@inorainllc/wellyon

Monitoring agent for the Wellyon server widget. Runs on your own Node.js/Linux server, connects outbound only to the Wellyon cloud and streams:

• host metrics (CPU, memory, disk, load, network)
• database statistics (slow queries, unused indexes, bloat, connections) for PostgreSQL, MySQL and MariaDB
• read-only query results for the widget's "Ask AI" charts
• managed database backups (opt-in): pg_dump / mysqldump runs locally, dumps stay on your server, only metadata is reported
• system monitoring (opt-in): Docker containers + stats, Nginx status, open ports (with optional close), and top-repeated system-log patterns with time-trend detection

Your database credentials never leave this server. Only metrics, schema metadata, statistics and query result rows are transmitted.

Quick start

1. In your Wellyon dashboard, add a server under Servers and copy the pairing token.
2. On the server:

npx @inorainllc/wellyon init    # interactive setup + enrollment
npx @inorainllc/wellyon start   # run the agent

For production, run start under systemd / pm2 / a container.

Configuration

init writes wellyon-agent.json (mode 600):

{
  "endpoint": "https://api.wellyon.ai",
  "pairingToken": "wsa_...",            // one-time; replaced by an agent token after enrollment
  "serverName": "prod-db-1",
  "databases": [
    {
      "id": "main",
      "kind": "postgres",
      "url": "postgres://wellyon_ro:...@localhost:5432/app",
      "backup": {
        "enabled": true,                     // permission gate — without this the agent refuses backup requests
        "dir": "/var/backups/wellyon/main",  // default: ~/.wellyon/backups/<id>
        "keep": 7,                           // dumps kept locally (older pruned)
        "timeoutMin": 60,
        "everyHours": 24                     // optional agent-side schedule; omit for on-demand only
      }
    },
    { "id": "orders", "kind": "mysql", "host": "127.0.0.1", "user": "wellyon_ro", "password": "...", "database": "orders" }
  ],
  "system": {
    "docker": true,                       // Docker containers + live CPU/mem (needs docker CLI + socket access)
    "nginx": true,                        // nginx version/running state
    "nginxStatusUrl": "http://127.0.0.1/nginx_status", // optional stub_status for live connections
    "ports": true,                        // listening TCP ports (lsof / ss)
    "allowClosePorts": false,             // gate — must be true for the widget to close a port
    "logs": {
      "journald": true,                   // Linux systemd journal
      "files": ["/var/log/nginx/error.log"], // extra log files to scan
      "windowMinutes": 15
    },
    "intervalSec": 30
  },
  "metricsIntervalSec": 15,
  "statsIntervalSec": 300
}

System monitoring

All opt-in via the system block. The agent shells out to read-only tools (docker, lsof/ss, pgrep, journalctl) — install what you want monitored. Closing a port terminates the process bound to it and is gated twice: system.allowClosePorts: true in this config AND the system:manage scope on the widget token. Log analysis normalizes each line to a template (masking timestamps/ids/ips/numbers), counts repeats, and compares windows to flag new / rising patterns — so a flood of one error surfaces at the top.

Managed backups

Backups are off by default and gated by the local config — the cloud can only trigger a dump for a database whose backup.enabled is true. Dumps run locally (pg_dump -Fc for Postgres, mysqldump --single-transaction | gzip for MySQL/MariaDB, both tools must be installed), land in backup.dir with mode 600, and are pruned to the newest keep. The dump file never leaves your server — Wellyon receives only name/size/duration/status metadata. Note: dumping usually needs broader privileges than monitoring; use a dedicated backup user or grant SELECT/LOCK TABLES to the configured one.

Lookup order: $WELLYON_CONFIG → ./wellyon-agent.json → ~/.wellyon/agent.json. Env overrides: WELLYON_ENDPOINT, WELLYON_PAIRING_TOKEN, WELLYON_SERVER_NAME.

Security model

• Read-only DB user strongly recommended — init prints the GRANT statements.
• Every SQL statement received from the cloud passes a local sandbox: single-statement lexical gate → SQL parser gate → read-only transaction/session with a server-side statement timeout (the hard guarantee), plus row/payload caps.
• Enrollment uses a one-time pairing token; afterwards a revocable agent token (stored in ~/.wellyon/agent-state.json, mode 600) authenticates the outbound socket. Revoking the server in the dashboard cuts the agent off.

Commands

| Command | Description | |---|---| | init | Interactive setup, writes config, enrolls the server | | start | Run the agent (default) | | test-connection | Verify Wellyon + database connectivity, then exit |