2025-05-06 - Mastra Release

Agents

• Ensure empty assistant messages after tool use are preserved in conversation flows, preventing errors caused by their removal. #3832
• The display of agent name initials on empty threads to show initials like "WA" for "Weather Agent" instead of just the first letter. #3991

Workflows

• Additional test cases to verify Mastra class access within step execution functions. #3980
• A serializedStepGraph to resolve a circular structure error in the code. #3982
• Parameters to filter workflow runs in both workflows and vNextWorkflows, including fromDate and toDate options for telemetry. #4006
• CommonJS imports in the vnext workflow. #4069
• Functionality to collect traces for the vnext version. #4076
• The getWorkflowRunById API to the vNext workflows core and server handlers. #4097
• GetWorkflowRunById in vnext to retrieve a run from this.#runs if it is not found in storage #4101
• The vnext workflow test to improve the project's testing process. #4103
• Dynamic input mappings, enabling more flexible and configurable input handling in the system. #4095

Deployer

• Fails gracefully instead of breaking when tools cannot be imported on platforms like Cloudflare. #4099
• Fixes a bug by ensuring directories are properly traversed when generating vercel.json so that all files are included in the includedFiles array. #4102

Playground

• Refactored the playground UI to receive traces as props instead of fetching them internally, enabling traces to be provided from external sources like Clickhouse. #4004
• The playground UI to display span and trace durations with three significant digits for improved precision. #4017
• Cookie consent component to ensure it only loads with user consent. #4020
• The threads component to unify and synchronize the UI between the cloud and playground environments. #4022

Client SDK - JS

• Remove a custom header, fixing workflow JSON issues, and updating the AI SDK dependency in the JavaScript client library. #3984
• API for retrieving workflow runs. #3996
• Add AGUI client into the project. #4092 [IMPORTANT]

MCP

• Deprecation warning only appears when using the deprecated class, not its subclass. #3988
• MCP tool schema conversion by switching to 'json-schema-to-zod' for more accurate schema handling in the playground, resolving form field type issues and allowing tools without input fields to be submitted without arguments. #3998

Memory

• A deprecation warning for working memory text-streaming in the @mastra/memory and @mastra/core packages. #4012
• A warning to notify users that, starting May 20th, the Memory component will no longer have default storage unless inherited from the Mastra instance, in preparation for an upcoming breaking change. #4083
• The token estimation accuracy in @mastra/memory from 96% to 99% by fine-tuning the base token amount added to messages and tools, resulting in more precise estimates and more reliable tests. #4087
• Support for handling textpart message content when upserting vector embeddings into the memory_messages table. #4086

Observability

• New agent getter methods from telemetry to prevent them from being tracked. #4003

RAG

• The new @mastra/fastembed package to encapsulate the current default embedder, adds warnings for users of the old embedder, and updates documentation to reference the new package in preparation for an upcoming breaking change. #4093

Storage

• Couchbase as a supported vector store in Mastra, enabling vector search and embedding operations via the MastraVector interface. #3739
• Fix typescript errors related to the default proxy storage implementation. #4015

CLI

• A new CLI command, mastra lint, to add linting functionality to the project. #3994 [IMPORTANT]
• cookie consent component to ensure it only loads with user consent. #4020

Quality of life

• The end-to-end tests by migrating the monorepo test setup to use Verdaccio, enabling compatibility with multiple package managers. #3993
• An unnecessary console log that appeared when using the hallucination metric. #3997
• The PostHog dependency to the most recent working version to avoid issues caused by broken ESM exports in newer releases. #4029
• Type mismatches related to the new DynamicArgument wrapper for models, resolving build and test failures by ensuring correct typing when passing models to tools. #4030
• The ai-sdk useCompletion example to correctly use the prompt field from the request body instead of the messages field, improving documentation accuracy. #4098
• The ai-sdk useObject example to correctly utilize the request body, ensuring the documentation accurately demonstrates proper usage. #4100