Mastra Release - 2025-08-21

This release introduces two major features: the new streamVNext and generateVNext methods that support multiple output formats (Mastra and AI SDK v5 formats), and output processors for transforming agent streams and results.

Highlights

New Stream and Generate Methods

• Introduced streamVNext and generateVNext methods with configurable output formats - choose between Mastra's native format or AI SDK v5 format #6877
• Added explicit return types for the streamVNext() function API #6876
• Extended Client JS SDK with new vnext methods #6889
• Integrated generateVNext support in the playground #6894
• Added vnext streaming functionality to agent networks #6910
• Updated message list component #6870

Output Processors

• Unified input and output processing with a single Processor interface, enabling stateful output processing for agent streams #6482
• Added tests for agent processors #6880
• Enhanced agent methods with customizable and structured output processing options

AI SDK v5 Support

• Added comprehensive AI SDK v5 support to MessageList, handling both v4 and v5 message formats simultaneously #6731
• Fixed MessageList class and streaming helpers to properly format v5 response messages #6766
• Added v5 model input support #6719
• Implemented chunk type support for convertMastraChunkToAISDKv5 function #6739
• Fixed provider metadata preservation during message conversions between AI SDK v5 and internal v2 format #6760
• Added AI SDK v5 support to model switcher with automatic version detection #6909

Observability & Tracing

• Updated AI Tracing interface based on recent design changes #6777
• Added experimental AI Tracing support for workflows and agents #6824

Playground Enhancements

• Added model switcher with improved suggestions UI to playground #6737

Agents

• Added Mastra LLM class and finalized processing loop #6742
• Integrated Google's Gemini Live API with real-time TTS, STT, and voice-to-voice capabilities #6677
• Added support for model version in agents handler #6881
• Fixed agent scorers to display agent name instead of key #6898
• Removed _wrapToolsWithAITracing function from agent tracing #6911

CI / Tests

• Fixed types-builder to display TypeScript compilation errors in console during builds #6717
• Fixed '@babel/preset-typescript' package resolution issue #6835
• Fixed safelyParseJson function to handle string values correctly and added tests #6862
• Added memory tests #6873
• Added tests for agent processors #6880
• Fixed linting issues #6885
• Added end-to-end test #6895
• Fixed loop issue in test suite #6896
• Added ts-ignore directive in streamObject and extended test timeout #6897
• Added comprehensive tests for DefaultExecutionEngine.executeConditional error handling during condition evaluation #6829
• Added comprehensive tests for mitt.off method, verifying event handler removal and edge cases #6813
• Patched peer dependencies across packages to support core upgrade #6913

CLI / Playground

• Added 'scorers' subcommand to CLI for managing and listing scorers #6704
• Added tracking functionality for create-mastra command #6808
• Added dynamic options to playground interface based on selected model version #6882
• Added generateVNext support in playground #6894
• Fixed default value behavior in playground component #6903

Deployer

• Fixed missing rollup json() plugin for Cloudflare deployer, resolving issue #6728 #6747
• Fixed build failures for projects using TypeScript syntax by adding @babel/preset-typescript support #6786
• Fixed relative imports resolution in transpiled packages by marking only absolute paths as external #6821
• Updated Cloudflare API reference and documentation #6727
• Consolidated bundling code by centralizing extraction logic and standardizing helper usage #6836
• Added deployer handlers for managing deployment processes #6886

Developer Tools & UI

• Fixed mobile overflow bug to improve layout on mobile devices #6724
• Fixed NumberField component to correctly apply 'min', 'max', and 'step' constraints from schema #6745
• Updated UI elements and Storybook #6748
• Fixed MessageList class and streaming output helpers to properly format v5 response messages #6766
• Improved Scorer page UI with fixed title, description, and input preview #6803
• Fixed model switcher to display correct icon for Anthropic models #6823
• Updated bundle resolution logic to use Rollup's native resolve instead of resolveFrom #6837
• Fixed workspace dependency detection in bundler by checking workspace map directly #6883
• Fixed compatibility issues with p-map version 7+ on Node.js 20.0–20.18 #6893

Evals

• Updated evaluation LLM prompts to ensure 'reason' output is always returned as string #6794
• Added runExperiment method for running and scoring agents with customizable scorers in test suites or CI #6761
• Fixed missing threadId and resourceId fields in scorers #6850

Examples

• Fixed example links in documentation index page #6758
• Set up TypeScript configuration for examples directory #6780
• Fixed agent example #6820

MCP

• Fixed MCP Hono server to send endpoint messages before keep-alive messages #6844
• Fixed duplicated switch-case in core module #6778
• Updated model loop logic #6868

Memory

• Fixed system message duplication by skipping messages retrieved from memory #6738
• Fixed system to load most recent snapshot #6807

Networks

• Added processing functionality for Mastra stream #6830
• Updated server handler functionality #6884
• Added vnext streaming functionality to agent networks #6910

Observability

• Added LangFuse observability exporter package for AI tracing #6783
• Added Keywords AI as new observability provider #6843

Storage

• Upgraded Chroma dependency to v3+ with Chroma Cloud support #6735
• Updated PostgreSQL type definitions #6872

Tools

• Fixed tool ID propagation through transformations for accurate AI tracing and logging #6847

Website

• Fixed '405: Method not allowed' error with feedback form in production #6819
• Fixed Swagger UI link #6906

Workflows

• Reorganized documentation by moving Run methods into subdirectory with redirects #6711
• Fixed data loss during suspend/resume operations in do-while loop workflows #6720
• Added comprehensive loop functionality for workflows with multiple tool format support and AI SDK v5 streaming #6730, #6733, #6734, #6736
• Implemented execution step for workflows #6740
• Added custom ID parameter to workflow .map method #6799
• Improved workflow suspension documentation #6855
• Added tests for .map function renaming #6856
• Added loop-related code and tests #6869
• Added error handling for v2 models and initial generateVNext plumbing #6871