Mastra
Open-source TypeScript AI agent framework — developed by the founder of Gatsby.
Mastra is an open-source TypeScript AI agent framework developed by Mastra Inc., which was founded in late 2024 by Sam Bhagwat, the co-founder of Gatsby. Similar to how React builds web interfaces by combining UI components, Mastra builds AI agent applications by combining four core primitives: Agent, Workflow, Memory, and Tool. It features a Model Router that connects over 90 LLM providers (including OpenAI, Anthropic, and Google Gemini) through a single interface, requiring only one line of code to be modified when switching providers. In April 2026, it secured a $22 million Series A investment led by Spark Capital (totaling $35 million), and has established itself as a leading AI agent framework in the TypeScript ecosystem, with 25.2k stars on GitHub and over 300,000 weekly downloads on npm.
The existing AI agent framework market has been dominated by Python-based tools such as LangChain, CrewAI, and AutoGen. For web full-stack developers to build agents, they had to set up a separate Python environment or bear the cost of language conversion, and there was also the issue of type safety being broken between the front-end, back-end, and agent. Mastra directly addresses this gap. Its graph-based workflow engine allows complex multi-step pipelines to be declaratively constructed using method chaining such as .then(), .branch(), and .parallel(), and it also incorporates human-in-the-loop pausing and resuming, state persistence, and time-travel debugging, providing a one-stop infrastructure for production-level agent operation. It also natively supports the creation of MCP (Model Context Protocol) servers, allowing agents and tools to be exposed externally using a standardized protocol.
From a practical application perspective, Mastra demonstrates strengths in three main scenarios. First, it can be used to embed AI agents into existing SaaS products, with companies like Replit, Sanity, and MongoDB integrating Mastra agents into their platforms to implement natural language-based data retrieval and workflow automation. Second, it can be used to build internal workflow automation copilots, allowing for the creation of long-term conversational assistants using a Memory Gateway in domains such as HR, sales, DevOps, and clinical research. Third, it can be used as a data analysis agent, where connecting Mastra agents to TypeScript-based research data portals or dashboards allows for database queries, report generation, and analysis pipeline triggering to be handled within a single conversational interface. Through Mastra Studio (localhost:4111), agent behavior can be visually monitored, and agent quality can be quantitatively measured using the built-in scorer and evaluation tools, allowing the entire lifecycle from prototyping to production deployment to be managed within a single framework.
💻 System Requirements
Not required (LLM uses an external API call method; if using a local LLM, follow the requirements of that model)
Framework ~200MB (including node_modules), additional storage per project
⚡ Installation
4-1. Quick Start
npm create mastra@latest
Or, using a package manager:
pnpm create mastra
yarn create mastra
bunx create-mastra
4-2. Detailed Installation and Basic Usage
# Create a project (interactive setup)
npm create mastra@latest my-agent-app --default --llm openai
# Run the development server + Mastra Studio
cd my-agent-app
npx mastra dev
# → Mastra Studio: http://localhost:4111
Example agent definition:
import { Agent } from '@mastra/core/agent'
export const myAgent = new Agent({
id: 'my-agent',
name: 'Research Assistant',
instructions: 'You are a helpful research assistant.',
model: 'openai/gpt-4o',
})
// Usage
const result = await myAgent.generate('Summarize recent findings on CRISPR delivery.')
console.log(result.text)
🧬 Bio Use Cases
SaaS Embedded Agent
Add a natural language interface to your web application, allowing users to query data, modify settings, and trigger automations through conversation. This is a pattern adopted by companies like Replit, Sanity, and Brex, and can be implemented by connecting Mastra Agent with existing REST API tools.
Multi-Step Data Analysis Pipeline
Automate the process of collecting data from multiple sources, cleaning it, analyzing it, and generating reports using the .branch() and .parallel() functions of the Workflow engine. Human-in-the-loop validation steps can be inserted for intermediate verification.
Internal Task Copilot
Build a domain-specific assistant that connects to 251 MCP tools, including Slack, Notion, and GitHub, while maintaining conversational context with the Memory Gateway. This can be used for onboarding new employees, automatically summarizing code reviews, and triaging issues.
FAQ
What is Mastra?
Mastra is an open-source TypeScript AI agent framework developed by Mastra Inc., which was founded in late 2024 by Sam Bhagwat, the co-founder of Gatsby. Similar to how React builds web interfaces by combining UI components, Mastra builds AI agent applications by combining four core primitives: Agent, Workflow, Memory, and Tool. It features a Model Router that connects over 90 LLM providers (including OpenAI, Anthropic, and Google Gemini) through a single interface, requiring only one line of code to be modified when switching providers. In April 2026, it secured a $22 million Series A investment led by Spark Capital (totaling $35 million), and has established itself as a leading AI agent framework in the TypeScript ecosystem, with 25.2k stars on GitHub and over 300,000 weekly downloads on npm. The existing AI agent framework market has been dominated by Python-based tools such as LangChain, CrewAI, and AutoGen. For web full-stack developers to build agents, they had to set up a separate Python environment or bear the cost of language conversion, and there was also the issue of type safety being broken between the front-end, back-end, and agent. Mastra directly addresses this gap. Its graph-based workflow engine allows complex multi-step pipelines to be declaratively constructed using method chaining such as .then(), .branch(), and .parallel(), and it also incorporates human-in-the-loop pausing and resuming, state persistence, and time-travel debugging, providing a one-stop infrastructure for production-level agent operation. It also natively supports the creation of MCP (Model Context Protocol) servers, allowing agents and tools to be exposed externally using a standardized protocol. From a practical application perspective, Mastra demonstrates strengths in three main scenarios. First, it can be used to embed AI agents into existing SaaS products, with companies like Replit, Sanity, and MongoDB integrating Mastra agents into their platforms to implement natural language-based data retrieval and workflow automation. Second, it can be used to build internal workflow automation copilots, allowing for the creation of long-term conversational assistants using a Memory Gateway in domains such as HR, sales, DevOps, and clinical research. Third, it can be used as a data analysis agent, where connecting Mastra agents to TypeScript-based research data portals or dashboards allows for database queries, report generation, and analysis pipeline triggering to be handled within a single conversational interface. Through Mastra Studio (localhost:4111), agent behavior can be visually monitored, and agent quality can be quantitatively measured using the built-in scorer and evaluation tools, allowing the entire lifecycle from prototyping to production deployment to be managed within a single framework.
When should I use Mastra?
Open-source TypeScript AI agent framework — developed by the founder of Gatsby.
What is a biomedical use case for Mastra?
SaaS Embedded Agent: Add a natural language interface to your web application, allowing users to query data, modify settings, and trigger automations through conversation. This is a pattern adopted by companies like Replit, Sanity, and Brex, and can be implemented by connecting Mastra Agent with existing REST API tools.
📝 Update Notes
No update notes yet.
🧪 Related Code of Life
No related Code of Life posts yet.