AgentMemo
Persistent memory and human approval for any AI agent — one API, any model, MCP-native.
⚠️ What This Skill Does
This skill connects your agent to the AgentMemo cloud API (api.agentmemo.net) to store and retrieve memories and request human approvals. Your agent's memory content is sent to and stored on AgentMemo's servers.
- - Requires: A free API key from agentmemo.net — set as
AGENTMEMO_API_KEY in your OpenClaw environment - Data: Memory content you store is sent to
api.agentmemo.net over HTTPS. You own your data and can delete it at any time. - Optional: The
agentmemo-mcp npm package is only needed for MCP client setups (Claude Desktop, Cursor, etc.) — not required for REST/SDK use - No data sharing: AgentMemo does not share your data with third parties. See privacy policy.
If you prefer fully local memory, this skill is not for you. If you're comfortable with a cloud API (like you'd use for any other SaaS tool), read on.
AgentMemo solves the two biggest pain points of autonomous AI agents:
- 1. Amnesia — agents forget everything between sessions. No more starting from zero.
- Dead ends — agents need to pause and ask a human before sensitive actions. Now they can.
Features
- - 🧠 Persistent memory — store, search, and retrieve memories across sessions
- ✅ Human approval gateway — agents pause, humans approve/reject, agents resume
- 🔌 MCP-native — one-line setup in Claude, Cursor, Windsurf, OpenClaw, or any MCP client
- 🌐 Works with any model — REST API, store in Claude, recall in GPT, use in local Llama
- 📦 npm SDK —
npm install agentmemo for TypeScript/JavaScript projects - 🆓 Free tier — 10K memories, 100 searches/day, no credit card needed
Quick Start
Get your free API key
Sign up at
agentmemo.net → free tier, instant access.
Option 1: MCP (Claude / Cursor / OpenClaw)
Add to your MCP config (claude_desktop_config.json or equivalent):
CODEBLOCK0
That's it. Your agent now has remember, recall, forget, request_approval, and check_approval tools.
Option 2: OpenClaw (this skill)
Set your API key in OpenClaw config or workspace env:
CODEBLOCK1
Then reference this skill in your agent instructions — see Usage below.
Option 3: REST API directly
CODEBLOCK2
Option 4: TypeScript/JavaScript SDK
CODEBLOCK3
CODEBLOCK4
Usage (as an OpenClaw skill)
When this skill is active, use AgentMemo to:
Store memories
Save important context that should persist across sessions:
CODEBLOCK5
Use POST /memories with your AGENTMEMO_API_KEY.
Search memories
Before starting any task, search for relevant prior context:
CODEBLOCK6
Use GET /memories/search?q=QUERY&namespace=NAMESPACE.
Request approval
Before any sensitive or irreversible action, request human approval:
CODEBLOCK7
Use POST /approve. Poll GET /approve/:id or set a callback_url webhook.
API Reference
Base URL: https://api.agentmemo.net
Auth: X-API-Key: YOUR_KEY header on all requests.
| Method | Endpoint | Description |
|---|
| INLINECODE19 | INLINECODE20 | Store a memory |
| INLINECODE21 |
/memories/search | Semantic search across memories |
|
GET |
/memories/:id | Retrieve memory by ID |
|
DELETE |
/memories/:id | Delete a memory |
|
GET |
/usage | Check usage stats and limits |
|
POST |
/approve | Submit action for human approval |
|
GET |
/approve/:id | Poll approval status |
POST /memories
CODEBLOCK8
Returns { id, namespace, content, metadata, created_at }.
GET /memories/search
Query params: q (required), namespace (optional), limit (optional, max 50).
Returns { query, namespace, count, results: [{ id, content, score, metadata, created_at }] }.
POST /approve
CODEBLOCK9
Returns { id, status: 'pending', ... }.
GET /approve/:id
Returns { id, status: 'pending'|'approved'|'rejected', decision_at, ... }.
MCP Tools
When using the MCP server (npx agentmemo-mcp), your agent gets these tools:
| Tool | Description |
|---|
| INLINECODE41 | Store a memory |
| INLINECODE42 |
Search stored memories |
|
forget | Delete a memory by ID |
|
list_memories | List recent memories in a namespace |
|
request_approval | Submit action for human review |
