Mastra

v1.x

Mastra AI (YC W25)

Agenttypescriptworkflowsopen-source

Strong

About This Agent

TypeScript-first AI agent framework from the Gatsby founders, combining agents, durable workflows, RAG, and evals in one toolkit. Provider-agnostic via Vercel AI SDK model routing, with a local dev playground and 1.0 stable release in January 2026.

Last Evaluated: June 10, 2026

Official Website

Trust Vector Analysis

Dimension Breakdown

🚀Performance & Reliability

task completion accuracy

Task completion testing across agent and workflow primitives

Evidence

Mastra Documentation — Agents combine tool calling, memory, and structured workflows; built-in evals allow teams to measure task accuracy

mediumVerified: 2026-06-10

tool use reliability

Tool invocation testing with typed schemas and MCP servers

Evidence

Mastra Tools Documentation — Zod-schema typed tools with input validation plus MCP client/server support reduce malformed tool calls

highVerified: 2026-06-10

multi step planning

Complex multi-step task testing using workflow engine

Evidence

Mastra Workflows Documentation — Graph-based durable workflows with branching, parallel steps, suspend/resume, and human-in-the-loop

highVerified: 2026-06-10

memory persistence

Memory system evaluation across threads and semantic recall

Evidence

Mastra Memory Documentation — Built-in working memory, conversation history, and semantic recall backed by pluggable storage adapters

highVerified: 2026-06-10

error recovery

Error injection testing on workflow retry and resume paths

Evidence

Mastra Workflows Documentation — Workflow steps support retries, error branches, and durable suspend/resume after failures

mediumVerified: 2026-06-10

agent collaboration

Multi-agent coordination testing via sub-agents and networks

Evidence

Mastra Agent Networks — Agents composable as workflow steps and sub-agents; agent network primitives for dynamic routing between agents

mediumVerified: 2026-06-10

🛡️Security

tool sandboxing

Security architecture review of tool execution model

Evidence

Mastra Architecture — No built-in tool sandboxing; tools run in the host Node.js process and isolation is the developer's responsibility

mediumVerified: 2026-06-10

access control

Access control assessment of server auth options

Evidence

Mastra Server Middleware Docs — Auth middleware hooks on the Mastra server; RBAC and SSO are enterprise features rather than core defaults

mediumVerified: 2026-06-10

prompt injection defense

Injection testing with and without guardrail processors configured

Evidence

Mastra Processors Documentation — Input/output processors enable moderation and PII filtering guardrails, but injection defense is not on by default

mediumVerified: 2026-06-10

data isolation

Data isolation architecture review of storage and memory scoping

Evidence

Mastra Storage Documentation — Memory and storage scoped per thread/resource with pluggable backends; multi-tenant isolation is application-level

mediumVerified: 2026-06-10

open source transparency

Source code and license structure review

Evidence

Mastra GitHub Repository — Apache 2.0 core with 22k+ stars; enterprise features are source-available under the Mastra Enterprise License

highVerified: 2026-06-10

🔒Privacy & Compliance

data retention

Privacy architecture review of self-hosted deployment

Evidence

Self-Hosted Framework — Framework runs in user infrastructure with user-chosen storage backends; full control of retention when self-hosted

highVerified: 2026-06-10

gdpr compliance

Compliance capabilities assessment of deployment configurations

Evidence

Mastra Documentation — GDPR compliance achievable through self-hosting and storage choice; no turnkey compliance tooling in core

mediumVerified: 2026-06-10

third party data sharing

Data flow analysis across model provider configurations

Evidence

Mastra Model Routing — Data flows to whichever provider is configured via Vercel AI SDK routing; local models supported through Ollama-compatible providers

mediumVerified: 2026-06-10

local deployment option

Deployment options assessment

Evidence

Mastra Deployment Documentation — Deploys as a standard Node.js server anywhere, including fully self-hosted environments with local model providers

highVerified: 2026-06-10

👁️Trust & Transparency

documentation quality

Documentation completeness and accuracy review

Evidence

Mastra Documentation — Polished docs with guides, templates, course content, and examples; quality reflects Gatsby founders' DX background

