Build and Run Your First VECTOR Task

This page is a Tutorial. You are a student. By the end, you will have VECTOR running and will have submitted your first autonomous task. Every step is guaranteed to work when followed exactly.

By the end of this tutorial you will have:

  1. The OpenClaw gateway running on your machine
  2. VECTOR initialized with its memory database
  3. The BEE cognitive plugin installed
  4. Your first task submitted and delivered

This takes about 15 minutes. Follow every step in order. Do not skip ahead.

Before You Begin

This tutorial requires:

If any of these are missing, install them now before continuing.

Part 1: Install OpenClaw

OpenClaw is the runtime that hosts VECTOR. Install it globally:

npm install -g openclaw

Confirm the installation worked:

openclaw --version

You will see a version number printed. That means OpenClaw is installed.

Part 2: Clone the VECTOR Workspace

The workspace is where VECTOR's memory, tasks, and configuration live:

git clone https://github.com/vectorcmd/vector-workspace ~/.openclaw/workspace

Move into the workspace directory:

cd ~/.openclaw/workspace

Install the workspace dependencies:

npm install

Wait for the installation to finish. You will see a summary like added 312 packages.

Part 3: Create Your Secrets File

VECTOR's credentials must never go in the main config file. Create a separate secrets file:

mkdir -p ~/.openclaw/agents/main/agent
touch ~/.openclaw/agents/main/agent/auth-profiles.json
chmod 700 ~/.openclaw/agents/main/agent/auth-profiles.json

Open the file in your editor and paste in this exact structure, replacing the placeholder values with your real API keys:

{
  "anthropic": {
    "mode": "api_key",
    "api_key": "YOUR_ANTHROPIC_KEY_HERE"
  },
  "openai": {
    "mode": "api_key",
    "api_key": "YOUR_OPENAI_KEY_HERE"
  }
}

Save the file. Do not put this file anywhere other than ~/.openclaw/agents/main/agent/. It is excluded from git by default.

Part 4: Validate the Configuration

VECTOR's runtime config lives in openclaw.json. Before you do anything else, validate it:

node scripts/validate-openclaw-config.js

You will see:

✓ openclaw.json is valid
✓ No credentials detected in config
✓ All required keys present

If you see any errors, read them carefully — they will tell you exactly which key is wrong and what it should be.

Part 5: Initialize the Database

VECTOR uses SQLite as its operational backbone. Initialize it:

node scripts/init-db.js

Confirm the tables were created:

sqlite3 state/vector.db ".tables"

You will see:

audit_log  content_calendar  cost_tracking  health_checks  pm_specializations  tickets

All six tables must be present. If any are missing, run node scripts/init-db.js --repair and check again.

Part 6: Install the BEE Plugin

BEE is VECTOR's cognitive architecture plugin. It lets the system learn from every task it runs:

npm install -g openclaw-bee --loglevel=error

Now open openclaw.json and find the plugins.entries section. Add the BEE entry:

{
  "plugins": {
    "entries": {
      "bee": {
        "enabled": true,
        "path": "openclaw-bee"
      }
    }
  }
}

Save the file, then validate again to confirm no mistakes crept in:

node scripts/validate-openclaw-config.js

Part 7: Start the Gateway

Start the gateway daemon:

openclaw gateway start

Check that it's running:

openclaw gateway status

You will see:

● Gateway running
  Port: 18789
  Agent: main
  Model: anthropic/claude-opus-4-6
  Plugins: bee ✓

The gateway is now running. VECTOR is live.

Part 8: Open the VECTOR Interface

Open your browser and navigate to:

http://localhost:3001

You will see the ACE Terminal interface. VECTOR boots automatically and runs its initialization sequence. Watch the console — you will see it:

  1. Read SOUL.md — loading its identity
  2. Read USER.md — loading your profile
  3. Check for an active sprint to resume
  4. Run its memory integrity check

When you see the VECTOR prompt appear, the system is ready.

Part 9: Submit Your First Task

Type this into the VECTOR interface and press Enter:

Create a markdown file called hello-world.md in the workspace root. 
It should contain a heading "Hello from VECTOR" and three bullet points 
describing what VECTOR is in one sentence each.

Watch what happens. VECTOR will:

  1. Create a ticket in the database
  2. Assess the task complexity
  3. Decide it's simple enough to handle directly (no PM spawn needed for a one-file task)
  4. Create the file
  5. Report back with confirmation

After VECTOR responds, confirm the file was actually created:

cat ~/.openclaw/workspace/hello-world.md

You will see the file VECTOR created. This is the critical principle: VECTOR verifies; it does not just claim. You have now seen this in practice.

Part 10: Submit a Multi-Agent Task

Now try something that requires VECTOR to spawn a Project Manager. Type this:

Research the top 3 open-source multi-agent frameworks released in the last 6 months 
and write a comparison table to projects/research/agent-frameworks.md. 
Include: framework name, stars, primary use case, and one-line verdict.

This time, VECTOR will:

  1. Create a ticket with a unique ID (like TASK-20260227-001)
  2. Spawn ORACLE (the research PM)
  3. ORACLE will plan the research, run Gate A, then spawn research workers
  4. Workers will search and synthesize
  5. ORACLE will review the output, verify the file exists, and close the ticket
  6. VECTOR will report back to you with the file path and a summary

This takes a few minutes. When it completes, check the output:

cat ~/.openclaw/workspace/projects/research/agent-frameworks.md

And check the audit trail — every action is logged:

sqlite3 ~/.openclaw/workspace/state/vector.db \
  "SELECT timestamp, actor, action, result FROM audit_log ORDER BY id DESC LIMIT 10;"

You will see the full chain: VECTOR creates ticket, spawns ORACLE, ORACLE runs gates, workers complete, ticket closes.

What You've Built

You now have a running VECTOR instance that:

This is the foundation. Everything else — PM profiles, custom skills, enterprise deployment, belief sharing — builds on what you've just set up.

Where to Go Next