Skip to content

summarize

Small-tier summarization for the Mastra plugin.

Two surfaces, both backed by the fast / small chat tier (model.ModelClass.ChatFast) resolved through the same /serving-endpoints pipeline as the main agents:

  • A dedicated summarize tool (see buildSummarizeTool) agents can call to condense arbitrary text without burning the heavyweight chat model.
  • The model + instructions Mastra’s memory uses to auto-name conversation threads (generateTitle), so titling reuses the same small tier rather than the agent’s primary model.

Mirrors the chart-planner wiring in chart.ts: a per-config cached Agent on the fast tier, invoked via agent.generate(...) inside the active asUser scope so tokens stay user-scoped.