TypeScript: AI SDK & Mastra

The TypeScript ecosystem has two main options depending on the complexity of the use case:

	AI SDK (Vercel)	Mastra
Best for	Streaming, structured output, tool use in web/API apps	Complex agents, workflows, persistent memory
API surface	Small, very focused	Broader, opinionated
React integration	Native hooks (`useChat`, `useCompletion`)	No (backend-first)
Multi-provider	Yes (OpenAI, Anthropic, Google, Mistral…)	Yes (via AI SDK internally)
Workflows	No (basic primitives)	Yes (typed step graph)
Memory	Manual	Built-in

For most Aircury TypeScript projects: start with AI SDK, migrate to Mastra when you need multi-step workflows or persistent memory across sessions.

AI SDK (Vercel)

The reference SDK for AI in TypeScript. Covers the most common use cases with a minimal, consistent API: text, streaming, structured output, and tool use with unified multi-provider support.

Covers the following cases well:

Basic and streaming calls to any provider
Typed structured output with Zod schemas
Tool use with automatic tool call loop
React hooks for conversational interfaces (useChat, useCompletion)
MCP server integration

Official docs: sdk.vercel.ai/docs

Mastra

Opinionated TypeScript framework for building AI agents and workflows. Where AI SDK is a low-level library, Mastra is a full platform with agents, persistent memory, graph-based workflows, and built-in observability.

Covers the following cases well:

Agents with typed instructions and tools
Short-term and long-term memory across sessions
Multi-step workflows with explicit branches and dependencies
MCP server integration
Telemetry and tracing via OpenTelemetry

Official docs: mastra.ai/docs

When to use each

Use AI SDK if you have a React/Next.js app, need streaming to the client, or your use case is relatively simple. Use Mastra if you’re building agents with memory across sessions, have multi-step workflows, or are managing multiple specialized agents.