tanstack-ai-memory-mem0
GitHub提供 mem0 内存适配器，通过 HTTP 连接 mem0 服务器实现 AI 会话的记忆持久化与检索。无需额外 SDK 依赖，支持用户级记忆保存、召回及搜索调优，适用于需要长期记忆能力的 AI 应用开发。
Trigger Scenarios
Install
npx skills add TanStack/ai --skill tanstack-ai-memory-mem0 -g -y
SKILL.md
Frontmatter
{
"name": "tanstack-ai-memory-mem0",
"description": "Use when wiring mem0() from @tanstack\/ai-memory\/mem0 — a hosted memory adapter that talks to a mem0 server over plain HTTP (no SDK peer). Requires a running mem0 server."
}
mem0 Memory Adapter
Hosted recall/save adapter backed by a mem0 server. mem0 owns extraction and ranking
server-side. Talks to the server over plain HTTP — no SDK peer dependency.
Setup
import { memoryMiddleware } from '@tanstack/ai-memory'
import { mem0 } from '@tanstack/ai-memory/mem0'
// Build per request from the server-validated session — never from req.body.
function memoryFor(session: { userId: string; threadId: string }) {
const memory = mem0({ user: session.userId }) // baseUrl defaults to MEM0_URL
return memoryMiddleware({
adapter: memory,
scope: { threadId: session.threadId, userId: session.userId },
})
}
Requires a running mem0 server (self-hosted or hosted). Point it via baseUrl (or
MEM0_URL); pass apiKey (or MEM0_ADMIN_API_KEY) when secured.
Options
user— mem0user_id(falls back toscope.userId, then'demo-user').baseUrl— mem0 server URL (defaultMEM0_URLorhttp://localhost:8000).apiKey— bearer token (defaultMEM0_ADMIN_API_KEY).rerank(defaulttrue),threshold(default0.1) — search tuning.
Scope fields: requests send user_id and run_id (scope.threadId). tenantId
and namespace are not sent — encode multi-tenant isolation into user if needed.
save posts the { user, assistant } turn to /memories; recall queries /search
and renders the results into the system prompt. mem0 exposes no LLM tools.
Version History
-
645757a
Current 2026-09-22 03:42
重构了 setup 示例，将 memory 实例化移至 per-request 函数中以确保安全性；更新了代码块以通过 kiira 类型检查并修正文档描述。
- 05280a5 2026-07-30 23:52
