to the project's own documentation, the framework provides four core primitives — Agents, Workflows, Tools, and Observational Memory — scaffolded together through a single CLI command.

Getting started. The create mastra CLI bootstraps a complete project with example agents, workflows, and tools. Developers supply an API key from a supported model provider and can immediately interact with their agent through Mastra Studio, a web-based prototyping interface that eliminates the need to build a custom UI during early development. The CLI supports non-interactive runs via the --default flag and template selection via --template.

Model routing. Mastra's unified model router claims support for more than 1,000 models across providers including OpenAI, Anthropic, and Google. The framework auto-detects provider API keys from environment variables, reducing configuration overhead. Each provider is configured through standard environment variables such as OPENAI_API_KEY or ANTHROPIC_API_KEY.

Workflows and memory. Beyond single-turn agent interactions, Mastra supports durable multi-step workflows with typed control flow. Its Observational Memory primitive persists conversation context and application state across sessions, enabling agents that maintain coherence over long-running tasks. Tools are defined through a structured createTool() factory requiring id, description, zod-based input schema, and an execute function — plain object definitions are explicitly rejected.

Agent Builder for teams. A distinguishing feature is the Agent Builder, which Mastra describes as an interface for non-developer team members — product managers, operations staff, and support personnel — to assemble tools, models, and workflows into agents they can test and publish. This departs from code-only frameworks in the AI Agent Development space, where agent creation typically requires TypeScript or Python proficiency.

AI-assisted development. Mastra packages ship with embedded documentation in dist/docs, including SKILL.md files, reference documentation, and source maps linking public exports to their locations in node_modules. The project also publishes an llms.txt index, making its documentation programmatically accessible to AI coding assistants. This design choice means developers using AI-assisted tools can receive framework-accurate suggestions without relying on stale training data.

Deployment model. Mastra projects can be self-hosted for free. The enterprise tier, priced as a custom flat annual fee, adds role-based access control, SSO, IAM integration, and the guarantee that no traces, prompts, or outputs leave the customer's VPC. This self-host-first posture contrasts with cloud-only agent platforms. For programmatic or headless usage, agents expose an agent.generate() method that runs without Studio.

Alternatives in adjacent niches include Genspark.ai for search-augmented agent workflows and Girikon.AI for enterprise AI orchestration with a different deployment model.