Self-host Verdyn
Verdyn is open source. The whole product — the agronomy engine, the web app, the automated executor, and the MCP server — runs from one repo you control. Your controller credentials never touch anyone else's infrastructure.
Quick start
git clone https://github.com/jasonbdiaz/verdyn cd verdyn && npm install cp web/.env.example web/.env.local # fill in what you need npm run dev -w @verdyn/web # http://localhost:3000
No database? Everything still works in-browser with a local profile — persistence, accounts, unattended execution and the MCP endpoint activate when you add one.
Environment
| DATABASE_URL | Postgres (Neon works great). Optional — the engine runs without it; persistence, auth and automation need it. |
| VERDYN_SESSION_SECRET | 32+ chars. Seals session cookies (AES-256-GCM). |
| CRON_SECRET | Bearer token protecting the /api/cron/run executor. |
| RESEND_API_KEY / EMAIL_FROM | Optional — magic-link sign-in emails (console-logged without it). |
| NEXT_PUBLIC_BASE_URL | Public origin, used in agent-connector URLs. |
Apply the schema with DATABASE_URL=… node web/scripts/migrate.mjs. For unattended watering, schedule /api/cron/run every 15 minutes (Vercel crons, GitHub Actions, or any scheduler) with the CRON_SECRET bearer header.
Safety model
The executor enforces defense-in-depth server-side: watering windows are hard-gated at the moment a valve would open, live weather is re-checked (wind + observed rain), every run is idempotently ledgered, and controller credentials are used once to mint a session token — never stored. Read SECURITY.md before pointing it at a real controller.
The other tiers
Don't want to run servers? The hosted Expert tier is the same code with the ops handled — sign up, connect your controller, optionally wire your AI agent via MCP.