OpenClaw and the Agent Fleet
Verdify sits inside a larger Vallery agent system. The greenhouse work is led by Iris, the AI planner, but Iris is not alone: Cortex provides inference capacity, Sentinel handles camera and vision work, Nexus observes the fleet, Web publishes the public site, and supporting agents keep infrastructure, backups, and operations aligned.
OpenClaw is the tool and agent coordination layer around that homelab. It gives agents a consistent way to call approved tools, retrieve context, hand work across hosts, and leave operational evidence. It is not the greenhouse safety controller.
Roles
Reads telemetry, forecasts, scorecards, lessons, site context, and tunable definitions, then writes bounded plan intent for dispatcher validation.
Runs local GPU-backed AI workloads for model serving, retrieval, embeddings, document analysis, and agent support.
Maintains the camera and visual-observation side of the system so scalar climate data can be checked against what the greenhouse visibly looks like.
Hosts Prometheus, Grafana, alerting, network diagnostics, and fleet health dashboards. Public Verdify pages mirror a narrow, safe subset of these metrics.
Turns operational data, plans, scorecards, photos, and documentation into the public Verdify site.
Coordinate broader homelab operations, scheduling, governance, backups, and infrastructure maintenance.
How This Helps the Greenhouse
The agent fleet makes the planning loop more evidence-rich. Iris can reason with greenhouse telemetry, prior plans, lessons learned, public documentation, forecast context, equipment definitions, and visual observations instead of treating the greenhouse as a generic controls problem.
That additional context is useful only because the lower layers stay strict. The MCP tools and dispatcher validate writes, tunable bounds are explicit, and firmware remains deterministic. If an AI plan is incomplete, stale, or unsafe, the validation path should reject it before it reaches the controller.
What Is Public
The public site now shows a subtle version of the homelab infrastructure story: Inference and Homelab Compute shows the fleet roles, and Resource Use separates greenhouse utility consumption from AI compute power.
The point is accountability, not spectacle. Verdify should show the cost and shape of the AI layer while making clear that plant safety depends on the ESP32 control loop, dispatcher validation, alerts, and operator review.