Mastra Review, Pricing & Alternatives (2026)

Mastra is an open-source TypeScript framework for agents and workflows, made by Kepler Software and the team behind Gatsby. Work began in October 2024, and Kepler Software was founded in 2024. Mastra joined Y Combinator’s Winter 2025 batch, giving the project an early company and funding context. The company announced a $13 million seed round in October 2025 and a $22 million Series A in April 2026, bringing reported total funding to $35 million. Developers can use Mastra with React, Next.js, Node.js, or a standalone server. Its GitHub repository lists nearly 29,000 stars and 3,000 forks. The company says more than 300 teams ran the 1.0 beta in production. The core framework uses Apache 2.0, while enterprise code has a separate license.

A typical Mastra project begins with an agent configured with a model, instructions, and typed tools. Agents can generate or stream responses, keep conversation history, retrieve information from APIs, databases, or files, and pause for human approval before taking an action. Workflows cover processes that need more predictable control, using typed steps, branches, parallel execution, retries, suspension, and resumption. Storage preserves workflow state and memory between runs. Mastra Studio provides a place to build and test agents, while evaluations support model-based or code-based scoring. Datasets make experiments repeatable, and tracing records model, tool, and workflow calls. The framework also supports MCP servers, server adapters, guardrails, metrics, and connections to external observability services, so teams can inspect behavior as projects move from tests into deployed applications.

Mastra uses a unified model interface, so applications do not need to rely on a separate provider SDK for every model connection. In October 2025, Mastra introduced a router covering more than 600 models from 45 providers, with automatic fallback chains when a provider fails. Its January 2026 1.0 release stabilized the public APIs and made server adapters the default deployment path. The release also added composite storage, thread cloning, AI SDK v6 support, a unified observability schema, and workflow changes. Composite storage lets teams choose different backends for memory, workflows, scores, and observability, rather than tying every function to one database. Mastra also offers Observational Memory, which uses background agents to compress growing conversation history. The 1.0 release followed production use by hundreds of teams during the beta.

Mastra separates its free framework from charges for the hosted platform. Platform Starter costs $0 per month and includes 100,000 observability events, 24 CPU hours, 10 GB of egress, 15-day retention, and unlimited users, deployments, and projects. Starter overages cost $10 per 100,000 events, $0.35 per CPU hour, and $0.10 per GB of egress. Teams costs $250 per month and includes 1 million events, 250 CPU hours, 100 GB of egress, six months of retention, multiple teams, SSO, and SOC 2 documentation. Its gateway adds 5.5 percent to market-rate model pricing. Memory, retrieval storage, database use, and persistent servers carry separate rates, including $100 per project for persistent uptime. Enterprise pricing is custom, with controls, support, retention, and service terms negotiated separately.

Mastra’s main limitation is scope. Agents, workflows, storage, memory, evaluations, deployment, and tracing give teams many parts to understand, creating a larger surface area than a single-purpose model SDK. The framework also requires TypeScript, and model charges remain separate from the framework price. LangGraph is the closer choice for teams centered on durable graph orchestration and explicit state transitions. Vercel AI SDK is a lighter foundation for applications that need model and user interface primitives without Mastra’s broader platform layer. Letta is more focused on persistent agent identity and memory than on combining deployment, workflows, evaluations, and traces. Mastra is best suited to TypeScript teams that want those capabilities in one framework, while smaller applications may prefer a narrower toolset.