October 7, 2026
Mastra introduces cross-trace Span Query APIs, cost/token trace aggregations across major storage backends, and self-embedding vector support. Additionally, @mastra/connect hits v1.0 with breaking changes.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
Mastra announced new observability and agent infrastructure features: - Span Query API : New endpoints and SDK methods ( storage.querySpans() , client.querySpans() , POST /api/observability/spans/query ) allow querying completed spans directly across traces with filtering, cursors, previews, and model costs. - Trace Aggregation Analytics : aggregateTraces() now computes token (input, output, total, reasoning, cached) and cost averages/sums across ClickHouse, DuckDB, and Postgres storage engines, supporting currency handling and coverage reporting. - Self-Embedding Vector Stores : Semantic recall can now run without a client-side embedder using MastraVector.isSelfEmbedding (such as MongoDBVector via autoEmbed ). - Session & Model Management : AgentController sessions now persist a single active model per thread across mode switches, and support starting sessions without creating an initial thread ( createInitialThread: false ). - Breaking Changes in @mastra/connect@1.0.0 : Discovered MCP tools no longer require manual approval by default (replaced autoApproveTools with opt-in requireApproval ). The API renamed integrations to providers , deprecated several legacy helpers/surfaces, and changed the Slack channel integration ID from slack to slack-channels . - Other Breaking Changes : AgentController.session.model.switch removed modeId / scope and now uses switch(modelId, options?) . @mastra/react hooks replaced positional arguments with a single object configuration accepting queryOptions .
可执行摘要
Mastra core updates bring querySpans across storage engines, cost/token metrics in aggregateTraces(), and self-embedding vector support via MastraVector.isSelfEmbedding. Breaking updates include @mastra/connect 1.0 (tools approved by default, integrations renamed to providers, Slack ID rename), AgentController model switching updates, and @mastra/react hook signature changes.
- Agent 实用度
- 94/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Upgrade @mastra/connect to 1.0.0 and configure 'requireApproval' if explicit approval is required for discovered MCP tools.
- Update @mastra/connect integration references: rename 'integrations' to 'providers' and change Slack channel integration IDs to 'slack-channels'.
- Refactor AgentController model switching logic to use the updated 'session.model.switch(modelId, options?)' signature.
- Migrate @mastra/react hook calls from positional arguments to a single options object.