|
check_approval | Poll approval status |
Pricing
| Plan | Price | Memories | Searches/day |
|---|
| Free | $0 | 10,000 | 100 |
| Starter |
$19/mo | 250,000 | 1,000 |
| Pro | $99/mo | 2,000,000 | 10,000 |
| Team | $499/mo | Unlimited | Unlimited |
Links
AgentMemo
为任何AI智能体提供持久化记忆与人工审批——一个API，兼容任何模型，原生支持MCP。
⚠️ 本技能说明
本技能将您的智能体连接到 AgentMemo 云API（api.agentmemo.net），用于存储和检索记忆以及请求人工审批。您的智能体记忆内容将被发送并存储在AgentMemo的服务器上。
- - 需要：从 agentmemo.net 获取免费API密钥——在OpenClaw环境中设置为 AGENTMEMOAPIKEY
- 数据：您存储的记忆内容通过HTTPS发送至 api.agentmemo.net。您拥有数据所有权，可随时删除。
- 可选：agentmemo-mcp npm包仅用于MCP客户端配置（Claude Desktop、Cursor等）——REST/SDK使用无需安装
- 无数据共享：AgentMemo不会与第三方共享您的数据。详见隐私政策。
如果您偏好完全本地化记忆，本技能不适合您。如果您接受使用云API（如同使用其他SaaS工具），请继续阅读。
AgentMemo 解决了自主AI智能体的两大痛点：
- 1. 健忘症 —— 智能体在会话间遗忘一切。不再需要从零开始。
- 死胡同 —— 智能体在执行敏感操作前需要暂停并询问人类。现在它们可以做到了。
功能特性
- - 🧠 持久化记忆 —— 跨会话存储、搜索和检索记忆
- ✅ 人工审批网关 —— 智能体暂停，人类批准/拒绝，智能体继续执行
- 🔌 原生MCP支持 —— 在Claude、Cursor、Windsurf、OpenClaw或任何MCP客户端中一行配置即可
- 🌐 兼容任何模型 —— REST API，在Claude中存储，在GPT中回忆，在本地Llama中使用
- 📦 npm SDK —— npm install agentmemo 用于TypeScript/JavaScript项目
- 🆓 免费套餐 —— 10,000条记忆，每天100次搜索，无需信用卡
快速开始
获取免费API密钥
在
agentmemo.net 注册 → 免费套餐，即时访问。
选项1：MCP（Claude / Cursor / OpenClaw）
添加到您的MCP配置（claudedesktopconfig.json 或等效文件）：
json
{
mcpServers: {
agentmemo: {
command: npx,
args: [agentmemo-mcp],
env: {
AGENTMEMOAPIKEY: yourapikey_here
}
}
}
}
就这样。您的智能体现在拥有 remember、recall、forget、requestapproval 和 checkapproval 工具。
选项2：OpenClaw（本技能）
在OpenClaw配置或工作区环境中设置您的API密钥：
bash
AGENTMEMOAPIKEY=amyourkey_here
AGENTMEMOAPIURL=https://api.agentmemo.net
然后在您的智能体指令中引用本技能——参见下面的使用说明。
选项3：直接使用REST API
bash
存储记忆
curl -X POST https://api.agentmemo.net/memories \
-H X-API-Key: YOUR_KEY \
-H Content-Type: application/json \
-d {content: 用户偏好深色模式和紧凑布局, namespace: preferences}
搜索记忆
curl https://api.agentmemo.net/memories/search?q=用户+偏好&namespace=preferences \
-H X-API-Key: YOUR_KEY
请求人工审批
curl -X POST https://api.agentmemo.net/approve \
-H X-API-Key: YOUR_KEY \
-H Content-Type: application/json \
-d {action: 发送邮件至client@example.com, context: 草稿已准备好审核}
选项4：TypeScript/JavaScript SDK
bash
npm install agentmemo
typescript
import AgentMemo from agentmemo;
const memo = new AgentMemo({ apiKey: process.env.AGENTMEMOAPIKEY });
// 存储记忆
await memo.memories.store({
content: 项目截止日期为3月31日,
namespace: project-alpha
});
// 搜索记忆
const results = await memo.memories.search(截止日期, { namespace: project-alpha });
// 请求人工审批
const approval = await memo.approvals.request({
action: 删除500个旧日志文件,
context: 释放2GB磁盘空间
});
使用说明（作为OpenClaw技能）
当本技能激活时，使用AgentMemo进行以下操作：
存储记忆
保存应在会话间持久化的重要上下文：
记住：[值得保留的内容]
命名空间：[项目/用户/智能体 — 可选，默认为default]
使用 POST /memories，携带您的 AGENTMEMOAPIKEY。
搜索记忆
开始任何任务前，搜索相关的先前上下文：
回忆：[您要查找的内容]
使用 GET /memories/search?q=QUERY&namespace=NAMESPACE。
请求审批
在任何敏感或不可逆操作前，请求人工审批：
请求审批：[操作描述]
上下文：[为什么需要执行此操作]
使用 POST /approve。轮询 GET /approve/:id 或设置 callback_url webhook。
API参考
基础URL：https://api.agentmemo.net
认证：所有请求需携带 X-API-Key: YOUR_KEY 头部。
| 方法 | 端点 | 描述 |
|---|
| POST | /memories | 存储记忆 |
| GET |
/memories/search | 跨记忆语义搜索 |
| GET | /memories/:id | 按ID检索记忆 |
| DELETE | /memories/:id | 删除记忆 |
| GET | /usage | 检查使用统计和限制 |
| POST | /approve | 提交操作等待人工审批 |
| GET | /approve/:id | 轮询审批状态 |
POST /memories
json
{
content: 字符串（必填）,
namespace: 字符串（可选，默认：default）,
metadata: {}
}
返回 { id, namespace, content, metadata, created_at }。
GET /memories/search
查询参数：q（必填），namespace（可选），limit（可选，最大50）。
返回 { query, namespace, count, results: [{ id, content, score, metadata, created_at }] }。
POST /approve
json
{
action: 字符串（必填）—— 智能体想要执行的操作,
context: 字符串（可选）—— 背景/推理,
callback_url: 字符串（可选）—— 决策通知的webhook
}
返回 { id, status: pending, ... }。
GET /approve/:id
返回 { id, status: pending|approved|rejected, decision_at, ... }。
MCP工具
使用MCP服务器（npx agentmemo-mcp）时，您的智能体将获得以下工具：
搜索已存储的记忆 |
| forget | 按ID删除记忆 |
| list_memories | 列出命名空间中的最近记忆 |
| request_approval | 提交操作等待人工审核 |
| check_approval | 轮询审批状态 |
定价
| 套餐 | 价格 | 记忆条数 | 每日搜索次数 |
|---|
| 免费 | $0 | 10,000 | 100 |
| 入门 |
$19/月 | 250,000 | 1,000 |
| 专业 | $99/月 | 2,000,000 | 10,000 |
| 团队 | $499/月 | 无限 | 无限 |
链接
