Google Vertex AI Memory Bank

Install and configure the OpenClaw Vertex AI Memory Bank plugin for persistent, cross-agent memory. Use when the user wants long-term memory, cross-session r...

Security Scan

The skill's requirements and instructions are coherent with its stated purpose (setting up a Vertex AI–backed memory plugin), but it runs external code (git clone + npm install) and requires a Google account with permissions — review the repository and limit credentials before use.

Purpose & Capability

Name and description claim a Vertex AI memory plugin; the SKILL.md and setup script perform GCP/Vertex AI operations, install an npm plugin, and configure openclaw.json. Required tooling (gcloud, Node.js, npm) and enabling Vertex AI are consistent with that purpose.

Instruction Scope

Runtime instructions and the included setup.sh stay within the stated task: checking tools, creating a Vertex AI reasoning engine, cloning the plugin repo, building it, and advising how to configure/restart OpenClaw. The script uses gcloud auth tokens and creates resources in the user project (expected for this integration). It does not reference unrelated system paths or ask for unrelated secrets.

Install Mechanism

There is no packaged install spec; the setup script clones https://github.com/Shubhamsaboo/openclaw-vertexai-memorybank and runs npm install && npm run build. Cloning from GitHub is reasonable, but running npm install executes whatever package scripts exist in that repo — this is normal for installing plugins but carries the usual risk of executing arbitrary code from the upstream repo. The script does not validate git is installed (minor robustness issue).

Credentials

The skill declares no required env vars, which matches the bundle. However, the setup relies on gcloud application-default credentials (or interactive gcloud login) and uses gcloud auth print-access-token to call the Vertex AI REST API. That requires a Google identity with permissions to enable services and create reasoning engines (project-level privileges). This is expected for the stated task but is high-privilege relative to typical local tools — use least-privilege credentials (service account with limited roles) rather than a broad personal account.

Persistence & Privilege

always is false and the skill does not request permanent platform-wide inclusion. The script writes to ~/.openclaw/plugins and suggests edits to ~/.openclaw/openclaw.json — scoped to the user's home. The plugin will persist user/agent memories to Vertex AI (by design), so consider privacy and retention settings, but no unexpected privilege escalation is present in the package itself.

Assessment

This skill appears to do what it says: set up a Vertex AI memory plugin. Before running the setup script: 1) Review the upstream repository and its package.json (postinstall/build scripts) so you know what code will run when npm install/build executes. 2) Avoid using a broad personal/project owner account interactively; prefer a service account with the minimum IAM roles required for enabling aiplatform and creating the reasoning engine (or review permissions first). 3) Understand that agent memories (including any sensitive data) will be stored in your GCP project — check retention/TTL, encryption, and privacy. 4) Test in a non-production project to observe costs and behavior. If you want higher assurance, manually perform each step (enable API, create reasoning engine, inspect plugin code) rather than running the script unreviewed.

Vertex AI Memory Bank Plugin

Give your OpenClaw agent persistent, cross-agent memory powered by Google's Vertex AI Memory Bank.

What This Does

After setup, your agent will:

• Auto-recall: Before each turn, relevant memories are retrieved and injected into context
• Auto-capture: After each turn, facts are extracted and stored automatically
• File sync: Workspace files (MEMORY.md, USER.md, SOUL.md) sync to Memory Bank with hash tracking
• Cross-agent: Tell one agent something, all agents remember it

Prerequisites

Before running the setup script, ensure:

1. Google Cloud SDK installed and authenticated (gcloud auth application-default login)
2. A GCP project with billing enabled
3. Vertex AI API enabled on the project
4. Node.js 18+ and npm installed

If the user doesn't have these, help them set up each one.

Installation

Run the setup script:

bash scripts/setup.sh

This script will:

1. Check for required tools (gcloud, npm, node)
2. Prompt for GCP project ID and region
3. Create a Vertex AI Agent Engine reasoning engine (Memory Bank instance)
4. Install the npm plugin package
5. Add the plugin configuration to openclaw.json
6. Restart the gateway to load the plugin

Manual Installation

If the script doesn't work for your environment, follow these steps:

Step 1: Create a Memory Bank Instance

# Set your project
gcloud config set project YOUR_PROJECT_ID

# Create a reasoning engine for Memory Bank
curl -X POST \
  "https://REGION-aiplatform.googleapis.com/v1beta1/projects/YOUR_PROJECT_ID/locations/REGION/reasoningEngines" \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  -d '{"display_name": "openclaw-memory-bank"}'

Note the reasoning engine ID from the response.

Step 2: Install the Plugin

cd /path/to/openclaw-vertex-memorybank
npm install
npm run build

Step 3: Configure openclaw.json

Add to your openclaw.json under plugins:

{
  "plugins": {
    "openclaw-vertex-memorybank": {
      "enabled": true,
      "path": "/path/to/openclaw-vertex-memorybank",
      "config": {
        "projectId": "YOUR_PROJECT_ID",
        "location": "us-central1",
        "reasoningEngineId": "YOUR_REASONING_ENGINE_ID"
      }
    }
  }
}

Step 4: Restart

openclaw gateway restart

Configuration Options

Option	Default	Description
projectId	required	GCP project ID or number
location	required	GCP region (e.g. us-central1)
reasoningEngineId	required	Agent Engine reasoning engine ID
autoRecall	true	Retrieve memories before each turn
autoCapture	true	Store memories after each turn
autoSyncFiles	true	Sync workspace .md files to Memory Bank
autoSyncTopics	true	Auto-configure memory topics at startup
topK	10	Max memories to retrieve per query
perspective	"third"	Memory perspective (first or third person)
backgroundGenerate	true	Fire-and-forget memory generation
ttlSeconds	none	Auto-expire memories after N seconds

Verifying It Works

After installation, check the gateway log:

tail -f ~/.openclaw/logs/gateway.log | grep memory

You should see:

• [memory-vertex] synced N topics on startup
• [memory-vertex] recall: N memories on each turn
• [memory-vertex] capture fired (bg) after each turn

CLI Commands

The plugin adds these commands:

• memorybank-search <query> - Search your memories
• memorybank-remember <fact> - Store a specific fact
• memorybank-forget <memory_id> - Delete a memory
• memorybank-sync - Force sync workspace files
• memorybank-status - Check plugin status
• memorybank-list - List all stored memories

Troubleshooting

• "401 Unauthorized": Run gcloud auth application-default login
• "Memory Bank not found": Check reasoningEngineId matches your instance
• No memories recalled: Check topK and maxDistance settings. Try memorybank-search to verify memories exist
• High token usage: Reduce topK or set introspection: "off" to remove similarity scores

Source

Full source code and documentation: https://github.com/Shubhamsaboo/openclaw-vertexai-memorybank