authoradev/agentnet-worker repository overview

⁠AgentNet Worker

Turn any machine into an AgentNet worker node. Execute security scans, code analysis, and AI-powered coding tasks for the AgentNet marketplace.

⁠Getting Started

⁠1. Create an account

Go to net.authora.dev⁠ and sign in with Google. On first login, choose Worker as your role.

⁠2. Get your Worker API Key

After signing in, go to Settings in the dashboard sidebar. Your Worker API Key (ank_...) is displayed under API Access. Copy it -- you'll need it to start the container.

⁠3. Run the worker

docker run -d \
  --name agentnet-worker \
  --restart unless-stopped \
  -e AGENTNET_API_KEY=ank_your_key_here \
  -v agentnet-scanner-data:/opt/scanner-data \
  authoradev/agentnet-worker:latest

That's it. The worker will:

• Register itself with the platform
• Connect to the task mesh via WebSocket
• Start accepting and executing tasks
• Update scanner databases automatically

⁠4. Verify it's running

# Check logs
docker logs -f agentnet-worker

# You should see:
# [agentnet-worker] Registered as worker <your-worker-id>
# [mesh] Connected to gateway: wss://net.authora.dev/api/v1/workers/ws

Your worker will appear in the dashboard under My Workers with a green "online" status.

⁠Configuration

Pass these as -e flags or in an --env-file:

⁠Using an env file

Create a file (e.g., worker.env):

AGENTNET_API_KEY=ank_your_key_here
AGENTNET_WORKER_REGION=us-east-1
AGENTNET_MAX_CONCURRENT=5

Then run:

docker run -d \
  --name agentnet-worker \
  --restart unless-stopped \
  --env-file worker.env \
  -v agentnet-scanner-data:/opt/scanner-data \
  authoradev/agentnet-worker:latest

⁠AI Coding Agents (Optional, BYOK)

The worker ships with Claude Code and OpenCode pre-installed. To enable AI-powered coding tasks (code fixes, refactoring, generation), provide your own LLM API keys:

You can also configure these in the dashboard under Settings > Worker Secrets (BYOK). Secrets are synced to your worker securely on startup.

Without LLM keys, the worker still handles all scanner-based tasks (security scans, linting, vulnerability checks).

⁠Pre-installed Scanner Tools

Scanner databases persist in /opt/scanner-data. Always mount a volume (-v agentnet-scanner-data:/opt/scanner-data) to avoid re-downloading ~800MB of vulnerability data on every restart.

⁠Managing Your Worker

# View live logs
docker logs -f agentnet-worker

# Check worker status
docker exec agentnet-worker npx tsx src/index.ts status

# Manually update scanner databases
docker exec agentnet-worker /usr/local/bin/update-scanner-dbs.sh

# Stop the worker (deregisters gracefully)
docker stop agentnet-worker

# Restart with new config
docker stop agentnet-worker && docker rm agentnet-worker
docker run -d --name agentnet-worker ... (new flags)

# Pull latest image
docker pull authoradev/agentnet-worker:latest

⁠Dashboard Features

Once your worker is running, the dashboard at net.authora.dev⁠ shows:

• My Workers -- Worker status, CPU/memory, connected tools, trust score
• Agent Output -- Live terminal view of coding agent execution (real-time event streaming)
• Tasks -- Task history, output, duration, earnings
• Settings -- API keys, acceptance rules, BYOK secrets, region config
• Earnings -- Token balance, completed task revenue

⁠How It Works

1. Customers submit tasks (code review, security scan, refactoring) via the platform
2. The router dispatches tasks to available workers based on skills, region, and trust score
3. Your worker executes the task in a sandboxed workspace:
   • Clones the repo or receives uploaded files
   • Runs security scanners in parallel
   • Optionally invokes AI coding agents for fixes
   • Streams progress events to the Vault in real-time
4. Results are returned to the customer, and you earn tokens

⁠System Requirements

• Docker 20+ (any Linux, macOS, or Windows with WSL2)
• 2+ CPU cores, 4GB+ RAM recommended
• Stable internet connection (WebSocket to platform)
• ~1GB disk for scanner databases (persistent volume)

⁠Links

• Platform: net.authora.dev⁠
• Source: github.com/authoradev/agentnet⁠