@ledgermem/mastra

LedgerMem-backed memory provider for Mastra agents. Combines a per-thread short-term buffer with semantic long-term recall.

Install

npm install @ledgermem/mastra @ledgermem/memory @mastra/core

Set LEDGERMEM_API_KEY and LEDGERMEM_WORKSPACE_ID, or pass them in.

Quickstart (30 seconds)

import { Agent } from "@mastra/core/agent";
import { openai } from "@ai-sdk/openai";
import { LedgerMemMemory } from "@ledgermem/mastra";

const memory = new LedgerMemMemory({ workingWindow: 20, recallLimit: 5 });

const agent = new Agent({
  name: "support",
  instructions: "Use long-term memory to personalize responses.",
  model: openai("gpt-4o"),
  memory,
});

await agent.generate("Remember that I prefer concise answers.", {
  threadId: "user-42",
  resourceId: "user-42",
});

What you get

• remember({ threadId, query }) — recent messages + top-k semantic hits
• rememberMessage({ threadId, message }) — appends to working buffer and persists user/assistant turns to LedgerMem
• Per-thread metadata (threadId, resourceId, role, createdAt) stored alongside each memory for filtering and audit

License

MIT