We're actively preparing for Mastra 1.0 and this week we're shipping Agent Studio - a unified interface for building and testing your Mastra agents, now with team collaboration features.
Release: @mastra/core@0.23.1
Let's dig in...
Announcing Mastra's Agent Studio
When Mastra went viral, Playground was what blew people's minds the most.
Then the community started taking projects into production. Users began asking us: how do I share the playground with my team?
Multiple teams told us about the time they were spending building their own admin interfaces, deploying dev playgrounds, and wiring it all up.
So we built a better way.
Agent Studio: for local development or team collaboration
Now, you can share access to Studio using Mastra Cloud.
We let you spin up a sandbox and run Studio inside of it, in an environment that's accessible to your team. That means you can show your teammates the agent you've been building.
We've renamed Playground to Studio to reflect its growth from a local development tool into a team-ready workspace where everyone can view, share, and give feedback to improve your agent in every stage of development.
Getting started
When you're developing locally, Studio works exactly like Playground always has. Just run mastra dev.
To spin up your Studio in Mastra Cloud, connect your pre-existing deployment (by selecting your Mastra API URL), project GitHub repo, or pick a Mastra template. Your team can now test agents and workflows, validate prompts, and provide feedback.
See our full announcementfor more details.
Other Notable Updates
AI SDK Compatibility
Fixed critical compatibility issues with Vercel's AI SDK:
- Fixed uiMessages duplication in chat routes
- Network routing agent streaming now properly supports text deltas
- Workflow stream refactoring with new fullStream property
See PR #8904, PR #8979, and PR #9048 for implementation details.
Arize Observability
New @mastra/arize package for AI tracing with OpenInference, enabling integration with Arize AX, Arize Phoenix, or any OpenInference compatible collector.
See PR #8827 for implementation.
Model Span Renaming
Breaking Change: LLM spans have been renamed to Model spans for clarity. This better reflects that these span types apply to all AI models, not just Large Language Models.
AISpanType.LLM_GENERATION→AISpanType.MODEL_GENERATIONLLMGenerationAttributes→ModelGenerationAttributes
See PR #9105 for the full migration guide.
Tool Call Approval
Added approval workflows for tool execution, enabling human-in-the-loop patterns for sensitive operations.
See PR #8649 for details.
Error Handling
Fixed error serialization in agent streaming to ensure errors are consistently exposed and preserved throughout the streaming lifecycle.
See PR #9192 for implementation.
1.0 Release Preparation
We're actively preparing for our 1.0 release with stable APIs and stronger documentation. This milestone will mark Mastra's production readiness with backward-compatible APIs and enterprise-grade features.
For the complete list of changes, check out the full release notes.
That's all for this update!
Happy building! 🚀