highVerified: 2026-06-10

execution traceability

Tracing and logging capabilities assessment

Evidence

Mastra Observability Documentation — OpenTelemetry tracing of agent runs and workflow steps plus a local playground for inspecting executions

highVerified: 2026-06-10

decision explainability

Explainability assessment of run visualization features

Evidence

Mastra Playground — Local studio visualizes agent steps, tool calls, and workflow graphs, aiding inspection of decisions

mediumVerified: 2026-06-10

open source code

Open source assessment of license split and code availability

Evidence

Mastra GitHub Repository — Apache 2.0 core, 1.0 stable released 2026-01-21; some enterprise modules under source-available Mastra Enterprise License

highVerified: 2026-06-10

community activity

Community engagement analysis of stars, downloads, and releases

Evidence

GitHub and npm Metrics — 22k+ GitHub stars, 300k+ weekly npm downloads, active Discord and rapid release cadence

highVerified: 2026-06-10

⚙️Operational Excellence

ease of integration

Integration complexity assessment with scaffolded project testing

Evidence

Mastra Quickstart — create-mastra CLI scaffolds a typed project with playground in minutes; idiomatic TypeScript APIs

highVerified: 2026-06-10

scalability

Scalability assessment across deployment targets

Evidence

Mastra Deployment Documentation — Deploys to serverless platforms (Vercel, Cloudflare, Netlify) and Node servers; durable workflows support long-running jobs

mediumVerified: 2026-06-10

cost predictability

Pricing model analysis

Evidence

Open Source Framework — Free Apache 2.0 core; costs from LLM APIs, hosting, and optional Mastra Cloud/enterprise offerings

highVerified: 2026-06-10

monitoring capabilities

Monitoring and evaluation features assessment

Evidence

Mastra Evals and Observability — Built-in evals (faithfulness, hallucination, relevance scorers) plus OTel export to external observability platforms

highVerified: 2026-06-10

production readiness

Production readiness assessment of API stability and release maturity

Evidence

Mastra 1.0 Release — 1.0 stable shipped 2026-01-21 with semver commitment after a year of rapid pre-1.0 iteration

highVerified: 2026-06-10

Strengths

+TypeScript-first with end-to-end type safety via Zod schemas
+Integrated suite: agents, durable workflows, RAG, memory, and evals in one framework
+Provider-agnostic model routing through the Vercel AI SDK
+Excellent developer experience: CLI scaffolding, local playground, polished docs
+1.0 stable (Jan 2026) with strong adoption: 22k+ stars, 300k+ weekly npm downloads
+Built-in eval scorers and OpenTelemetry observability

Limitations

!No built-in tool sandboxing; tools execute in the host Node.js process
!Enterprise features (RBAC, SSO, some modules) are source-available, not Apache 2.0
!Young 1.0; some advanced patterns and integrations still maturing
!No Python SDK, limiting access to the Python ML/data ecosystem
!Security guardrails (processors) are opt-in rather than default

Metadata

license: Apache 2.0 (core); Mastra Enterprise License (source-available enterprise features)

supported models

0: OpenAI

1: Anthropic Claude

2: Google Gemini

3: Any provider via Vercel AI SDK routing

4: Local models via OpenAI-compatible endpoints

programming languages

0: TypeScript

1: JavaScript

deployment type: Self-hosted or Mastra Cloud

tool support

0: Zod-typed custom tools

1: MCP client and server

2: Vercel AI SDK tools

3: Built-in RAG tools

github stars: 22000+

first release: 2024 (YC W25; 1.0 stable 2026-01-21)

pricing: Free (Apache 2.0 core) - Costs from LLM APIs, hosting, and optional enterprise/cloud offerings

node requirement: Node.js >=20

adoption: 300k+ weekly npm downloads; built by the Gatsby founders

Use Case Ratings

code generation

Typed TypeScript tooling and MCP support fit developer-workflow agents well

customer support

Memory, RAG, workflow human-in-the-loop, and guardrail processors suit support bots

content creation

Workflows plus evals enable quality-controlled content pipelines

research assistant

RAG primitives and agent networks handle multi-source research tasks

data analysis

Capable via tools and workflows, though Python data ecosystem is unavailable

education

Memory threads and structured workflows work well for tutoring applications