Player not loading? Watch on YouTube
This walkthrough introduces Mastra's TypeScript framework and the platform services used to operate agents. An agent definition combines a model, tools, and system instructions, with Studio providing a local development environment for testing. The speaker explains how specialized agents can work under a supervisor or through a workflow, and how the model router lets each agent use a different model.
The example is a travel AI agent that uses RAG to answer airline policy questions and books flights after traveler approval. The speaker then discusses preparation for production: guardrails intended to block prompt injection and limit token usage, built-in evaluation scorers for tool call accuracy, and custom scoring with an LLM judge.
According to the walkthrough, the framework captures logs and traces for agent runs and tracks inference spending. Mastra Observability uses ClickHouse to store exported telemetry, with trace filters and dashboards for cost and performance trends. The platform exporter sends telemetry to Mastra Platform even when agents deploy elsewhere.
Mastra Server is presented as a managed deployment option that returns an authenticated API. Studio can also deploy as a shared workspace for prompt versioning, datasets built from traces, and measured experiments. The video covers local testing and deployment choices, but does not establish support for offline model inference or give hardware requirements.
