Another big week for Mastra! We're thrilled to announce Mastra Templates and our second hackathon, plus major improvements to Memory, Evals, and more.
Let's dive in:
Introducing Mastra Templates
Bootstrap your AI projects instantly with pre-built starting points for common agentic use cases.
npx mastra@latest create --template [template-name]
Browse our templates library featuring ready-to-use projects for:
- PDF chatbots
- Text-to-SQL
- Browser agents
- CSV analysis
- Docs chatbots
- And more
Skip the boilerplate and get straight to building.
MASTRA.BUILD Templates Hackathon
Our second virtual hackathon is focused entirely on building Mastra Templates! It's your turn to build the best templates and win prizes.
Join us from July 28 - August 11 for:
- Cash prizes
- Feature slots in our templates library
- Judges from Chroma, Arcade, WorkOS, Smithery, Browserbase, and Recall
Kickoff ceremony will be livestreamed Monday, July 28th at noon PT.
Register at https://mastra.build.
StreamVNext Protocol
We've shipped a new unified streaming protocol that enables nested streaming and usage tracking for all Mastra primitives — agents, tools, and workflows.
This major improvement provides:
- Unified streaming across all components
- Support for nested streaming operations
- Built-in token usage tracking and monitoring
Complete Eval Scores System
We've shipped a new evals infrastructure with code and LLM-based scoring capabilities. This comprehensive system includes:
- Scorer interface with abstract methods for custom implementations
- Code-based scoring for deterministic evaluations
- LLM-based scoring for subjective assessments
- Scorer hooks that can be triggered for agents and workflows
- Storage support across all major databases (PostgreSQL, MongoDB, DynamoDB, LanceDB, ClickHouse, LibSQL)
Now you can properly evaluate and benchmark your agents' performance across different scenarios with both objective and subjective metrics.
Memory Improvements
New MSSQL Provider
Microsoft SQL Server is now supported as a backend for memory storage, complete with implementation and tests. This expands your options for enterprise deployments.
Enhanced MongoDB Performance
We've improved query performance by pre-filtering on properties before vector search and fixed multi-property filter bugs. Your MongoDB-backed agents should see noticeable speed improvements.
Memory in Playground
Memory features are now available in the playground with semantic recall and configurable options. Test and develop memory-enabled agents directly in the UI — no more guessing what your agent remembers.
Runtime Context Configuration
Memory can now be configured using runtime context, making it easier to dynamically adjust memory settings based on your application's needs.
Developer Experience Improvements
Auth Improvements
The dev playground now bypasses authentication for local development when auth is enabled. This makes testing much easier — no more juggling tokens during development!
That's all for this week! We're especially excited about Templates and the new StreamVNext protocol — they represent major steps forward in making Mastra more powerful and easier to use.
For the complete list of changes and fixes, check out the full release notes.
Happy building! 🚀