Mastra Blog 功能更新汇总

106 posts on Features from the Mastra blog.

Introducing Mastra Connect

Your agents can now use Linear, Notion, Slack, and 20 other services with a single key.

Managed PostgreSQL for VPC-isolated deployments.

Introducing Jira, GitLab and incident.io Integrations for Mastra Factory

More ways to manage issues, feature requests, and incident follow-ups.

Introducing Memory Hooks

Monitor your agent's memory cycle, or intercept what it sees and stores.

Introducing Classifiers with Jev

Evaluation model support for making fast, predictable decisions.

Introducing Vitest Integration for Testing

Catch regressions and errors before you deploy.

Introducing Experiment Tool Mocks

Run experiments without hitting live APIs.

Introducing Experiment Lifecycle Hooks

Configure each stage of an experiment run.

Introducing Experiment Targets

Run experiments on workflows and scorers too, not just agents.

Query feedback by region, environment, thread and source.

Real-time updates for failed deploys and crashed services.

Diagnose and fix failed deploys with helpful recommendations.

Monitor server requests and inspect by path, status, or latency.

Reusable skills for multi-agent systems.

Find relevant content by keyword, semantic-similarity, or both.

Read and write to multiple filesystems using the same file interface.

Give your agents access to filesystems your team already use.

Control, status, hooks, persistence, and background processes.

Give your agents a computer to use, just like you do.

Introducing Harness Channels

Run a multi-player harness from Slack, iMessage, WhatsApp, Discord, and more.

Real-time caps on context window and response lengths.

Real-time spending caps configurable by thread, user, session, org, and more.

Kubernetes best practices for your self-hosted Mastra project.

Load agent skills on demand to reduce context bloat, token usage and cost.

Load agent tools on demand for sharper tool selection, lower token usage, and less cost.

Assert multi-turn conversations using deterministic gates and LLM-as-judge scorers.

Per-user, per-resource permissions for routes, agents, workflows, tools, memory, and MCP servers.

AI SDK v7 support in Mastra

Image generation, reasoning control, multimodal prompts, stream handlers, and UI helpers.

A preconfigured coding agent with sandbox, task list, error retries, and a goal-judge built-in.

Five ready-to-use tools for common agent tasks.

Introducing Dynamic Workflows for Mastra

Add, update, and remove workflows from a running Mastra server — without touching the code.

Introducing iMessage Support for Mastra Agents

Let users chat to your agents using iMessage on iPhone, iPad, or Mac.

Automatically redact sensitive customer information from agent traces.

Deploy your entire Mastra project to E2B, Daytona, or Vercel sandboxes.

Capture human feedback for every agent response and query them by agent, source, or time.

You can now give every environment a persistent filesystem and secure sandboxes, provisioned and billed through the Mastra platform.

Evaluate agent output against a rubric of criteria.

Introducing Environments for Mastra Platform

You can now deploy the same project to production, staging, and preview environments each with its own URL, variables, and database.

You can now pin a deployment environment to a region and your server, hosted databases, and observability data all run there.

Extract structured data from agent conversations.

Introducing Goals for Mastra Agents

Durable objectives for long-running agents with LLM-judged evaluations.

Observe and control every tool call your agent makes.

Catch agent regressions and ship with more confidence.

Send notification signals to your agents from any external source.

A unified API for managing automated recurring tasks

Attach skills directly to your agents.

Define your agents using folders and files.

Record and replay your agent's browser sessions.

Your agents can now react to webhook events from GitHub, Slack, Stripe, or your own API.

Build and control long-running agent experiences with sessions, modes, permissions, subagents.

Introducing Durable Agents

Survive client disconnects, browser refreshes, or network blips with cacheable resumable streams.

Send or receive events with Redis Streams and Google Cloud Pub/Sub.

Connect external sources like GitHub and Slack and trigger them to wake up your agent.

Keep your agents on track through long-running, complex work.

Introducing deployable templates in Mastra platform

Create and deploy a new Mastra project using pre-built templates.

Run, debug, and improve your Mastra agents from the terminal.

Introducing auto-deployment in Mastra platform

Create and deploy your Mastra platform project from GitHub

Provision a database for your Mastra project in seconds.

Introducing Claude Code, Cursor, and Codex agents in Mastra

Build your own coding agents using Claude, Cursor, or OpenAI's agent SDKs.

Your agents can now run multi-tool computations with more accurate results.

Delegate coding tasks to Claude Agent, Codex CLI, Cursor, or any ACP-compatible harness.

Mastra Workflows, Enhanced

Mastra's new Temporal integration brings durable execution, advanced retries, and richer observability to your workflows.

Introducing Agent-to-Agent support for Mastra

Build cross-framework multi-agent systems with A2A-enabled agents.

You can now cut costs and reduce latency with cached responses.

Your workflows can now run on Temporal with durable, fault-tolerant execution.

Your agents can now stream progress feedback during long-running tool calls.

