Memory (AgentMemo)

OpenClaw external memory plugin for AgentMemo. Enables semantic memory search and auto-capture via a self-hosted AgentMemo server. Install to use AgentMemo a...

Security Scan

The plugin's code, instructions, and config are coherent with its stated purpose: it proxies semantic memory to an AgentMemo HTTP service and may auto-recall or auto-capture conversation data; there are no unexplained credentials, downloads, or unrelated capabilities.

Purpose & Capability

Name/description match the code and SKILL.md: the plugin connects to an AgentMemo HTTP server (serverUrl, optional apiKey, optional userId) and implements search/add/health and auto-recall/auto-capture hooks. No unrelated environment variables or binaries are requested.

Instruction Scope

SKILL.md and code are aligned: the plugin will inject retrieved memories into context before an agent turn and (if enabled) will analyze user messages after conversations and POST captured snippets to the configured AgentMemo server. This is expected functionality for a memory plugin, but it means user conversation text may be transmitted to the configured server.

Install Mechanism

There is no external download/install script in the skill bundle. The package.json lists an npm spec for normal installation via OpenClaw/ClawHub. Nothing in the manifest points to arbitrary remote archives or shorteners.

Credentials

No required credentials are declared. An API key is optional and reasonable for authenticated AgentMemo instances. However, conversation content can be sent to the configured server (especially when autoCapture is enabled), so providing a remote serverUrl implicitly grants that server access to any captured messages.

Persistence & Privilege

The skill does not request always:true and does not modify other skills or system settings. It registers normal lifecycle hooks (before_agent_start, agent_end) appropriate for a memory provider.

Assessment

This plugin behaves as advertised: it stores and retrieves semantic memories from an AgentMemo HTTP service. Before installing, verify you trust the target serverUrl (prefer a localhost/self‑hosted instance) because captured conversation text can be transmitted there when autoCapture is enabled. If you don't want user messages sent remotely, leave autoCapture disabled (default: false) and consider running an AgentMemo server locally. If you configure values using ${ENV_VAR}, ensure those environment variables are set (the code will throw if a referenced env var is missing). Review the repository URL and source (package.json points to a GitHub repo) and confirm you are comfortable with its provenance before using a remote server for sensitive data.

External memory plugin for OpenClaw that delegates semantic memory storage and retrieval to a self-hosted AgentMemo server.

Install

openclaw install @openclaw/memory-agentmemo

Or via ClawHub:

clawhub install memory-agentmemo

Configuration

In your openclaw.json, add the plugin config:

{
  "plugins": {
    "memory-agentmemo": {
      "serverUrl": "http://localhost:8000",
      "apiKey": "your-api-key",
      "userId": "karl"
    }
  }
}

Config Options

Field	Required	Description
serverUrl	✅	AgentMemo HTTP server URL
apiKey	❌	API key if your server requires auth
userId	❌	User ID for memory namespacing
autoCapture	❌	Auto-capture memories from conversations (default: false)
autoRecall	❌	Auto-inject relevant memories into context (default: true)

AgentMemo Server

Run your own AgentMemo server: https://github.com/agentmemo/agentmemo

pip install agentmemo
agentmemo server --port 8000