Memory federation SDK

@praxa/sdk/memory is published in @praxa/sdk@0.3.0. It is a dependency-free, read-only federation layer for provider clients and transports that your backend already owns.

Install

npm install @praxa/sdk@0.3.0

Query an existing LangGraph store

import {
  MemoryFederation,
  createLangGraphMemorySource,
} from "@praxa/sdk/memory";

const memory = new MemoryFederation({
  sources: [
    createLangGraphMemorySource({
      store,
      mapNamespace: ({ tenantId, subjectId }) => [
        tenantId,
        subjectId,
        "memory",
      ],
    }),
  ],
  maxConcurrency: 4,
  defaultTimeoutMs: 5_000,
});

const result = await memory.recall({
  query: "preferred editor",
  namespace: { tenantId: "acme", subjectId: "person-1" },
  limit: 10,
});

The package covers Mem0, LangGraph, Zep, Graphiti, Letta, and OpenAI Agents sessions. Each adapter:

accepts an injected client or structural transport;
requires an explicit namespace mapper or session resolver;
performs no credential discovery;
exposes no provider-write API;
bounds time, result count, and payload size;
preserves source identity, record kind, provenance, and contradictions;
exposes partial provider failure instead of returning a false empty success.

Release boundary

The 0.3.0 release passed package verification and a clean install from the public npm registry. Adapters accept caller-owned clients, require caller-supplied tenant and subject namespace resolution, and implement no provider write, import, mirror, cutover, or hosted /v1/memory client. Use the CLI dry runs to inspect local project metadata without a credential or provider write. The hosted candidate API has a separate deployment and authorization boundary.

Get started

Core Concepts

Releases

Memory federation SDK

Install

Query an existing LangGraph store

Release boundary

​Install

​Query an existing LangGraph store

​Release boundary

Install

Query an existing LangGraph store

Release boundary