@mastra/core@1.72.0 @mastra/core@1.73.0

medium risk

hcengineering/platformSep 30, 2026 → Oct 5, 20265 days between releases

What's New

Adds default error recovery, Google Workspace toolsets, and tool-pause UX improvements.

Breaking Changes

  • Durable streams no longer emit a separate tripwire chunk; clients must check `finishReason === 'tripwire'` and `output.tripwire` instead.

New Features

  • Default agent error recovery with repair-first strategy and safe retries.
  • Google Workspace toolsets for Docs, Drive, and Sheets.
  • Observational memory can display attachments to the model.
  • New `tool-call-resumed` stream chunk for better UI state management.
  • Exports `findGatewayForModel` and `getGatewayId` from `@mastra/core/llm`.

Improvements

  • Reliability and scale fixes for durable/evented execution.
  • Invalid tool-call ID rewriting for Anthropic models.
  • New `resolveSubagentModel` option with requestContext support.
  • Improved reliability for background tasks via heartbeat and lease fencing.

Bug Fixes

  • Reliability and scale fixes for durable/evented execution.

Upgrade Notes

Clients consuming durable streams must update logic to check `finishReason === 'tripwire'` and `output.tripwire` instead of expecting a separate chunk. The new default error recovery is enabled but can be disabled via `errorProcessorDefaults: false`.

The Two Releases

@mastra/core@1.72.0

September 29, 2026

Sep 30, 2026· 22 days ago

This release adds live channel resolvers for runtime channel connections without redeploy, implements multi-worker safety for background tasks via persisted ownership leases, strengthens durability and recovery for agents/workflows, and introduces observability upgrades with trace aggregation.

Full release notes
@mastra/core@1.73.0

October 1, 2026

Oct 5, 2026· 17 days ago

This release introduces default agent error recovery with repair-first strategy and safe retries, improves tool-pause UX with a new tool-call-resumed stream chunk, adds Google Workspace toolsets, and enables observational memory to show attachments to the model.

Full release notes