Welcome to Mastra Changelog #3. Let's dive in.
Major Updates
New mastra build command
We've released a new mastra build command.
This command lets you bundle a production version of your Mastra agents and workflows into a Hono server.
This is good for deploying Mastra on a cloud server or containerized environment (ie, EC2, Digital Ocean, etc).
Evals added to Dev Playground
The big news here is that we've added evals to the dev playground.
This means you can now test your agents and workflows locally and get feedback on their performance.
Metrics can be accessed both during CI and in live environments through the playground.
Here's a screenshot:
More Dev Playground improvements
The local dev playground is now more powerful and easier to use:
- Migrated to Hono server for better performance
- Enhanced OpenAPI documentation
- Added Swagger UI support
- Automatic rebuild/refresh on file changes
Some build and performance improvements here too:
- Separated playground dependencies from CLI package
- Improved build & deploy configurations
- Enhanced tsconfig setup
These changes make the development experience more responsive and reliable.
More Evals
Building on our recently released @mastra/evals framework, we've added new eval metrics:
- Bias detection
- Hallucination measurement
- Contextual relevance
- Summarization quality
In addition, we've reorganized metrics into NLP and LLM categories for better organization, and metrics now return a standardized format with score and info object containing detailed information.
New Model Support
Added support for:
- Perplexity Sonar model
- o1 model support
- DeepSeek-v3 and DeepSeek-r1 model support
- PlayAI TTS with streaming
New @mastra/deployer package
We've released a new @mastra/deployer package to streamline deployment workflows.
The package handles deploying Mastra into a serverless environment (ie, bundling and code splitting appropriately), managing secrets, and writing files.
Because we've abstracted it out into a package, other providers can extend it. If you're interested in adding support for another provider, take a look at the @mastra/deployer package and the individual provider code: Cloudflare, Vercel, and Netlify.
Mastra Memory (coming soon)
We're finalizing a number of advanced memory features:
- Hierarchical memory storage
- Long-term memory via memory compression
- Vector search memory
The code is currently in a draft PR on a branch: https://github.com/mastra-ai/mastra/pull/1525
Sneak peek:
Technical Improvements
RAG & Memory
- Added transaction support to pgvector
- Improved vector query search functionality
- Enhanced memory state management
Bug Fixes
- Fixed agent generate/stream with structured output in dev playground
- Resolved workflow trigger handling issues
- Addressed CLI build and deployment edge cases
Package Version Updates
@mastra/core(0.1.27-alpha.68)@mastra/cli(0.1.57-alpha.106)@mastra/deployer(0.0.1-alpha.6)@mastra/evals(0.1.0-alpha.8)@mastra/engine(0.0.5-alpha.62)@mastra/memory(0.0.2-alpha.48)@mastra/mcp(0.0.1-alpha.8)@mastra/rag(0.0.2-alpha.54)@mastra/tts(0.0.1-alpha.15)
Source: Mastra Changelog 2025-01-17, Mastra Changelog 2025-01-10