Make your Mastra application production-ready with observability storage that scales.

Your agents can now work with remote filesystems like S3, GCS, and Azure.

Introducing Channels for Mastra Agents

Give users access to your agent from Slack, Discord, Telegram, and more.

LSP Inspection for Mastra Workspaces

Give agents semantic understanding of your codebase.

Your agents can now browse and take actions on the web, just like you.

Introducing Agent Editor

Edit agent instructions, tools, and display conditions from Studio with draft/publish versioning, or programmatically through the editor API.

Introducing Metrics and Logs

Mastra Studio now supports logs, traces, and metrics, so you can understand how your agents behave in production.

Secure your deployed Mastra Studio with login screens, third-party identity providers, and role-based permissions.

Three new sandbox providers for safely running agent code in isolated cloud containers.

A new video course designed to get you from zero to deploying your first agent in under two hours.

Change, Run, and Compare with Experiments in Mastra Studio

Run test cases against agents and workflows, score the results, and track quality over time.

Introducing Datasets in Mastra Studio

Define test cases, run experiments, and track response quality for every change you make.

Introducing Mastra Skills

Mastra Skills teach coding agents about Mastra through structured knowledge files, giving any compatible agent the context it needs to help you build.

File access, command execution, and reusable skills for agents that actually get things done.

Mastra’s new Server Adapters automatically expose your agents, workflows, tools, and MCP servers as HTTP endpoints on the server you’re already running.

Mastra UI Dojo: Choose your frontend

We have a new UI dojo. It's a collection of working examples showing Mastra agents integrated with the most popular AI UI frameworks. Try it out at ui-dojo.mastra.ai.

Introducing Mastra’s Agent Studio

We've renamed Playground to Studio and it's now shareable with your team.

HITL: Ask Before Acting

Learn how to use HITL (Human-in-the-Loop) to safely build tools that require human approval.

You can suspend/resume workflows in playground

Mastra workflows now support suspend and resume in Playground. You can also use the new resumeStream API to close streams on suspend and resume them later.

The evolution of AgentNetwork

We've unified multi-agent coordination under a new .network() primitive—read about our journey, the experiments, and how you can use it to orchestrate collaborative agents simply.

Introducing Mastra Model Router: 600+ models, one API, zero package installs

Access 600+ LLM models from 40+ providers with a single string. Full TypeScript autocomplete turns your IDE into a model search engine.

Model fallbacks: Your safety net for production AI

Model fallbacks ensure your application stays online by automatically switching to backup models when your primary provider fails.

Mastra now supports AI Tracing

Mastra introduces AI Tracing: filter out noise and use multiple observability platforms.

Adding agent templates into existing projects

You can now easily merge Mastra agent templates into your existing projects with just a few clicks.

The next evolution of Mastra streaming

We're making our new streaming architecture the default experience starting from `@mastra/core v0.19.0.`

Introducing Output Processors

Introducing output processors: the gatekeepers that sit between your language model's response and your users.

Introducing Cloneable Scorers

Cloneable scorers let you copy, customize, and extend Mastra’s evals into your project.

Introducing Mastra's improved agent orchestration with AI SDK v5 support

Mastra now controls the agent loop and tool calling with increased orchestration capabilities—while maintaining backward compatibility with AI SDK v4 and v5.

Introducing Scorers in Mastra

We're excited to announce the release of scorers in Mastra, a new way to evaluate and rank the quality of your agent's responses.

Nested streaming support in Mastra

Mastra's nested streaming support provides real-time visibility into agent and workflow execution, with comprehensive cost tracking and a unified messaging interfaces.

Introducing Mastra Templates

Announcing Mastra Templates: pre-built, production-ready, customizable agent and workflow projects. Plus: join our MASTRA.BUILD hackathon to create and showcase your own templates.

Agent Network (vNext) introduces intelligent AI orchestration that automatically routes and executes complex multi-agent tasks without predetermined workflows.

Explore how our new docs chatbot leverages the Mastra MCP Docs server to enhance user experience and streamline access to documentation.

Introducing vNext Workflows: the next generation of Mastra Workflows

vNext brings stronger control flow, better type safety, and multi-engine support.

Create and share your AI tools with the Model Context Protocol (MCP) - now available in our latest stable release.

Mastra's MCP Registry Registry simplifies the process of finding MCP registries.

We've expanded Mastra Voice with real-time speech-to-speech capabilities for Agents.

Introducing Mastra MCP Documentation Server

We built a Mastra MCP documentation server that provides AI assistants and IDE's with direct access to Mastra.ai's complete knowledge base.

Introducing Evals in Mastra

A guide to Mastra's evaluation metrics suite, demonstrating how to measure and validate AI model outputs across dimensions including answer relevancy, completeness, content similarity, context usage, and tone consistency.

Introducing TTS in Mastra

Use Mastra's TTS system with OpenAI's speech models to create text-to-speech applications, covering voice selection, audio generation, streaming capabilities

Ship better agents